Application of a computer vision algorithm to quantify the frequency and duration of children's microactivities in different play scenarios.

Application of a computer vision algorithm to quantify the frequency and duration of children’s microactivities in different play scenarios - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Expo Sci Environ Epidemiol . Author manuscript; available in PMC: 2026 Apr 17. Published in final edited form as: J Expo Sci Environ Epidemiol. 2025 Aug 13;36(4):713–724. doi: 10.1038/s41370-025-00800-3 Search in PMC Search in PubMed View in NLM Catalog Add to search Application of a computer vision algorithm to quantify the frequency and duration of children’s microactivities in different play scenarios Sara N Lupolt Sara N Lupolt 1 Department of Environmental Health & Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. 2 Risk Sciences and Public Policy Institute, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. Find articles by Sara N Lupolt 1, 2, ✉ , Qinfan Lyu Qinfan Lyu 1 Department of Environmental Health & Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. Find articles by Qinfan Lyu 1 , Guofeng Zhang Guofeng Zhang 3 Department of Computer Science, Johns Hopkins Whiting School of Engineering, Baltimore, MD, USA. Find articles by Guofeng Zhang 3 , Jiahao Wang Jiahao Wang 3 Department of Computer Science, Johns Hopkins Whiting School of Engineering, Baltimore, MD, USA. Find articles by Jiahao Wang 3 , Stacey Tang Stacey Tang 4 Department of Pediatrics, Johns Hopkins School of Medicine, Baltimore, MD, USA. Find articles by Stacey Tang 4 , Jamie Cho Jamie Cho 5 Department of Neuroscience, Johns Hopkins Krieger School of Arts and Sciences, Baltimore, MD, USA. Find articles by Jamie Cho 5 , Christina Huynh Christina Huynh 1 Department of Environmental Health & Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. Find articles by Christina Huynh 1 , Alan Yuille Alan Yuille 3 Department of Computer Science, Johns Hopkins Whiting School of Engineering, Baltimore, MD, USA. 6 Department of Cognitive Science, Johns Hopkins Krieger School of Arts and Sciences, Baltimore, MD, USA. Find articles by Alan Yuille 3, 6 , Kristin Voegtline Kristin Voegtline 4 Department of Pediatrics, Johns Hopkins School of Medicine, Baltimore, MD, USA. 7 Department of Obstetrics and Gynecology, Weill Cornell Medicine, New York, NY, USA. 8 Department of Population Health Sciences, Weill Cornell Medicine, New York, NY, USA. Find articles by Kristin Voegtline 4, 7, 8 , Keeve E Nachman Keeve E Nachman 1 Department of Environmental Health & Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. 2 Risk Sciences and Public Policy Institute, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. 9 Department of Health Policy and Management, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. Find articles by Keeve E Nachman 1, 2, 9, ✉ Author information Article notes Copyright and License information 1 Department of Environmental Health & Engineering, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. 2 Risk Sciences and Public Policy Institute, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. 3 Department of Computer Science, Johns Hopkins Whiting School of Engineering, Baltimore, MD, USA. 4 Department of Pediatrics, Johns Hopkins School of Medicine, Baltimore, MD, USA. 5 Department of Neuroscience, Johns Hopkins Krieger School of Arts and Sciences, Baltimore, MD, USA. 6 Department of Cognitive Science, Johns Hopkins Krieger School of Arts and Sciences, Baltimore, MD, USA. 7 Department of Obstetrics and Gynecology, Weill Cornell Medicine, New York, NY, USA. 8 Department of Population Health Sciences, Weill Cornell Medicine, New York, NY, USA. 9 Department of Health Policy and Management, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. ✉ Correspondence and requests for materials should be addressed to Sara N. Lupolt or Keeve E. Nachman. [email protected] ; [email protected] AUTHOR CONTRIBUTIONS SNL: Writing—original draft, conceptualization, methodology, supervision, funding acquisition, project administration. QL: Methodology, data collection, data analysis, writing—review, and editing. GZ: Methodology, investigation, data analysis, writing—review, and editing. JW: Methodology, investigation, data analysis, writing—review and editing. ST: Methodology, data analysis. JC: Methodology, data analysis. CH: Recruitment, data collection, writing—review, and editing. AY: Conceptualization, methodology, funding acquisition, supervision, writing—review and editing. KV: Conceptualization, methodology, funding acquisition, supervision, writing—review and editing. KEN: Conceptualization, methodology, writing- review and editing, funding acquisition, supervision, project administration. Issue date 2026 Jul. Reprints and permission information is available at http://www.nature.com/reprints PMC Copyright notice PMCID: PMC13085206  NIHMSID: NIHMS2155119  PMID: 40796652 The publisher's version of this article is available at J Expo Sci Environ Epidemiol Abstract BACKGROUND: Microactivity rates, including hand- and object-to-mouth contacts, are inputs for modeling children’s exposure to chemicals in soil and dust and from toys. EPA’s confidence in its current recommended microactivity frequency estimates for use in exposure assessments is low. OBJECTIVE: We aimed to quantify children’s hand- and object-to-mouth microactivities using a novel computer vision method and explored differences by lifestage and fine and gross motor development. METHODS: We recorded 20-minute videos of 61 children aged 6 to ≤18 months playing in their homes under two play contexts. We employed a novel computer vision approach that uses multiview video to estimate 3D body keypoints and measure hand-to-mouth distances, enabling quantification of each child’s hand-and object-to-mouth contact without the need for human coders. We explored differences in the frequency and duration of microactivity contacts by age and motor development. RESULTS: We observed at least one instance of a microactivity event (i.e., hand- or object-to-mouth contact) among 41 of the 61 participants observed. The median rate of object-to-mouth contacts (23 contacts/hour) was greater than hand-to-mouth contacts (6 contacts/hour). We did not observe significant differences in the frequency of either hand-to-mouth and object-to-mouth contacts by age or motor development, but we did observe greater variation in object-to-mouth contacts than previously reported. The median durations of hand- and object-to-mouth contact were 0.24 min/hour and 1.57 min/hour, respectively. SIGNIFICANCE: Our observed rates of microactivities are comparable and in some cases, less than the current EPA central tendency estimates recommended for use in risk assessments. Our high-end (95th percentile) estimate for object-to-mouth contacts, however, underscores the need for better characterization of population variability in order to protect the most highly exposed children. IMPACT: We apply a novel computer vision algorithm to quantify the microactivity frequencies and durations of 61 children 6 to ≤18 months old. We examine differences in these frequencies and durations by child lifestage and motor development. Demonstration of this method paves the way for future, larger studies that observe children for longer durations to develop confident estimates of population variability in microactivity behaviors. These and future data could inform analyses in support of the revision of EPA recommendations for children’s soil and dust ingestion rates. Keywords: Microactivity, Soil, Dust, Videography, Hand-to-mouth, Object-to-mouth INTRODUCTION Young children’s frequent mouthing behaviors present unique pathways for the non-dietary ingestion of chemical and biological contaminants [ 1 , 2 ]. Estimates of their hand- and object-to-mouth contacts, or microactivities, have been used as inputs in models for estimating children’s exposures to pesticides and other contaminants in soil and dust in human health risk assessment and management [ 3 - 5 ]. The United States Environmental Protection Agency (EPA) provides recommended default exposure assumptions for these microactivities, but the agency’s confidence in these estimates of microactivities is low due to an array of issues relating to poor quality assurance and characterization of variability in the population and limited descriptions of key uncertainties [ 1 ]. Robust distributions of children’s microactivity contact rates are needed to inform models for estimating high quality soil and dust ingestion rates [ 6 ]. There are several approaches for collecting data on children’s microactivities [ 7 , 8 ], including direct observation and manual recording in real-time [ 9 ], video annotation (either manually or with a computer software program) [ 10 - 17 ], and questionnaires completed by parents/caregivers [ 13 ]. Each of these approaches is subject to measurement inconsistencies attributable to differences in definitions of mouthing behaviors, and all require significant person-time and effort for analysis [ 18 ]. Given these barriers, many key and relevant studies have small sample sizes, limiting their representativeness to the overall U.S. population. Artificial intelligence and machine learning-based methods, which are increasingly used in a variety of consumer and research settings, can reinvigorate the field of children’s microactivity assessment [ 19 ]. Approaches that leverage computer vision have been shown to produce accurate and faster counts of microactivities comparable to traditional human-based methods, and with much less human effort. These methods can also simultaneously process and leverage multiple camera views of recorded video, ensuring the key points on the child’s body needed to quantify microactivities (e.g., mouth, hand, or object) are rarely occluded. Preventing occlusion can thus ensure that video data continuously recorded over the observation period are usable. These advances also support the expansion of microactivity assessment among more children in diverse environments and contexts. The EPA uses a lifestage approach for assessing exposure and risk among children [ 20 ]. This approach considers the broad body of evidence demonstrating important windows of susceptibility that occur during childhood [ 21 , 22 ]. In addition to these periods of physiological development when children may be more vulnerable to the effects of chemicals relative to other time periods, children are known to have higher exposures to soil and dust due to their higher rates of inhalation per body weight and hand- and object-to-mouth activities [ 22 ]. These microactivities are developmentally appropriate and necessary behaviors that help children learn about their environments [ 23 ]. The EPA recommends use of four lifestages for children under 1 year (i.e., birth to <1 month, 1 to <3 months, 3 to <6 months, and 6 to <12 months) and six lifestages for children ≥1 and <21 years (i.e., 1 to <2 years, 2 to < 3 years, 3 to <6 years, 6 to <11 years, 11 to <16 years and 16 to <21 years) [ 20 ]. While these lifestages mark important developmental phases, it is also known that development is not necessarily linear and children may develop at different rates [ 24 ]. For example, children may take their first steps at 9 months or 13 months. The extent to which specific physical or behavioral milestones may influence rates and patterns of microactivity behaviors, however, is not well understood. For instance, before a child can put an object in their mouth, they must develop the gross motor skills to control the movement of the hand and arms and the fine motor skills to grasp the object securely. To explore the relationship between microactivity and development among children ages 6 to ≤18 months, we pair an assessment of their microactivity behaviors (hand-to-mouth and object-to-mouth contacts) with a concurrent assessment of their fine and gross motor development. The goals of this study are to (1) describe the frequency and duration of hand and object-to-mouth contacts and durations during play scenarios for children in two EPA-designated lifestages (6 to <12 months and 1 to <2 years); and (2) explore differences in these key microactivities by EPA lifestage and fine and gross motor development. METHODS Recruitment Recruitment was implemented as part of the broader Innovations to Generate Estimates of children’s Soil/dust inTake (INGEST) study, whose goal is to advance methods for estimating rates of soil and dust ingestion among children. We screened 593 and recruited and enrolled 64 parents of children between 6 and 18 months residing near Baltimore, MD, USA. This age range includes two distinct EPA-designated lifestages: 6– < 12 months and 1– < 2 years. Participants were recruited through BuildClinical [ 25 ], an online data-driven platform and service that connects potential participants with research studies that might interest them. We also recruited participants from the Harriet Lane Clinic at Johns Hopkins Children’s Health Center and Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) sites in Baltimore. Flyers posted in community centers, childcare centers, and public libraries supplemented direct recruitment. Parents were screened and enrolled via telephone. We administered a modified version of the Demographics of the Child and the Primary Caregiver Family Forms from the Environmental Influences on Child Health Outcomes (ECHO) Study. To assess the child’s physical development, we administered the fine and gross motor skills domains of the Ages and Stages-3 Questionnaire, a comprehensive and reliable screening tool for assessing development in the first five years of life [ 26 ]. The ASQ-3 was administered to caregivers at enrollment via telephone to assess each child’s motor development (see Child motor assessment ). We scheduled home visits to collect video footage at a time convenient to the family. Data collection Video data were collected between April 2023 and June 2024. Two study team members visited each participant’s home to collect up to 20 min of video of the child engaged in play. First, while the child and caregiver were out of the room, the team set up four GoPro (Hero9) cameras mounted on all four sides of the room where the child most often plays. To begin data collection, we recorded up to 2 min of video footage needed to calibrate each view for the algorithm. Briefly, two black and white checkered boards of different sizes were used for extrinsic and intrinsic calibration [ 27 ]. Intrinsic calibration determines a camera’s internal parameters, including focal length, principal point, and lens distortion, which are unique to each camera and necessary for mapping 3D points to 2D image coordinates. Extrinsic calibration defines the spatial relationships between cameras by estimating their relative positions and orientations. A study team member put the large checkerboard at the center of all four cameras (which is removed prior to the child entering the scene) and presented the small checkerboard to each camera in turn, tilting the board toward and away from the camera in a figure 8 pattern. While the cameras were recording, we invited caregivers and the participating child into the play area and instructed caregivers to play with the child (i.e., unstructured play), for up to 10 min. We then asked caregivers to introduce a novel toy (i.e., Baby Einstein Take Along Tunes Musical Toy) and play with the child (structured play) an additional 10 min. Caregivers were generally engaged with the child for the duration of the observation period and helped keep the child in the room (if needed). The calibration procedure was repeated after the structured play ended. All 61 caregivers provided written informed consent to participate in the study. Forty-six participants provided written permission for the video of their child to be published for research purposes. Participants were offered a $90 gift card and invited to keep the Baby Einstein Take Along Tunes Musical Toy. The Johns Hopkins Bloomberg School of Public Health Institutional Review Board (IRB00020023) approved all study protocols. Computer vision algorithm method We developed an automated computer vision system to identify when infants bring their hands or objects to their mouths during play using video recordings from four synchronized cameras ( Fig. 1 ). The method first detects the child in each video frame, identifies key body points (like hands, mouth, and joints) in two dimensions (2D), and then combines views from multiple cameras to reconstruct the child’s body in three dimensions (3D). The system tracks the distance between the hands (or held objects) and the mouth in 3D space to flag moments when this distance falls below a pre-specified threshold of 6 cm, indicating a possible mouthing contact event. A classification model then distinguishes between hand-to-mouth and object-to-mouth actions. Finally, contact events are counted based on repeated detections over short time windows, enabling automated measurement of specific infant behaviors during natural play. Our method is the first fully automated method that allows for identification and quantification without human labor. Fig. 1. Computer vision workflow. Open in a new tab The caregiver of the child in this figure provided written informed consent to participate in the study and written permission for the video (and still images) of their child to be published for research purposes. Video curation and application of computer vision algorithm to generate microactivity data We preprocessed the play videos using Adobe Premiere Pro to synchronize streams from each of the four GoPros into one timeline. Extraneous footage was removed from each of the video files to ensure the processed videos captured the calibration and play footage and were of equal length. We generated a 4-view-in-1 video in which all four video streams were displayed synchronously in a single frame. We applied a computer vision method (described in full elsewhere [ 19 ]) to detect and quantify microactivity events during structured and unstructured play by analyzing the 4-view-in-1 video footage at a rate of 10 frames/second. Application of the method began with detecting and generating bounding boxes around the child and caregiver in each frame using Faster R-CNN [ 28 ] from the MMPose toolbox [ 29 ]. Then, we used HRNet (High-Resolution Network) [ 30 ] to identify 2D key points (specific locations on the body like the eyes, wrists, elbows, knees, etc.) in COCOWholeBody format [ 31 ] of each detected person. Grounded SAM [ 28 , 32 ] was then used to isolate the child and remove other people from the scene. Leveraging camera calibration parameters, the child’s 2D key points from all views were triangulated into 3D key points using EasyMocap [ 33 ]. EasyMocap is an open-source toolbox from Zhejiang University that reconstructs 3D human motion from regular RGB videos. It uses a method called triangulation, where 2D keypoints detected in multiple camera views are back-projected into 3D space. If the camera poses are known, the back-projected lines from each view ideally intersect at the true 3D position of the point ( Fig. S1 ), allowing EasyMocap to estimate accurate 3D poses frame by frame. These 3D key points allowed us to calculate distances among body parts and objects, enabling identification of microactivities such as hand- and object-to-mouth behaviors. A ResNet50-based binary classifier [ 34 ] was trained to distinguish between hand- and object-to-mouth behaviors. For the computer vision method to log a microactivity contact, two criteria must be satisfied. First, the distance between the hand (or held object) and the mouth must be less than a pre-specified distance D, set at 6 cm. Second, the distance must remain less than D for at least three frames in that second for the microactivity contact to be counted. A tolerance interval of four seconds was used to distinguish between multiple contacts versus a single contact of extended duration, meaning that contacts occurring within a space of less than or equal to four seconds were not considered separate microactivities. For each documented microactivity contact, we also quantified its duration and calculated the total duration of microactivities for each participant by microactivity type. We evaluated the performance of our method using 21 collected videos that were annotated by human coders. Our computer vision method shows an average counting error (i.e., the difference between the count from the computer vision algorithm and the count based on human behavioral coding for each of the three action types) of 0.23 on left-hand to mouth action, 0.04 on right-hand to mouth action and 2.18 on object to mouth action, indicating that our method is reliable. In addition, the negative effects of inaccurate camera calibration are minimized. Since the camera setups vary across home visits, camera calibrations need to be conducted every visit. However, once the cameras are calibrated in a home visit, their parameters typically stay consistent throughout that visit. But, if a camera is moved—perhaps by a child—then a second calibration is conducted at the end of the recording to capture the altered parameters. Child motor skill assessment To assess children’s fine and gross motor development at enrollment, we administered the fine and gross motor skills domains from the Ages and Stages Questionnaire-3 (ASQ-3) [ 26 ], a comprehensive and validated tool for assessing development in the first few years of life. The fine and gross motor skill domains of the ASQ-3 are each 6-item questionnaires. The child’s calendar age determines the exact set of six items administered—segmented in two-month intervals from ages 6–24 months. For each item, caregivers were asked about a particular activity (e.g., “Does your child climb onto furniture or other large objects?”). Their response options were: “My child consistently does this,” “My child sometimes does this,” or “My child has not yet done this.” Fine and gross motor scores were summed from scores for each of the 6-item responses: a response of “consistently” contributes 10 points toward the score, while responses of “sometimes” and “not yet” contribute 5 and 0 points, respectively. The resulting fine and gross motor scores each range from 0 (indicating low motor function for the child’s age) to 60 (indicating excellent motor function). We compared each fine and gross motor score to national sample values [ 35 ] – mean and standard deviation values for a specific age range - to establish cutoff points for classification. Motor status rankings were determined as follows: scores exceeding the national sample mean for the child’s ASQ-3 age version were categorized as “above average”; scores greater than one standard deviation below the mean were categorized as “sufficient”; scores greater than two standard deviations below the mean were categorized as “poor”; and scores more than two standard deviations below the mean were categorized as “very poor”. Data analysis Counts of hand- and object-to-mouth contacts and their durations were generated at the participant level separately for structured and unstructured play activities. We used descriptive statistics to summarize the frequency of contact events in R Studio 2024.12.0 + 467 with R 4.4.2. We reported time-weighted activity counts for each hand- and object-to-mouth microactivity by extrapolating the rate of contacts during the observation period (generally ~20 min) to an hourly rate. We calculated adjusted contact duration (seconds/minute) for each microactivity type by dividing the overall duration of that microactivity contact by the total recording time. Given our small sample sizes and the non-normality of the data, we conducted non-parametric Wilcoxon rank sum tests to evaluate differences in duration and frequency of microactivities by lifestage (6 to <12 months vs. 1 to <2 years) and motor development status (less developed [status of “very poor” and “poor”] vs. more developed [status of “sufficient” and “above average” for both gross and fine motor skills]). We used Pearson correlation analyses to evaluate the linear relationships between left-hand and right-hand mouthing events and total hand- and object-to-mouth activities. RESULTS Participants We recruited 64 children for videography data collection, but excluded three participant videos from analysis due to video quality issues, leaving 61 videos of children playing. The mean recording time of each video was 20.5 min (SD = 1.5). The children ranged in age from 6.0 to 17.8 months, with a mean age of 11 months ( Table 1 ). Participants were predominantly non-Hispanic (93%) and white (87%). Only three were born preterm (i.e., before 37 weeks gestation). Ninety percent of children were enrolled in the study by their biological mother. Seventy-four percent of the enrolling parents had acquired a master’s degree or higher. Table 1. Child and caregiver characteristics ( n = 61). Child characteristics ( N = 61) Age (in months) Mean Range 11.1 6.0–17.8 Sex n % Female 32 52.5 Male 29 47.5 Ethnicity Non-Hispanic 57 93.4 Hispanic 4 6.6 Race (not mutually exclusive) White 53 86.9 Black 8 13.1 American Indian or Alaska Native 0 0.0 Asian Indian 4 6.6 Chinese 4 6.6 Filipino 1 1.6 Japanese 0 0.0 Korean 1 1.6 Vietnmanese 1 1.6 Other Asian 0 0.0 Native Hawaiian 0 0.0 Guamanian or Chamoor 0 0.0 Samoan 0 0.0 Other Pacific Islander 1 1.6 Born Preterm (i.e., < 37 weeks) 3 4.9 Caregiver Characteristics Relationship to child Biological Mother 55 90.2 Biological Father 6 9.8 Ethnicity Non-Hispanic 59 96.7 Hispanic 2 3.3 Race (not mutually exclusive) White 53 86.9 Black 6 9.8 American Indian or Alaska Native 0 0.0 Asian Indian 1 1.6 Chinese 2 3.3 Filipino 1 1.6 Japanese 0 0.0 Korean 0 0.0 Vietnamese 1 1.6 Native Hawaiian or other Pacific Islander 1 1.6 Born in US 48 78.7 Born outside US 13 21.3 Marital Status Married to a partner of opposite sex 53 86.9 Married to a partner of the same sex 1 1.6 Not married but living with a partner of opposite sex 3 4.9 Not married but living with a partner of same sex 0 0.0 Widowed 0 0.0 Separated 0 0.0 Divorced 0 0.0 Single, never married 4 6.6 Education GED or equivalent 2 3.3 Associates degree 1 1.6 Bachelor’s degree 13 21.3 Master’s degree 33 54.1 Professorial or Doctorate degree 12 19.7 Open in a new tab Frequency and duration of microactivities Of the 61 participants, 41 (67%) engaged in at least one instance of either hand- or object-to-mouth contact. The median frequency of all microactivity contacts among these doers was 28 contacts/hour ( Table 2 ). Twenty-one (34%) participants engaged in at least one instance of hand-to-mouth contact; among them, the median frequency was six contacts/hour. Thirty-four (56%) participants engaged in at least one instance of object-to-mouth contact; among them, the median frequency was 23 contacts/hour. Table 2. Summary distributions of time-weighted activity frequencies and durations for hand- and object-to-mouth activities and contact duration. Frequency of events, all participants (contacts/hour) Participants Range Mean (SD) p25 p50 p75 p95 Total microactivities 61 (0–130) 24 (30) 0 12 34 85 Hand-to-mouth activities 61 (0–71) 4 (11) 0 0 3 17 Left hand to mouth 61 (0–56) 3 (9) 0 0 3 17 Right hand to mouth 61 (0–15) 1 (2) 0 0 0 3 Object-to-mouth 61 (0–123) 19 (29) 0 9 26 84 Frequency of events, doers a only (contacts/hour) Doers Total microactivities 41 (3–130) 35 (31) 12 28 42 87 Hand-to-mouth activities 21 (3–71) 12 (16) 3 6 12 33 Left hand to mouth 16 (3–56) 13 (15) 3 7 13 39 Right hand to mouth 12 (3–15) 5 (4) 3 3 4 12 Object-to-mouth 34 (3–123) 35 (32) 12 23 43 89 Duration per contact event (seconds) Events Hand-to-mouth activities 86 (1–27) 4 (4) 1 2 4 13 Object-to-mouth 407 (1–47) 6 (6) 1 3 7 19 Total contact duration per child (seconds) Doers Hand-to-mouth activities 21 (1–132) 15 (30) 2 5 10 47 Object-to-mouth 34 (1–352) 66 (90) 11 34 68 289 Duration (minutes/hour) Doers Total microactivities 41 (0–18) 3 (4) 0.41 1.55 3.12 13.21 Hand-to-mouth activities 21 (0–7) 1 (2) 0.1 0.24 0.54 2.35 Left hand to mouth 16 (0–6) 1 (1) 0.1 0.37 0.61 3.15 Right hand to mouth 12 (0–1) 0 (0) 0.05 0.05 0.17 0.78 Object-to-mouth 34 (0–18) 3 (4) 0.5 1.57 3.16 13.36 Open in a new tab a We define a doer as a participant with at least one hand-to-mouth or object-to-mouth contact. Hourly counts of all microactivities (i.e., total of hand- and object-to-mouth contacts) among all participants were greater during structured play, i.e., when the child was given a new toy as a novel stimulus to explore compared to unstructured play when the child was allowed to play with all of their usual toys ( Tables S2 - 5 ). The median for structured play was 11 contacts/hour, compared to 6 contacts/hour for unstructured play. Though not statistically significant, this difference was most pronounced for object-to-mouth activities. We observed a moderately strong positive, linear relationship between the frequency of left-hand-to-mouth and right-hand-to-mouth contacts (Pearson’s correlation, r = 0.68, p < 0.001). We did not observe a linear relationship between hand-to-mouth frequency and object-to-mouth frequency (Pearson’s correlation, r = −0.099, p = 0.45) ( Fig. S5 ). When hand-to-mouth contact events occurred, the duration ranged from 1 to 27 s, with a median of 2 s. When object-to-mouth contact events occurred, the duration ranged from 1 to 47 s, with a median of 3 s ( Table 2 ). Among doers, the median durations of hand- and object-to-mouth contact were 0.24 min/hour and 1.57 min/hour, respectively. Microactivities by age and lifestage Among children aged 6 < 12 months who engaged in at least one hand-to-mouth contact (“doers”; n = 12), we observed rates of hand-to-mouth activity ranging from 3 to 71 contacts/hour, with a median of 4 contacts/hour ( Table 3 ). No statistical differences in the frequency of the microactivities (i.e., hand- and object-to-mouth) were found by these doers by lifestage. Children aged 1–2 years were observed to have a narrower range of hand-to-mouth contact frequency (3–33 contacts/hour) with a higher median frequency of 9 contacts/hour. Children aged 6 to <12 months had higher hourly object-to-mouth contacts (median = 28) than children aged 1 to <2 years (median = 18), but this difference was not statistically significant. No significant linear relationships were observed between object-to-mouth (r = −0.24, p = 0.06) or hand-to-mouth contact frequency (r = 0.001; p = 0.99) and calendar age (in months) ( Fig. S5 ). Table 3. Summary distributions of time-weighted activity frequencies and durations for hand- and object-to-mouth activities, by lifestage. Events per hour (all participants) Age group N Range Mean (SD) p25 p50 p75 p95 Total microactivities 6– < 12 months 38 (0–130) 26 (33) 0 12 39 88 1– < 2 years 23 (0–85) 19 (25) 0 12 33 78 Both hands-to-mouth 6– < 12 months 38 (0–71) 4 (12) 0 0 3 12 1– < 2 years 23 (0–33) 5 (10) 0 0 6 31 Right hand-to-mouth 6– < 12 months 38 (0–15) 1 (3) 0 0 0 4 1– < 2 years 23 (0–6) 1 (2) 0 0 0 3 Left hand-to-mouth 6– < 12 months 38 (0–56) 2 (9) 0 0 0 9 1– < 2 years 23 (0–33) 5 (9) 0 0 4 26 Object-to-mouth 6– < 12 months 38 (0–123) 23 (32) 0 9 34 86 1– < 2 years 23 (0–85) 14 (24) 0 3 18 77 Events per hour (doers only a ) Total microactivities 6– < 12 months 25 (3–130) 40 (34) 12 32 57 95 1– < 2 years 16 (3–85) 28 (25) 11 21 35 83 Both hands-to-mouth 6– < 12 months 12 (3–71) 11 (19) 3 4 11 38 1– < 2 years 9 (3–33) 14 (12) 3 9 17 33 Right hand-to-mouth 6– < 12 months 8 (3–15) 5 (5) 3 3 4 13 1– < 2 years 4 (3–6) 4 (2) 3 3 4 6 Left hand-to-mouth 6– < 12 months 8 (3–56) 12 (18) 3 5 9 40 1– < 2 years 8 (3–33) 13 (11) 5 9 20 31 Object-to-mouth 6– < 12 months 22 (3–123) 39 (33) 12 28 54 96 1– < 2 years 12 (3–85) 27 (28) 12 18 28 83 Duration per contact event (seconds) Total microactivities 6– < 12 months 25 (1–11) 4 (3) 3 4 5 9 1– < 2 years 16 (1–9) 3 (2) 2 4 4 6 Both hands-to-mouth 6– < 12 months 12 (1–9) 3 (3) 1 3 5 7 1– < 2 years 9 (1–4) 2 (1) 1 2 3 4 Right hand-to-mouth 6– < 12 months 8 (1–9) 2 (3) 1 1 2 8 1– < 2 years 4 (1–3) 1 (1) 1 1 1 2 Left hand-to-mouth 6– < 12 months 8 (1–6) 4 (2) 2 4 5 6 1– < 2 years 8 (1–4) 2 (1) 2 3 3 4 Object-to-mouth 6– < 12 months 22 (1–11) 4 (3) 3 4 5 10 1– < 2 years 12 (1–9) 4 (2) 2 4 5 7 Total contact duration per child (seconds) Total microactivities 6– < 12 months 25 (1–352) 72 (90) 13 51 72 271 1– < 2 years 16 (2–300) 48 (76) 7 25 49 179 Both hands-to-mouth 6– < 12 months 12 (1–132) 16 (37) 2 5 10 69 1– < 2 years 9 (1–47) 14 (18) 2 9 10 45 Right hand-to-mouth 6– < 12 months 8 (1–24) 5 (8) 1 1 5 19 1– < 2 years 4 (1–5) 2 (2) 1 1 2 4 Left hand-to-mouth 6– < 12 months 8 (1–108) 19 (36) 4 5 12 76 1– < 2 years 8 (1–47) 15 (17) 2 9 17 43 Object-to-mouth 6– < 12 months 22 (1–352) 73 (93) 13 52 71 279 1– < 2 years 12 (1–300) 53 (86) 8 21 45 211 Duration (minutes/hour) Total microactivities 6– < 12 months 25 (0–18) 4 (5) 1 2 4 13 1– < 2 years 16 (0–13) 2 (3) 0 1 2 8 Both hands-to-mouth 6– < 12 months 12 (0–7) 1 (2) 0 0 0 3 1– < 2 years 9 (0–2) 1 (1) 0 0 1 2 Right hand-to-mouth 6– < 12 months 8 (0–1) 0 (0) 0 0 0 1 1– < 2 years 4 (0–0) 0 (0) 0 0 0 0 Left hand-to-mouth 6– < 12 months 8 (0–6) 1 (2) 0 0 1 4 1– < 2 years 8 (0–2) 1 (1) 0 0 1 2 Object-to-mouth 6– < 12 months 22 (0–18) 4 (5) 1 2 3 13 1– < 2 years 12 (0–13) 2 (4) 0 1 2 10 Open in a new tab a We define a doer as a participant with at least one hand-to-mouth or object-to-mouth contact. The median hand-to-mouth duration for children aged 6–12 months was 5 s compared to 9 s for children 1–2 years ( Table 3 ). Children aged 1 to <2 years had lower object-to-mouth contact event duration (median = 21 s) compared to 6–12 month olds (median = 52 s). No statistical differences ( p > 0.05) in the adjusted duration of the microactivities (i.e., hand- and object-to-mouth) were found by lifestage. Though not statistically significant, both adjusted hand-to-mouth duration and object-to-mouth duration were observed to have a negative linear relationship with calendar age for doers (Pearson’s correlation, r = −0.09, p = 0.71; r = −0.27, p = 0.13; Fig. S5 ). Microactivities by fine and gross motor development Based on caregivers’ responses to their child’s age-specific ASQ-3, 39 (64%) of children in our study were considered to have “sufficient” or “above average” gross motor development ( Table 4 ). Among children who had at least one instance of either hand- or object-to-mouth contact, 27 (66%) of them were considered to have “sufficient” and “above average” gross motor development based on their age-specific ASQ-3. Children with “poor” gross motor development tended to have the highest rate of hourly mouthing activities (hand- and object-to-mouth) (median = 41 contacts/hour). Similarly, among children who had at least one hand-to-mouth contact (i.e., doers), the median rates of hourly hand-to-mouth contacts and median hand-to-mouth contact duration were highest among children with “above average” gross motor development, 8 contacts/hour and 8 s, respectively. Among children who had at least one object-to-mouth contact ( n = 34), those with “poor” gross motor development had the highest median object-to-mouth frequency of 42 contacts/hour. The median object-to-mouth contact duration was highest among children with “poor” gross motor development, being four times longer than that of children with “above average” gross motor skills (91 s vs. 21 s). Table 4. Summary distributions of time-weighted activity frequencies and durations for hand- and object-to-mouth activities, by gross motor development. Microactivity, by gross motor status (all participants) N Range Mean (SD) p25 p50 p75 p95 Total microactivities Very Poor 15 (0–87) 21 (32) 0 6 23 85 Poor 7 (0–82) 30 (30) 4 33 42 71 Sufficient 12 (0–97) 29 (34) 2 18 43 89 Above Average 27 (0–130) 21 (29) 0 12 33 67 Hand-to-mouth Very Poor 15 (0–71) 7 (18) 0 0 3 29 Poor 7 (0–33) 5 (12) 0 0 0 23 Sufficient 12 (0–11) 2 (3) 0 0 3 7 Above Average 27 (0–33) 4 (8) 0 0 4 16 Object-to-mouth Very Poor 15 (0–85) 14 (29) 0 0 12 84 Poor 7 (0–82) 25 (32) 0 9 42 71 Sufficient 12 (0–97) 27 (34) 0 14 41 87 Above Average 27 (0–123) 17 (27) 0 12 22 64 Microactivity, by gross motor status (doers a only) Total microactivities Very Poor 9 (3–87) 35 (35) 8 22 71 86 Poor 5 (9–82) 42 (26) 33 41 43 74 Sufficient 9 (3–97) 38 (34) 11 28 57 91 Above Average 18 (3–130) 32 (30) 12 26 38 85 Hand-to-mouth Very Poor 6 (3–71) 17 (27) 3 6 11 56 Poor 1 (33–33) 33 33 33 33 33 Sufficient 4 (3–11) 5 (4) 3 3 5 10 Above Average 10 (3–33) 10 (9) 4 8 13 26 Object-to-mouth Very Poor 6 (3–85) 36 (38) 7 22 68 84 Poor 4 (9–82) 44 (30) 33 42 53 76 Sufficient 8 (3–97) 40 (34) 15 30 63 91 Above Average 16 (9–123) 29 (30) 12 18 31 87 Duration per contact event (seconds) Hand-to-mouth Very Poor 6 (1–132) 24 (53) 2 4 5 100 Poor 1 (47–47) 47 47 47 47 47 Sufficient 4 (1–17) 8 (7) 3 7 11 16 Above Average 10 (1–41) 9 (12) 2 8 10 28 Object-to-mouth Very Poor 6 (1–300) 112 (149) 9 45 226 296 Poor 4 (8–138) 82 (57) 54 91 119 134 Sufficient 8 (5–352) 81 (116) 14 35 84 271 Above Average 16 (4–195) 37 (47) 10 21 48 97 Total contact duration per child (seconds) Hand-to-mouth Very Poor 6 (1–6) 2 (2) 1 1 3 5 Poor 1 (4–4) 4 4 4 4 4 Sufficient 4 (1–9) 5 (3) 3 4 5 8 Above Average 10 (1–6) 2 (1) 1 2 3 5 Object-to-mouth Very Poor 6 (1–10) 5 (4) 2 6 9 10 Poor 4 (3–8) 5 (2) 4 5 6 8 Sufficient 8 (2–11) 5 (3) 3 4 5 9 Above Average 16 (1–6) 3 (1) 2 3 4 5 Duration (minutes/hour) Hand-to-mouth Very Poor 6 (0–7) 1 (3) 0 0 0 5 Poor 1 (2–2) 2 2 2 2 2 Sufficient 4 (0–1) 0 (0) 0 0 1 1 Above Average 10 (0–2) 0 (1) 0 0 1 1 Object-to-mouth Very Poor 6 (0–14) 5 (6) 0 2 11 14 Poor 4 (0–7) 4 (3) 2 4 6 6 Sufficient 8 (0–18) 4 (6) 1 2 4 14 Above Average 16 (0–10) 2 (2) 0 1 2 5 Open in a new tab a We define a doer as a participant with at least one hand-to-mouth or object-to-mouth contact. Based on caregivers’ responses to their child’s age-specific ASQ-3, 49 (80%) of children in our study were considered to have “sufficient” or “above average” fine motor development ( Table 5 ). Among children who had at least one instance of either hand- or object-to-mouth contact, the majority (85%) of them were considered to have “sufficient” and “above average” fine motor development based on their age-specific ASQ-3. Among children with at least one hand-to-mouth contact, children with “very poor” fine motor skills had the most frequent hand-to-mouth contacts (median = 22 contacts/hour). Children with less developed fine motor skills had significantly longer hand-to-mouth contact duration (median = 32 s) than those with “more developed” fine motor skills (Wilcoxon Rank Sum Test, p = 0.008). Children with “above average” fine motor skills ( n = 16) were observed to have the lowest object-to-mouth contact frequency (median = 21 contacts/hour) and duration (median = 15 s). Table 5. Summary distributions of time-weighted activity frequencies and durations for hand and object to mouth activities, by fine motor development. Microactivity, by fine motor status (all participants) N Range Mean (SD) p25 p50 p75 p95 Very Poor 9 (0–71) 21 (24) 0 22 33 58 Poor 3 (0–82) 27 (47) 0 0 41 74 Sufficient 20 (0–87) 26 (29) 2 12 40 83 Above Average 29 (0–130) 22 (32) 0 12 33 92 Hand-to-mouth Very Poor 9 (0–71) 13 (24) 0 0 11 56 Poor 3 (0–0) 0 (0) 0 0 0 0 Sufficient 20 (0–13) 2 (3) 0 0 3 8 Above Average 29 (0–33) 3 (7) 0 0 3 15 Object-to-mouth Very Poor 9 (0–36) 8(13) 0 0 17 30 Poor 3 (0–82) 27 (47) 0 0 41 74 Sufficient 20 (0–84) 24 (28) 0 12 30 80 Above Average 29 (0–123) 19 (32) 0 3 24 92 Microactivity, by fine motor status (doers a only) Very Poor 5 (22–71) 38 (19) 28 33 38 64 Poor 1 (82–82) 82 82 82 82 82 Sufficient 15 (3–87) 34 (29) 11 24 50 84 Above Average 20 (3–130) 33 (34) 11 22 39 98 Hand-to-mouth Very Poor 4 (3–71) 29 (30) 9 22 42 65 Poor 0 Sufficient 7 (3–13) 5 (4) 3 3 7 11 Above Average 10 (3–33) 10 (10) 3 8 12 26 Object-to-mouth Very Poor 3 (17–36) 25 (10) 19 22 29 34 Poor 1 (82–82) 82 82 82 82 82 Sufficient 14 (3–84) 34 (28) 12 24 54 81 Above Average 16 (3–123) 34 (36) 11 21 39 103 Duration per contact event (seconds) Hand-to-mouth Very Poor 4 (9–132) 51 (56) 15 32 68 119 Poor 0 Sufficient 7 (1–10) 3 (3) 1 2 4 8 Above Average 10 (1–41) 9 (12) 3 6 10 28 Object-to-mouth Very Poor 3 (14–55) 41 (23) 34 53 54 55 Poor 1 (138–138) 138 138 138 138 138 Sufficient 14 (1–283) 56 (73) 11 34 68 178 Above Average 16 (1–352) 74 (111) 9 15 69 313 Total contact duration per child (seconds) Hand-to-mouth Very Poor 4 (4–9) 6 (2) 4 5 7 9 Poor 0 Sufficient 7 (1–4) 2 (1) 1 1 2 4 Above Average 10 (1–6) 2 (2) 1 2 4 5 Object-to-mouth Very Poor 3 (2–7) 4 (2) 3 4 5 7 Poor 1 (5–5) 5 5 5 5 5 Sufficient 14 (1–10) 4 (2) 3 4 4 7 Above Average 16 (1–11) 4 (3) 2 4 5 10 Duration (minutes/hour) Hand-to-mouth Very Poor 4 (0–7) 3 (3) 1 2 3 6 Poor 0 Sufficient 7 (0–1) 0 (0) 0 0 0 0 Above Average 10 (0–2) 0 (1) 0 0 0 1 Object-to-mouth Very Poor 3 (1–3) 2 (1) 2 2 2 3 Poor 1 (7–7) 7 7 7 7 7 Sufficient 14 (0–14) 3 (4) 1 2 3 9 Above Average 16 (0–18) 4 (5) 0 1 3 14 Open in a new tab a We define a doer as a participant with at least one hand-to-mouth or object-to-mouth contact. No statistically significant differences in hand- or object-to-mouth frequency were observed between children with less developed (i.e., “poor” and “very poor”) and more developed (i.e., “sufficient” and “above average”) for both gross and fine motor skills ( p > 0.05). DISCUSSION We used a novel computer vision method to quantify hand and object-to-mouth microactivities from video footage of 61 children playing in their homes. We found that object-to-mouth events were more common than hand-to-mouth events and tended to increase in frequency when children were presented with a new toy (i.e., a novel stimulus). We explored the relationships between microactivities and lifestage and motor development and found that microactivities were more frequent among younger children (6 to <12 months vs. 1 to <2 years) and those with the least gross and most fine motor development. While we noted key descriptive differences in microactivities across these factors, none were statistically significant. It is possible that lifestage, fine or gross motor development or exposure to a new stimulus could impact a child’s rates of microactivity contacts. Compared to other studies, our study reported fewer hand-to-mouth and more object-to-mouth microactivities for children of similar age in indoor environments. Two notable meta-analyses of children’s hand-to-mouth [ 36 ] and object-to-mouth [ 37 ] activities were published in 2007 and 2010, respectively. They rely on 10 studies of children’s microactivities, many of which used human annotation of recorded videos. Xue et al.’s meta-analysis of hand-to-mouth activities found almost twice as many contacts/hour (median = 14 contacts/hour for 6 to <12 month olds and 14 contacts/hour among 1 to <2 year olds) compared to our study (median = 0 contacts/hour for 6 to <12 month olds and 0 contacts/hour among 1 to <2 year olds). This pattern held at the 95th percentile as well; Xue et al. estimated 52 contacts/hour among 6 to <12 month olds and 63 contacts/hour among 1 to <2 year olds. We observed 12 contacts/hour among 6 to <12 month olds and 31 contacts/hour among 1 to <2 year olds. Our study yielded object-to-mouth microactivity counts that were much more variable than the previous meta-analysis. For example, among 6 to <12-month-olds, we observed a median of 9 contacts/hour and 95th percentile of 86 contacts/hour compared to Xu et al.’s estimated median of 19 contacts/hour and 39 contacts/hour. Similarly, among 1 to <2-year-olds, we observed a median of 3 contacts/hour and 95th percentile of 77 contacts/hour compared to Xue et al.’s estimated median of 12 contacts/hour and 34 contacts/hour. Our high rates of object-to-mouth contact may be related to our study design (which deliberately included the child experiencing the introduction of a new toy). While not significant, we did observe higher rates of object-to-mouth contacts during structured play involving the new toy, especially for participants at the higher end of the distribution. Based on this, further investigation may be warranted to explore the influence of novel stimuli (and other environmental factors) on the frequency of object-to-mouth contacts. The EPA relies on these meta-analyses and underlying studies to inform recommended non-dietary ingestion exposure factors for the EPA Exposure Factors Handbook [ 1 ]. Based on this, the agency recommends using means of 20 object-to-mouth contacts/hour for children 6 to <12 months and 14 object-to-mouth contacts/hour for 1 to <2-year-olds ( Table S6 ). These recommended exposure factors are less than our observed mean for 6 to <12-month-olds (23 object-to-mouth contacts/hour) and equal to our observed mean for 1 to <2-year-olds (14 object-to-mouth contacts/hour). The EPA’s recommended values for mean hand-to-mouth contacts are consistently higher than estimates from our study. The EPA recommends using a mean of 19 hand-to-mouth contacts/hour for 6 to <12-month-olds, and 20 hand-to-mouth contacts/hour for 1 to <2-year-olds as default exposure factors. Our study yielded means of 4 and 5 hand-to-mouth contacts/hour for each age group, respectively. In this study, we pioneer a new method for quantifying children’s microactivities that is more efficient and arguably more objective than past studies that use human coders for microactivity annotation. Our computer algorithm determines a contact event by objectively measuring the distance between key points representing the hand and mouth. To this end, our results are less prone to misclassification due to human error. Our study also targeted children between the ages of 6 and 18 months, which includes lifestages previously observed to have the highest rates of hand- and object-to-mouth behaviors [ 1 , 36 , 37 ]. We also observed children in their home environments where their activities are less likely to be impacted by artificial experiences in a lab. An important limitation of our study was the shorter observation time (i.e., up to 20 min during active play) for each participant. Similar studies of children’s microactivities had observation periods of at least 1 h [ 14 , 15 , 38 ] and up to 8 or 10 h on a single day [ 12 ]. To compare our results to other studies, we extrapolated our data to provide hourly rates, assuming that the 20 min we observed were consistent with the remaining 40 min we did not observe. About a third of the children in our study did not engage in either hand- or object-to-mouth activities during the observation period. It is certainly possible (and maybe likely) that, given a longer observation period, these non-doers would engage in at least one microactivity contact event; as a result, we assume that our results underestimate the lower end of the distribution of events. This contributes to uncertainty in extrapolating microactivity contact rates from our data to estimate longer-term (e.g., per day) contact event rates. As children’s play often involves exploration of their environments, we targeted this macroactivity for observation and introduced a new toy as a novel stimulus to explore its influence on microactivity contacts. It is critical to recognize that the rates of microactivities may be different during other non-play macroactivities like eating, bathing, or sleeping and that children may experience exposures to novel stimuli in an array of contexts and at different rates. This also contributes to uncertainty in extrapolating our microactivity rates to represent a single day or longer. We also note that because all caregivers were present and instructed to play with their child as they normally would during video collection, we could not evaluate the impact of the caregivers’ presence on children’s microactivities. Future investigations to examine differences across microactivities and under different conditions and play scenarios are warranted. At enrollment, 15 percent of caregivers reported their child routinely used pacifiers. Because we observed pacifier use in the recordings of only six participants, we had insufficient data to train the computer vision algorithm to identify and quantify pacifier-specific contacts. Pacifier and comfort object use is a unique, but not well-characterized consideration for generating estimates of soil/dust ingestion for children. Pacifiers may occupy the mouth and thus reduce hand- or object-to-mouth contacts [ 11 ] (and thus reduce soil/dust ingestion), but when dropped, soil/dust may also adhere to the pacifier, resulting in additional opportunities for ingestion when used again. One modeling study estimated pacifier use contributed about 20 mg/day to the median dust ingestion estimate for infants [ 6 ]. More data on the frequency, duration, and parenting and cultural practices motivating pacifier use are needed to further assess the potential impact and contribution of pacifier use on soil and dust ingestion. The duration of mouthing events may be important for children’s exposure to soil and dust. Our estimates of object-to-mouth durations were lower than EPA recommendations for both lifestages ( Table S6 ). However, children who mouth their hands or objects for longer periods of time may produce and transfer more saliva, increasing the potential for greater adherence of soil/dust particles to the objects between mouthing events by the child. The EPA Exposure Factors Handbook does not currently provide recommended estimates for the duration of hand-to-mouth events because the algorithm used to estimate exposure is unrelated to time [ 1 ]. Our computer vision method quantified the duration of microactivities, generating a critical input that could advance exposure estimates associated with mouthing behavior, specifically for chemicals that may leach from toys [ 39 - 42 ]. For example, a 2021 study developed a mechanistic material-saliva migration model to estimate the exposure and risk for several chemical-material-specific combinations under average and upper-bound mouthing scenarios [ 43 ]. These models require more complex mouthing-related inputs, including the duration of contact events. Given the importance of objects, a longitudinal tracking of objects and characterization of children’s interactions is warranted. Our microactivity contact and duration data are important inputs into future efforts to develop lifestage specific soil and dust ingestion rates for the U.S. population. These efforts would be further improved by larger, population-based studies to increase characterization of the variability of these factors in diverse populations. More robust distributions of microactivity contact frequencies could significantly improve the confidence of recommendations for soil and dust ingestion rates developed via comprehensive modeling efforts, such as EPA’s Stochastic Human Exposure and Dose Simulation Soil and Dust model (SHEDS-Soil/Dust) [ 6 ]. Supplementary Material Supplementary Material NIHMS2155119-supplement-Supplementary_Material.pdf (644.4KB, pdf) Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41370-025-00800-3 . ACKNOWLEDGEMENTS We thank the parents and their children who participated in this study. We are grateful to our student research assistants, Tionna Tolefree and Sofia Harrison who visited participants’ homes and recorded the video data. FUNDING This project was supported by a grant from the US Environmental Protection Agency: Estimating Children’s Soil and Dust Ingestion Rates for Exposure Science EPA-G2020-STAR-D1. SNL also received financial support from the National Institute of Environmental Health Sciences (NIEHS, Grant ID P30ES032756). Footnotes COMPETING INTERESTS The authors declare no competing interests. ETHICAL APPROVAL This study has been approved by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board (IRB00020023). All methods were performed in accordance with relevant guidelines and regulations. DATA AVAILABILITY The dataset and codebook will be available from the authors upon request. REFERENCES 1. United States Environmental Protection Agency. Exposure Factors Handbook Chapter 4 2011 [Available from: https://www.epa.gov/expobox/exposure-factors-handbook-chapter-4 . 2. United States Environmental Protection Agency. Exposure Factors Handbook Chapter 5 2017 [Available from: https://www.epa.gov/expobox/exposure-factors-handbook-chapter-5 . 3. Xue J, Liu SV, Zartarian VG, Geller AM, Schultz BD. Analysis of NHANES measured blood PCBs in the general US population and application of SHEDS model to identify key exposure factors. J Expo Sci Environ Epidemiol. 2014;24:615–21. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Xue J, Zartarian V, Wang S-W, Liu SV, Georgopoulos P. Probabilistic modeling of dietary arsenic exposure and dose and evaluation with 2003–2004 NHANES data. Environ Health Perspect. 2010;118:345–50. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Xue J, Zartarian V, Tornero-Velez R, Tulve NS. EPA’s SHEDS-multimedia model: Children’s cumulative pyrethroid exposure estimates and evaluation against NHANES biomarker data. Environ Int. 2014;73:304–11. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Ozkaynak H, Glen G, Cohen J, Hubbard H, Thomas K, Phillips L, et al. Model based prediction of age-specific soil and dust ingestion rates for children. J Expo Sci Environ Epidemiol. 2022;32:472–80. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Ferguson AC, Canales RA, Beamer P, Auyeung W, Key M, Munninghoff A, et al. Video methods in the quantification of children’s exposures. J Expo Sci Environ Epidemiol. 2006;16:287–98. [ DOI ] [ PubMed ] [ Google Scholar ] 8. U.S. EPA. Exposure Factors Handbook Chapter 4 2011 [Available from: https://www.epa.gov/expobox/exposure-factors-handbook-chapter-4 . 9. Juberg DR, Alfano K, Coughlin RJ, Thompson KM. An observational study of object mouthing behavior by young children. Pediatrics. 2001;107:135–42. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Tsou MC, Ozkaynak H, Beamer P, Dang W, Hsi HC, Jiang CB, et al. Mouthing activity data for children age 3 to <6 years old and fraction of hand area mouthed for children age <6 years old in Taiwan. J Expo Sci Environ Epidemiol. 2018;28:182–92. [ DOI ] [ PubMed ] [ Google Scholar ] 11. Tsou MC, Ozkaynak H, Beamer P, Dang W, Hsi HC, Jiang CB, et al. Mouthing activity data for children aged 7 to 35 months in Taiwan. J Expo Sci Environ Epidemiol. 2015;25:388–98. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Zartarian VG, Streicker J, Rivera A, Cornejo CS, Molina S, Valadez OF, et al. A pilot study to collect micro-activity data of two- to four-year-old farm labor children in Salinas Valley, California. J Expo Anal Environ Epidemiol. 1995;5:21–34. [ PubMed ] [ Google Scholar ] 13. Reed KJ, Jimenez M, Freeman NC, Lioy PJ. Quantification of children’s hand and mouthing activities through a videotaping methodology. J Expo Anal Environ Epidemiol. 1999;9:513–20. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Beamer P, Key ME, Ferguson AC, Canales RA, Auyeung W, Leckie JO. Quantified activity pattern data from 6 to 27-month-old farmworker children for use in exposure assessment. Environ Res. 2008;108:239–46. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Ferguson A, Dwivedi A, Adelabu F, Ehindero E, Lamssali M, Obeng-Gyasi E, et al. Quantified activity patterns for young children in beach environments relevant for exposure to contaminants. Int J Environ Res Public Health. 2021;18. https://www.mdpi.com/about/announcements/784 . [ Google Scholar ] 16. Beamer PI, Luik CE, Canales RA, Leckie JO. Quantified outdoor micro-activity data for children aged 7-12-years old. J Expo Sci Environ Epidemiol. 2012;22:82–92. [ DOI ] [ PubMed ] [ Google Scholar ] 17. Auyeung W, Canales RA, Beamer P, Ferguson AC, Leckie JO. Young children’s hand contact activities: an observational study via videotaping in primarily outdoor residential settings. J Expo Sci Environ Epidemiol. 2006;16:434–46. [ DOI ] [ PubMed ] [ Google Scholar ] 18. Asan O, Montague E. Using video-based observation research methods in primary care health encounters to evaluate complex interactions. Inf Prim Care. 2014;21:161–70. [ Google Scholar ] 19. Lupolt SN, Zhang G, Wang J, Tang S, Cho J, Lyu Q, et al. Development and evaluation of computer vision algorithm for quantification of children’s microactivities. J Expo Sci Environ Epidemiol. 2025. (in press). [ Google Scholar ] 20. United States Environmental Protection Agency. Childhood Lifestages relating to Children’s Environmental Health 2025. [Available from: https://www.epa.gov/children/childhood-lifestages-relating-childrens-environmental-health . 21. Cohen Hubal EA, de Wet T, Du Toit L, Firestone MP, Ruchirawat M, van Engelen J, et al. Identifying important life stages for monitoring and assessing risks from exposures to environmental contaminants: Results of a World Health Organization review. Regul Toxicol Pharm. 2014;69:113–24. [ Google Scholar ] 22. Cohen Hubal EA, Moya J, Selevan SG. A lifestage approach to assessing children’s exposure. Birth Defects Res Part B: Dev Reprod Toxicol. 2008;83:522–9. [ Google Scholar ] 23. Groot EML, MC, Steenbekkers LPA Mouthing behaviour of young children; An observational study (summary report). RIVM report 613320 002 RIVM Bilthoven, The Netherlands, September 1998. [ Google Scholar ] 24. Feldman DH Beyond universals in cognitive development, 2nd ed. Westport, CT, US: Ablex Publishing; 1994. [ Google Scholar ] 25. BuildClinical. 2025. [Available from: https://www.buildclinical.com/ . 26. Squires J, Bricker D Ages and Stages Questionnaires: A Parent-Completed Child Monitoring System: Paul H. Brookes Publishing Co. Inc.; 2009. [ Google Scholar ] 27. Szeliski R Computer vision: algorithms and applications: Springer Nature; 2022. [ Google Scholar ] 28. Ren T, Liu S, Zeng A, Lin J, Li K, Cao H, et al. Grounded sam: Assembling openworld models for diverse visual tasks. arXiv preprint arXiv:240114159. 2024. [ Google Scholar ] 29. MMPose Contributors. Openmmlab pose estimation toolbox and benchmark 2020. [Available from: https://github.com/open-mmlab/mmpose . 30. Sun K, Xiao B, Liu D, Wang J, editors. Deep high-resolution representation learning for human pose estimation. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition; 2019. [ Google Scholar ] 31. Jin S, Xu L, Xu J, Wang C, Liu W, Qian C, et al. , editors. Whole-body human pose estimation in the wild. Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part IX 16; 2020: Springer. [ Google Scholar ] 32. Kirillov A, Mintun E, Ravi N, Mao H, Rolland C, Gustafson L, et al. , editors. Segment anything. Proceedings of the IEEE/CVF International Conference on Computer Vision; 2023. [ Google Scholar ] 33. Shuai Q, Geng C, Fang Q, Peng S, Shen W, Zhou X, et al. EasyMocap-make human motion capture easier. Github. 2021;1:6. [ Google Scholar ] 34. He K, Zhang X, Ren S, Sun J, editors. Deep residual learning for image recognition. Proceedings of the IEEE conference on computer vision and pattern recognition; 2016. [ Google Scholar ] 35. Squires J, Twombly E, Bricker D & Potter L ASQ-3 User’s Guide. Baltimore, MD: Paul Brookes Publishing.; 2009. [ Google Scholar ] 36. Xue J, Zartarian V, Moya J, Freeman N, Beamer P, Black K, et al. A meta-analysis of children’s hand-to-mouth frequency data for estimating nondietary ingestion exposure. Risk Anal. 2007;27:411–20. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Xue J, Zartarian V, Tulve N, Moya J, Freeman N, Auyeung W, et al. A meta-analysis of children’s object-to-mouth frequency data for estimating non-dietary ingestion exposure. J Expo Sci Environ Epidemiol. 2010;20:536–45. [ DOI ] [ PubMed ] [ Google Scholar ] 38. Freeman NC, Jimenez M, Reed KJ, Gurunathan S, Edwards RD, Roy A, et al. Quantitative analysis of children’s microactivity patterns: the Minnesota Children’s Pesticide Exposure Study. J Expo Anal Environ Epidemiol. 2001;11:501–9. [ DOI ] [ PubMed ] [ Google Scholar ] 39. Xie M, Wu Y, Little JC, Marr LC. Phthalates and alternative plasticizers and potential for contact exposure from children’s backpacks and toys. J Expo Sci Environ Epidemiol. 2016;26:119–24. [ DOI ] [ PubMed ] [ Google Scholar ] 40. Guney M, Zagury GJ. Contamination by ten harmful elements in toys and children’s jewelry bought on the North American Market. Environ Sci Technol. 2013;47:5921–30. [ DOI ] [ PubMed ] [ Google Scholar ] 41. Brandsma SH, Leonards PEG, Koekkoek JC, Samsonek J, Puype F. Migration of hazardous contaminants from WEEE contaminated polymeric toy material by mouthing. Chemosphere. 2022;294:133774. [ DOI ] [ PubMed ] [ Google Scholar ] 42. Ionas AC, Dirtu AC, Anthonissen T, Neels H, Covaci A. Downsides of the recycling process: Harmful organic chemicals in children’s toys. Environ Int. 2014;65:54–62. [ DOI ] [ PubMed ] [ Google Scholar ] 43. Aurisano N, Fantke P, Huang L, Jolliet O. Estimating mouthing exposure to chemicals in children’s products. J Expo Sci Environ Epidemiol. 2022;32:94–102. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Material NIHMS2155119-supplement-Supplementary_Material.pdf (644.4KB, pdf) Data Availability Statement The dataset and codebook will be available from the authors upon request. ACTIONS View on publisher site PDF (694.2 KB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top