High-Tech Video Capture and Analysis for Counting Park Users

Abstract

Today’s technology could contribute substantially to measuring physical activity. The current study evaluated traditional and novel approaches for assessing park use. The traditional approach involved a trained observer performing the System for Observing Play and Recreation in Communities (SOPARC) at 14 parks while wearing a point-of-view, wearable video device (WVD). The novel approach utilized computer vision to count park users in the WVD videos taken during in-person SOPARCs. Both approaches were compared to criterion counts from expert reviews of the WVD videos. In the 676 scans made during in-person SOPARCs, 293 individuals were observed while 341 were counted by experts in the corresponding WVD videos. When using scans/videos having individuals in them (84 scans/videos), intra-class correlations (ICC) indicated good-to-excellent reliability between in-person SOPARC and experts for counts of total women and men, within age groups (except seniors), of Blacks and Whites, and within intensity categories (ICCs > .87; p < 0.001). In a subsample of 42 scans/videos, 174 individuals were counted using computer vision and 213 by experts. When using 27 of the 42 WVD videos with individuals in them, ICCs indicated good reliability between computer vision and expert reviews (ICC = .83; p < 0.001). Bland-Altman analysis showed the concurrence of expert counts with both in-person SOPARC and computer vision counts decreased as the number of individuals in a scan/video increased. The results of this study support the use of a highly discrete method for obtaining point-of-view videos and the application of computer vision for automating the counting of park users in the videos.

Field Test of a Passive Infrared Camera for Measuring Trail-Based Physical Activity

“…For example, with cameras located at every entry point of a trail system, it becomes possible to track each user and estimate PA FITT in that setting by using unique features (e.g., facial features, clothing) to distinguish individual trail users. Identifying specific people and coding images will require technological innovation, such as using computer vision algorithms ( 36 , 37 ). By achieving this capability, researchers could answer Koohsari et al ( 4 ) call for evidence of causal inferences.…”

“…Systematic observation to assess PA in parks and recreation areas has evolved through the often-used System for Observing Play and Recreation in Communities (SOPARC) ( 54 – 57 ). Initially reliant on in-person observers using hand-written forms, SOPARC has evolved to incorporate iPads (iSOPARC) ( 58 ), wearable video detection devices ( 37 ), wide-angle cameras mounted on an elevated tripod ( 36 ), and remotely controlled aerial vehicles (drones) for data collection ( 59 ).…”

“…We adapted agreement measures from SOPARC studies ( 36 , 37 ) to fit our data collection methods which are continual over a period of time rather than snapshots of scanned target areas of a larger setting (i.e., park) ( 54 ). With limited guidance on the “best practice” of how to establish agreement and concordance among the devices, this study's findings suggest there is a need to establish a method to assess the validity of PIC counts against infrared sensor with concurrent in-person counts in controlled and real-world scenarios, following guidance of similarly purposed SOPARC studies ( 32 , 36 , 37 ). Calculating agreement based on a modified technique often used in qualitative methods demonstrated to be insufficient.…”

“…Descriptive analyses included frequencies with valid percentages for categorical or count indicators and measures of central tendency with variance for continuous indicators. To explore the agreement or consistency (i.e., concordance) between the hourly counts from the infrared sensor and hourly user counts derived from the PIC photos, this study followed methods outlined by Johnson and Waller ( 35 ) and similar to those used in studies of systematic observation methodologies of PA in park spaces ( 36 , 37 ). Specifically, Bland–Altman plots ( 38 ), paired t -test, Concordance Correlation Coefficient (CCC) ( 39 , 40 ), and Kendall's Tau-b ( 41 ) were used to test for agreement and bias between instruments.…”

A Direct Observation Video Method for Describing COVID-19 Transmission Factors on a Micro-Geographical Scale: Viral Transmission (VT)-Scan

“…To address some of these limitations, the current study developed an observation protocol that uses a high-resolution, wearable video device (WVD) to capture data on COVID-19 transmission behaviors and related factors across a wide range of environmental settings. The use of video has been shown to improve the accuracy of observations when studying physical activity behavior [25,26]. In addition, this study provides a comprehensive set of rules for extracting data from video that is vital for scientific rigor and reproducibility.…”