Analyzing the Effects of Human Detection in Top-Down Pose Estimation for Crowd Situation Recognitions

A comprehensive analysis of the human detection within the CrowdPose scheme [1], a representative top-down approach proposed to apply human pose estimation to crowd situations, was conducted. The results of this performance analysis can prove to be invaluable for designing multi-person pose recognition systems with the aim of identifying abnormal events in crowd situations. As the candidate object detectors, YoloV3, YoloX, and Faster R-CNN are selected and used, which are representative detectors used in existing crowd-related research. Various analyses were performed using 8,000 crowd-situation test images provided by the CrowPose research team and the detailed analysis results have been presented.