Beyond Human Detection: A Benchmark for Detecting Common Human Posture

Human detection is the task of locating all instances human beings present in an image, which has a wide range applications across various fields, including search and rescue, surveillance, autonomous driving. The rapid advancement computer vision deep learning technologies brought significant improvements detection. However, for more advanced like healthcare, human–computer interaction, scene understanding, it crucial to obtain information beyond just localization humans. These require deeper understanding behavior state enable effective safe interactions with humans environment. This study presents comprehensive benchmark, Common Postures (CHP) dataset, aimed at promoting informative encouraging mere benchmark dataset comprises diverse collection images, featuring individuals different environments, clothing, occlusions, performing postures activities. aims enhance research this challenging by designing novel precise methods specifically it. CHP consists 5250 images collected from scenes, annotated bounding boxes seven common poses. Using well-annotated we have developed two baseline detectors, namely CHP-YOLOF CHP-YOLOX, building upon identity-preserved posture detectors: IPH-YOLOF IPH-YOLOX. We evaluate performance these detectors through extensive experiments. results demonstrate that effectively detect on dataset. By releasing aim facilitate further pose estimation attract researchers focus task.