Object recognition for human behavior analysis

This paper discusses the deformable part-based models for object detection in low contrast images. The objects wheeled walker, walking frame and chair are chosen for the activities walking, sitting and standing. Relationships between detected objects and persons are indicators for those activities. Hence, we enhance a stereo vision system for the purpose of high-level behavior analysis. In order to train models of the objects using the algorithm and get an optimum performance, a sufficient set of images was recorded and annotated. For evaluation, precision and recall curves are reported.