Counting Human Actions in Video During Physical Exercise
We present a simple yet effective human action detection and counting scheme during physical exercise using video stream data. Counting human actions automatically is more meaningful for data analysis in sports centers, and for healthiness observations in rehabilitation centers. The identification of the action starts with the detection of crucial body regions, namely skeletal joints. We observed that hand-wrist, arm-elbow, and arm-shoulder points are crucial for human arm motion, whereas during leg motion ankle-knee, knee-waist, and waist-ankle points are critical. These body junctions get different angle values during physical exercise, which helps us track and count the action. We assumed a simple, cheap and effective solution for multi-tracking these joints, which are marked with a distinctive color. Color filtering and color-based tracking steps are then performed to detect and count the actions by tracing the angle variations between joints. The developed application and performance evaluation tests show that our technique provides a reasonable performance while providing a simple and cheap video setup.