A real-time upper limb motion estimation system

Detection and estimation of upper limb motion is a challenge issue in real-time human-machine interaction. Realtime processing and naturality are key requirements when designing the system. In this paper, we introduce an efficient system to detect and recognize the limb motion of human beings. Considering naturality, we only employ one video camera without any auxiliary marking tools while establishing the system. A modified frame difference approach is presented to detect the limb motion. We then model the upper limb as a straight line to simplify the modeling. Based on the model, histograms for each type of motion are generated and analyzed to evaluate the motion effectively. Multiple experimental results on test sequences demonstrate the high performance of our system.