View Distance Based Human Motion Analysis

This paper presents a novel approach is described for realtime human/vehicle classification and motion analysis in real visual surveillance scene. Spatio-temporal 1-D signals based on the distances between the outer contour of binarized silhouette of a motion object and a bounding box placed around the silhouette are chosen as the basic image features called the distance vectors. The spatio-temporal distance vectors are extracted using four view directions to the outer of the silhouette from the bounding box, they are top-, bottom-, left-, and right-views. Correlation-based a similarity function in the time domain is calculated to classify the motion objects and a similarity function in the frequency domain is then also extracted to analysis human motions. Experimental results on the different test image sequences demonstrate that the proposed algorithm has an encouraging performance with relatively robust and low com-