Detecting human and judging their pose in video

Human is often the most important role in surveillance scenes. How to detect intact human and judge their pose correctly is discussed. Moving human is firstly extracted by background subtraction method in HSV colorspace. In order to alleviate disturbances from different human´s motion poses, the upper half of human is projected to the horizon axis. Utilizing the upper projection histogram, the major axis is located. Walking poses of human are also judged according to the upper projection histogram by introducing a weighting kernel function. Experiments show that our method can detect moving human successfully and can improve the accuracy of locating human´s major axis, also can judge human´s walking pose by 94% accuracy. The algorithm can be implemented with completely real-time speed.