Human Detection using HOG Features of Head and Shoulder Based on Depth Map
Qing Tian, Bo Zhou, Wenhua Zhao, Yun peng Wei, Wei-wei Fei · Journal of Software · 2013
Conventional moving objects detection and tracking using visible light image was often affected by the change of moving objects, change of illumination conditions, interference of complex backgrounds, shaking of camera, shadow of moving objects and moving objects of self-occlusion or mutual-occlusion phenomenon. We propose a human detection method using HOG features of head and shoulder based on depth map and detecting moving objects in particular scene in this paper. In-depth study on Kinect to get depth map with foreground objects. Through the comprehensive analysis based on distance information of the moving objects segmentation extraction removal diagram of background information, by analyzing and comprehensively applying se g mentation a method based on distance information to extract pedestrian’s Histograms of Oriented Gradients (HOG) features of head and shoulder [1] , then make a comparison to the SVM classifier. SVM classifier isolate regions of interest (features of head and shoulder) and judge to achieve real-time detection of objects (pedestrian). The human detection method by using features of head and shoulder based on depth map is a good solution to the problem of low efficiency and identification in traditional human detection system. T he detection accuracy of our algorithm is approximate at 97.4% and the average time processing per fram e is about 51.76 ms.