Multi-scale Human Detection Based on Window Gradient Potential Energy with Partial Occlusion Handling

Ouyang Yi,Sanyuan Zhang,Zhang Yin

DOI: https://doi.org/10.3724/SP.J.1146.2011.00827

2012-01-01

Abstract:In order to improve accuracy of the human detection,this paper proposes the conception of the Window edge of the Gradient of Potential Energy(WGPE) and a fast human detection method based on potential energy. By using sparse-dense gradient potential windows set,the detection time of the multi-scale detection can be shortened.Cascading Support Vector Machine(SVM) training using weighted positive and negative samples,the occlusion sample of the human body is weighted to detect the human body under occlusion.Filter positive in the detection window,the algorithm does not require too much computational overhead increases when the detection window is filtered.In the smooth background image,the proposed method compared to the multi-level Histograms of Oriented Gradients(HOG) detection and HOG-LBP(Local Binary Pattern) methods accuracy at the same rate, spents less testing time.Experiments show that the human detection accuracy and efficiency has increased,the case for the human body in partial occlusion detection,the accuracy rate is improved markedly.