Fast Human Detection Based On Feature Of Gradients And Cascade Classification

As a important research direction of target detection field , Pedestrian Detection Technology has become a hot research topic recent years. Pedestrian Detection Technology has become a core technology and has a wide range of applications in area of the Intelligent Vehicle, automatic navigation, motion analysis and advanced human-machine interface. Traditional detection methods have some defects and shortcomings. Some of the current research methods mainly use statistical machine learning. Machine learning consists of two stages: extract image features firstly, train human body model from the training database secondly. The database includes a large number of positive and negative training samples. So the pedestrian detection speed has been improved.Based on the gradient orientation histogram features (HOG) proposed by Dalal and boosted cascade algorithm proposed by Viola, we combine these both of them and apply them to the pedestrian detection in this paper. In order to reduce the detection complexity, the detection time and improve detection performance, we have them completed in several aspects: a) complete the one-step cascade detection algorithm. b) complete the four-step cascade detection algorithm, which speeds up the detection speed compared to the original. c) under a boosted cascade algorithm framework, discover the feature point distribution in the image,which can guide the study of the pedestrian detection algorithm.In this paper, we completed the algorithm above, using the INRIA static upright human database. The experiment proved that the performance of the detection system in the paper is close to that of the current level of pedestrian detection system. When FPPW(false positive per window) is 10-3, the detection rate of our detector can reach 87%. For a 240×320 image, the detection speed of our detection algorithm is twice more than support vector machine algorithm.