| Pedestrian Detection, which has been widely used in intelligent transportation, robotics development, intelligent video surveillance and other fields, is the hotspot research topic and one of the difficult issues in computer vision. In this paper, beginning with weighting precision and speed and solving the partial occlusion. Using the theory of image processing, computer vision and other fields, I proposed my own pedestrian detection algorithm based on the existing algorithm.Firstly, according to the problem of the pedestrian detection algorithm based on HOG-LBP features, which detection speed is slow and time can’t meet the real-time. By optimizing the HOG features and introducing two layers of structure detection, I proposed pedestrian detection algorithm based on multiple features cascade. First of all, optimized HOG has been used to detect that can quickly eliminate most negative samples. Second multi-features have been used to detect that can remove hard samples. Then the speed of detection has been improved with the accuracy of detection guaranteed.Secondly, by optimizing the HOG features and advantage of the overall detector and part detector respectively, I proposed the pedestrian detection algorithm based on occlusion reasoning. First of all, I judged the overall detector detection score whether the score in the range of possible occlusion. The second, improved occlusion reasoning algorithm has been used to determine the window, which the score in the range if there are occlusions. The last, the DPM model has been used to detect the occluded window that the pedestrian is detected in the window. Then the accuracy of detection has been improved.Finally, by comparing with existing methods, the pedestrian detection based on multi-feature cascade is verified that the effectiveness of improving detection speed with the accuracy of detection guaranteed, the pedestrian detection based on occlusion reasoning is verified that the feasibility of solving partial occlusion. |
PDF Full Text Request