Design And Implemention Of Pedestrian Detection And Early Warning System Based On Mounted Camera

With the continuous growth of car ownership in China,traffic safety issues are becoming more and more prominent.Pedestrians are the most vulnerable as vulnerable groups in road traffic.The detection and early warning of pedestrians in front of vehicles is an important way to ensure pedestrian safety.Therefore,pedestrian detection and early warning based on vehicle environment has significant research value and application prospects.Based on the single camera to obtain the road information in front of the vehicle,this paper conducts related research on pedestrian detection and safety warning in front of the vehicle.The main work is as follows:First of all,this paper studies the pedestrian foreground segmentation algorithm.Aiming at the problem of large computational complexity of traditional multi-scale sliding window segmentation pedestrian candidate frame,an efficient BING detection model is introduced,and related improvements are made.The pedestrian scale feature is integrated into the model,enabling it to be quickly initially locate the pedestrian candidate box.Then,for the problem that the candidate frame has an offset,a candidate frame position correction method based on pedestrian edge information is proposed.The results show that the proposed algorithm greatly reduces the number of pedestrian candidate frames compared with the traditional multi-scale sliding window method and the original BING model,which lays a foundation for the rapid extraction and classification of pedestrian features in the next stage.Secondly,aiming at the problem of high HOG feature miss detection rate and high vector dimension,the HOG is improved and enhanced,and the PHOG feature with low feature vector dimension is constructed.Then based on the improved PHOG feature,a two-level pedestrian detection algorithm comBining the detection of the whole and part features is proposed.The experimental results show that the proposed algorithm is robust to occlusions and scenes with large changes in pedestrian attitude.Then,the pedestrian tracking algorithm is studied.A spatio-temporal context tracking algorithm comBined with particle filtering is proposed.Aiming at the problem of spatio-temporal context tracking failure when the tracking target is completely occluded,a pedestrian occlusion discrimination method based on HSV histogram similarity is proposed to trigger particle filtering by judging whether the pedestrian is completely occluded and then retrack pedestrians.Experimental results show that the improved algorithm inherits the advantages of spatio-temporal context tracking algorithm and enhances the robustness of the algorithm to occlusion.Finally,based on the vehicle environment,a pedestrian anti-collision warning model and a monocular ranging model are established,and the above algorithm is integrated to construct a pedestrian detection and early warning system.The system is tested and verified by the vehicle camera video.The results show that the system can detect the pedestrians in front of the vehicle and make...