Font Size: a A A

Research On Pedestrian Detection Algorithm In Complex Scene

Posted on:2019-06-05Degree:MasterType:Thesis
Country:ChinaCandidate:X H ShangFull Text:PDF
GTID:2428330566495889Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
The task of pedestrian detection is to determine whether a given image contains pedestrian targets or not.If there are pedestrians,it gives location coordinates,which has wide application prospects.Pedestrian detection algorithm based on artificial design features can not effectively solve pedestrian detection problems in complex scenes,resulting in serious misdiagnosis and missed detection.However,the performance of pedestrian detection algorithm based on deep learning is not ideal in complex scenes.In view of the above problems,this paper has done the following research work:1.In the light of the pedestrian detection algorithm based on artificial design feature can not effectively solve pedestrian detection in complex scenes,Pedestrian detection algorithm based on multilevel feature fusion is proposed.Based on the existing network structure,the improved network model combining multi level feature of convolutional neural network.The algorithm has better detection performance in the datasets which has more complex scene,and has better detection performance and robustness.2.According to the selective search algorithm generate low quality candidate regions with a low speed,a pedestrian candidate region redundant algorithm is proposed,which use multi strategy fusion to generate higher quality pedestrian candidate areas to improve the detection performance of the algorithm.By combining the Binarized Normed Gradients algorithm and the pedestrian detector,the redundancy of the candidate region is reduced.The experiment shows that a high quality pedestrian candidate area can be produced at a fast speed for the multi strategy fusion method for the pedestrian detection task.3.In order to further reduce the impact of complex background on algorithm performance and effectively utilize the semantic information of the image itself,a pedestrian detection algorithm based on the segmentation semantic optimization is proposed.Using pedestrian mask generated by semantic segmentation algorithm to correct the score of pedestrian boxes detected by pedestrian detection algorithm.Experiments show that the algorithm can get higher detection accuracy on the test set and reduce the impact of complex background on the performance of the algorithm.
Keywords/Search Tags:pedestrian detection, feature fusion, deep network, candidate region selection, complex background
PDF Full Text Request
Related items