Research On Pedestrian Detection Algorithm Under Occlusio

As a branch of target detection,pedestrian detection has a strong development momentum and has shown great technical value in various fields of intelligence.This technology can not only provide security for the development of the intelligent field,but also one of the core technologies in today’s traffic safety system.Pedestrian detection algorithms emerge in an endless stream,but most of them have good results in simple environments with few people.It is powerless for severely occluded environments,and occlusion is the bottleneck that affects the performance of pedestrian detection.In this paper,an attention-enhanced pedestrian and head-shoulders cascade detection algorithm is proposed.The experiment is based on the R-FCN model of the ResNet50+DCN network.Firstly,a new channel attention mechanism called fully convolutional channel attention mechanism is proposed.Secondly,suitable channel and spatial attention mechanisms are integrated for classification and regression tasks to enhance effective detection features and suppress background feature information.Finally,a pedestrian detector is designed in which the whole pedestrian is cascaded with the pedestrian head and shoulders area,and the pedestrian head and shoulders area detection branch is added on the basis of the pedestrian detection model.For undetected pedestrians,the head and shoulders regions are detected,and then the pedestrian’s overall bounding box is generated according to the pedestrian’s inherent body structure ratio,so as to improve the detection effect of mutual occlusion between pedestrians.Especially in the case of severe occlusion of the lower body,the missed detection rate of occluded pedestrians is greatly reduced.The experimental results show that the logarithmic average missed detection rate on the Reasonable(reasonable subset)of the Caltech dataset is reduced to 5.37%,especially the logarithmic average missed detection rate on Occ=heavy(severely occluded subset).to 23.33%.At the same time,the algorithm also has good detection performance on ETH pedestrian detection datasets.Compared with the current pedestrian detection algorithms with better performance,the method proposed in this paper has strong robustness and high recall rate.In order to combine algorithm with practice,this paper designs a pedestrian detection system.The pedestrian detection system uses the pedestrian detection algorithm in this paper to judge the pedestrian and regression the position of the input picture,which can be embedded in various detection systems,or can be used as an independent client software.