Research On Pedestrian Detection Algorithm Based On Vehicle Camera

With more and more applications of artificial intelligence in daily life,the phenomenon of using depth learning in the field of target detection is increasing day by day.The form of extracting image features has been gradually converted from early artificial feature extraction to using convolutional neural networks to extract features.As one of the important branches of target detection based on computer vision,pedestrian detection algorithm has been widely used in various industries and fields.Unmanned driving,as the number one research goal in the field of artificial intelligence,has attracted strong attention all over the world because of its broad application prospects.In recent years,the development of deep learning technology has promoted the continuous progress of target detection algorithms,which has made significant progress and breakthroughs in unmanned driving technology in the commercial field.The application of deep learning to pedestrian detection technology in driverless cars has been an inevitable trend in the development of this technology.Although many pedestrian detection methods have been proposed at present,pedestrian detection is still a challenge due to the complexity of environmental background,the different kinds of pedestrian postures and the universality of pedestrian occlusion,which requires more accurate algorithms.To solve these problems,this thesis proposes an improved pedestrian detection algorithm MCDET.The algorithm has been improved in three places on M2 DET network to solve the target scale problem.Firstly,the feature extraction network VGG network is improved to make the extracted features more representative;Secondly,the original 8 TUM modules of M2 DET network are reduced to 4,which can greatly reduce the operation parameters,thus effectively improving the running speed of the network;Finally,the convolution block attention mechanism is added to the network to improve the ability of extracting shallow information,and then it is fused with depth features to enrich the detection feature layer.Through the improvement of the network in this thesis,the MCDET pedestrian detection algorithm proposed in this thesis can meet the accurate detection of pedestrians in actual traffic scenes.The experimental results on the current mainstream public data set of pedestrian detection show that,the depth neural network model designed in this thesis has a small improvement in the detection accuracy of pedestrian targets to a certain extent.The experimental results also verify that this method has higher accuracy and faster speed,meets the real-time requirements,and has good robustness and generalization.In this thesis,the detection is also carried out on the collected pedestrian data set,the test results show that the proposed pedestrian detection algorithm can meet the detection requirements of the actual traffic environment.