Pedestrian Detection And Pedestrian Recognition System Based On Deep Convolutional Neural Network

In recent years,with the development of self-driving car and intelligent video surveillance,the amount of video data generated has increased,and the pedestrian detection and pedestrian recognition technology in video has received more and more attention from researchers.In this context,this paper designs a pedestrian detection and pedestrian recognition system based on deep convolutional neural network.The specific work content is summarized as follows:1)This paper designs a deep convolutional neural network model based on YOLO to realize pedestrian detection.The model uses the convolutional neural network to extract features,and outputs the probability and category of the regression target to realize the detection of the human target in the complex scene.The experimental results show that the deep convolutional neural network model has real-time and robustness.2)In order to effectively identify pedestrians,this paper constructs a full convolutional siamese network model based on the similarity measurement idea to realize pedestrian tracking.Firstly,the natural video frame under the camera is used as the image to be processed;Secondly,the pedestrian target is detected on the image by a deep convolutional neural network;Then using the full convolution siamese network to predict and track the target between adjacent frames;Finally,it is determined by the coincidence ratio of the obtained detection target frame and the tracking candidate frame to judge whether the same target;In addition,due to the possibility of detection and tracking errors,this paper also uses the cosine similarity matching method as the second target matching method.The experimental results show that the full convolution siamese network model has a good tracking effect.3)In order to accurately identify pedestrians,this paper proposes two recognition methods based on the traced pedestrian image sequences.The first method is the gait recognition method: Firstly,based on the traced pedestrian image sequence,the DNN pose estimation algorithm is used to extract the key nodes of the human body from the tracked pedestrians;Then,the key node of the pedestrian is obtained by the resnet network and the LSTM algorithm to obtain the gait feature of the pedestrian including the spatial and temporal information;Finally,the SVM algorithm is used to classify the gait features to achieve the purpose of pedestrian gait recognition.The second method is face recognition: Firstly,based on the Dlib face detector,the multi-angle face detection model is trained by CNN to obtain a multi-angle detector of the face;Secondly,the PCA algorithm is used to construct the face feature extraction network.Finally,the SVM algorithm is used to classify the face,so as to realize face recognition under natural walking conditions.The experimental results show that pedestrians can be identified without the active cooperation of pedestrians in the natural walking state.4)According to the actual needs,this paper designs a pedestrian detection and pedestrian recognition system based on the above algorithm under natural walking conditions.The system can complete video acquisition,pedestrian detection,face detection,pedestrian tracking and pedestrian identification.In addition,the paper gives the operation flow of the system in detail,and carries out related tests,which shows that the system has good application and promotion value in practical application.