Pedestrian Detection By Vehicle Video Based On Deep Neural Network

In the field of video intelligent surveillance,aided driving,pedestrian detection technology has been a hot research.It is important for the pedestrian tracking,behavior pattern analysis in the video intelligent monitoring.The premise is to accurately detect the person based on video information.The pedestrian is the first recognition goal in the field of auxiliary driving and automatic driving.Pedestrian detection technology based on vehicle video is getting more and more attention.However,there are also great challenges to the pedestrian detection technology based on video,such as real-time background,different shape,different light intensity,and mutual occlusion.The core of pedestrian detection algorithm is the feature extraction,and the pedestrian features include the features of artificial setting and the features based on deep learning.The advantages of manual setting are simple and direct,such as the histogram of gradient direction,the color channel self-similarity and so on,but it is not easy to define fast robust features,and the expression of artificial features is limited,and the detection accuracy is not high.In recent years,the feature extraction method based on deep learning has received more and more attention,and its feature description ability is strong,and the classification accuracy is high,but the model is always complex,the training time is long,and it often has overfitting problems.In this paper,a pedestrian detection algorithm based on the deep neural network is used to combine the artificial setting features and the deep learning method,which makes use of the advantages of the deep network model description ability to improve the detection accuracy.In addition,the method based on the geometric constraint conditions is used to reduce the number of candidate windows in the region segmentation.In the target recognition part,this paper optimizes the neural network’s excitation function,and improves the convergence effect of the network parameters.CUDA is used to speed up the GPU algorithm,and the efficiency of the algorithm is improved in this paper.Comparing the experimental results with other pedestrian detection algorithms,it can be found that the algorithm of this paper has low false detection rate and high accuracy.