Research On Pedestrian Detestrian Based On Deep Neural Network

With the rapid development of deep learning,it is widely used in many fields such as computer vision,which can achieve image classification,target location and detection tasks.Pedestrian detection is one of the main research hot pots in the field of computer vision.It plays a very important role in intelligent monitoring,unmanned driving,intelligent robots and other fields.It is also the basis of human posture estimation,pedestrian tracking and other pedestrian related visual tasks.The task of pedestrian detection is to detect the pedestrian from the given image and get the specific location information.The accuracy of pedestrian detection has been greatly improved when the application of deep learning technology in pedestrian detection,but there are still some problems.First,the deep neural network detection algorithm is not accurate and has low recall rate when detecting pedestrians.Second,under different environments,pedestrians have occlusion,which makes the feature extraction effect of pedestrians poor and makes the occlusion of pedestrians more challenging in pedestrian detection.In view of the above problems,The main research contents are as follows:(1)In the pedestrian detection method based on deep neural network,related theories of deep neural network and commonly used pedestrian detection algorithms are elaborated.Finally,YOLO detection algorithm is adopted in this paper and improved to realize pedestrian detection.(2)To solve the problem that the tiny YOLOV4 target detection algorithm has low accuracy and low recall rate in pedestrian detection,the feature extraction network and prediction network are improved.In the part of feature extraction network,the traditional convolution network is replaced by a depthwise separable convolution network to reduce parameters and computation.The attention mechanism module is added in the feature extraction network to enhance the area of interest of detecting object and improve the detection accuracy.A prediction scale is added in the prediction network,and the added scale is enhanced by features to improve the recall of detection of target objects.The experimental results show that compared with the original algorithm,the improved tiny YOLOv4 algorithm improves the accuracy by 7.1%,and the recall rate also increases by 6.6%.(3)To study the problem of pedestrian occlusion detection,improve the network structure of YOLOX.In the feature extraction network,the improved CBAM is used for feature extraction.Add the ASFF adaptive feature fusion mechanism module;Aiming at the situation that pedestrians are prone to occlusion between pedestrians and environmental occlusion,pedestrian occlusion data set was added to the training set,and improved Focal Loss was used to improve the uneven model to samples.The final results show that the detection performance of the improved YOLOX algorithm model is improved.