Research On Pedestrian Detection And Recognition Based On Convolutional Neural Network

Pedestrian detection is one of the most important researches in the field of computer vision.Pedestrian detection is identified and tracked by techniques such as image processing,computer vision algorithms,and machine learning.There is an increasing application for pedestrian detection technology in many fields such as civil and military applications,the driverless cars,intelligent visual monitoring and other fields.In recent years,deep learning represented by convolutional neural networks has achieved great success in the classification and detection of targets,attracting the attention of many researchers and becoming one of the hot topics in the field of computer vision.Pedestrian detection methods based on convolutional neural networks have become faster and more accurate.However,most of the methods still have the effects of network depth and excessive parameters,and run slowly on low-performance computers.YOLO real-time object detection method is currently one of the most advanced real-time target detection network.For low-hardware computers,a fast target detection method based on depth residual network is proposed based on YOLO model.The method has a 20-layer convolution neural network design and incorporates the bottleneck design of the depth residual network to effectively reduce the training parameters,and using pre-activation mechanism to ensure network performance and accuracy.Based on the above model,we propose a new pedestrian detection method.In order to reduce the rate of false detection of pedestrians,we use a Resnet-50 network model.And we increase the horizontal dimension of the feature map in view of the pedestrians.Second,our network model is pre-trained on the ImageNet dataset.And A hybrid dataset training method is used to enhance the performance of the network.Finally,a priori box is designed by dimensional clustering.In this paper,a pedestrian detection method based on depth residual network is proposed.The method uses the pedestrian recognition characteristics and the basic principle of depth residual network,and combines the methods of pre-training,joint dataset training and dimension cluster to design the pedestrian detection method.Research and experiments show that this method has better pedestrian detection effect than the traditional pedestrian detection method on the INRIA dataset.