The Research On Pedestrian Detection Method Based On Image Enhancement And Deep Network

With the rapid development of the rail transit network,scientific and effective management of urban public transportation is of great significance for maintaining public order and planning traffic operations.Real-time and accurate passenger flow monitoring has become an important basis for decision-making planning.Among them,pedestrian detection based on deep network models is the key technology of passenger flow monitoring systems.Deep learning methods learn features from a large number of samples,and have a high dependence on the number and quality of the samples.The samples collected from actual detection problems often have problems such as insufficient numbers and poor image quality.At the same time,deep convolutional neural networks usually adopt the method of increasing the number of network layers to improve the learning ability.This method leads to a large number of model parameters and high time complexity,which affects real-time detection performance.This paper studies the pedestrian detection technology in subway passenger flow images..First,this paper proposes to use the scene migration method to augment the data of subway pedestrian detection samples.This method extracts the characteristics of content images and style images based on the Fast Photo Style style migration network that has a good migration effect on real photos.Based on this method,we can obtain new samples with style picture styles and content picture contents,then we smooth the stylized results to reduce distortion.This method can achieve realistic new samples based on new scene migration when there are few samples.At the same time it can keep the shape and position of the detection target without need to re-label the detection labels and achieve data augmentation of the detection samples.Secondly,this paper proposes an image enhancement algorithm that combines the dark channel defogging algorithm with the adaptive color equalization algorithm.The use of dark channel dehazing algorithm can effectively improve the problems of grayness,uneven illumination and weak contrast in subway pedestrian samples.At the same time,an adaptive color equalization algorithm is added to further enhance the image and improve the sample quality.Finally,the deep neural network can learn features from images of various qualities and improve the model generalization ability.Thirdly,this paper designs a lightweight network structure based on the YOLOv3 network framework.This method performs deconvolution and optimization of the convolution mode on the original network.Aiming at the problems of pedestrian appearance change and complex background,the darknet53 network with better performance is selected as the feature extraction network.At the same time,according to the metro pedestrian detection speed requirements in this article,a large number of repeated convolution modules in the deep network are deduplicated and the depthwise convolution method with fewer parameters is used to replace the original convolution method.Based on the above method,this paper reduces the time complexity of YOLOv3 network and improves the detection speed.Finally,this paper trains a pedestrian detection model for subway passenger flow based on the research of the above several key technologies and achieves a good balance between the accuracy and speed of the detection model.The validity of the method in this paper is verified.