Design And Implementation Of Pedestrian Re-identification System Based On Deep Learning

With the development of artificial intelligence,big data and other fields,deep learning plays a vital role in computer vision,image processing,and natural language processing.The face recognition application scene is extremely rich,such as mobile phone unlocking,work punching,ticket checking,monitoring and so on.But sometimes the camera can't capture a clear face,in which case a new recognition algorithm and system research is imminent.Pedestrian re-identification,as an emerging field of computer vision,has many similarities with face identification,but the difference is that it can solve the problem that the face can't be captured,so it is a society that is crucial to its research.The pedestrian re-identification intelligent system can be divided into two modules: pedestrian detection and pedestrian identification.The pedestrian detection uses the pedestrian classifier to detect whether there is a pedestrian in each frame of the input video stream,and if it exists,mark it.Pedestrian re-identification uses the special algorithm to extract the features of the detected pedestrian images.The extracted features are compared with the features in the pedestrian feature database for similarity calculation.If the similarity exceeds a certain threshold,the pedestrian ID corresponding to the pedestrian image is output.The main work of this paper is:(1)Using the deep learning of the current hottest PyTorch framework to train and implement algorithmic models.The pedestrian detection part adopts the current more efficient YOLOv3 model,and the pedestrian re-identification part adopts the PCB model with higher current recognition rate.And the combination of the two realizes a set of pedestrian re-identification intelligent system including front and back.(2)In order to increase the robustness of the model and reduce the impact of the environment on the recognition results,this paper integrates the current three mainstream large-scale pedestrian re-identification datasets,and uses the CycleGAN to generate the network to migrate the image styles captured by different cameras.Finally,a fusion dataset was produced.The work of this paper greatly increases the amount of training data of the PCB model and the accuracy of the model after training.(3)In order to simplify the management of monitoring equipment and accelerate the transmission rate of surveillance video,this paper invents a Raspberry Pibased pedestrian recognition processing terminal,which can compress and cache images before video transmission,and simultaneously through internal implantation.The program is very good at controlling the monitoring equipment.This terminal successfully applied for a utility model patent.