Research On Pedestrian Re-identification Based On Adversarial Generative Network And Hash Matching

At present,there are surveillance cameras in most public places.How to use these surveillance video data to better facilitate people's life has become a problem that researchers in the field of computer vision need to think about.As early as the beginning of the 21st century,China established the "Skynet System",so in the Internet era,how to make more effective and efficient use of the "Skynet System" to track and locate criminals quickly has become an urgent problem to be solved in the field of computer vision.When children get lost in crowded places such as amusement parks,how to use surveillance video big data to quickly help parents locate the children is also a problem that needs to be considered in the field of computer vision.Pedestrian re-recognition task is mainly applicable to scenes shot by multiple cameras.When searching single image,the given single pedestrian image is compared with the image data in the large image database to judge whether the pedestrian image under different cameras belongs to the same pedestrian.In the process of practical application,pedestrian re-identification task is restricted by two factors.First,the collection of pedestrian re-identification data set is limited.Due to policy,personal privacy and other reasons,scholars and general enterprises are unable to collect pedestrian data on a large scale for pedestrian re-identification task.In addition,there are deviations between existing pedestrian re-recognition data,which mainly come from the deviations between camera perspectives,the deviations between the attitudes of the captured crowd,and the deviations caused by cross-domain problems in the data set.The existing pedestrian re-recognition systems are mainly trained on public data sets,and these pedestrian re-recognition systems usually achieve good results on a single training set.However,they are often unsatisfactory when aiming at cross-domain data problems.In a word,the main problems existing in the existing pedestrian re-recognition system include:unable to achieve cross-domain effect,poor detection effect,detection speed is not fast enough,etc.Based on this,this paper proposes a pedestrian re-recognition algorithm based on antagonistic generating network and hash retrieval.Its main innovations are as follows:1.Adopted confrontation generation network based on human body posture;In this paper,the method is combined with the target attitude to generate the corresponding pedestrian attitude image;It makes up for the deviation of pedestrian attitude in pedestrian re-identification data set.2.A deep residual network based on the attention mechanism is adopted to further extract the key features of pedestrians,making the extracted pedestrian features more convenient for the re-recognition task;At the same time,the method of difference weight is used for feature fusion.3.The image retrieval method of hashing matching is adopted,and the hashing layer is added into the deep residual network to learn the hashing function while extracting pedestrian features;Hash function can transform pedestrian images into binary codes.Different pedestrian corresponding images have different binary code distances.The distance between binary codes is sorted from small to large to obtain the retrieval results.The method proposed in this paper achieves good results in terms of retrieval speed and retrieval accuracy.PyTorch is used to build a deep learning network.The final experimental results show that the pedestrian re-recognition method adopted in this paper is more suitable for solving related problems in the field of pedestrian re-recognition based on antagonistic generating network and hash matching.