The Research On Person Re-identification Based On Visual Invariance With Unsupervised Learning

Person re-identification aims to determine whether a given pedestrian appears in the view of other cameras in a nonoverlapping video monitoring network,which is a more challenging research direction of computer vision than pedestrian detection.In recent years,the state has proposed using artificial intelligence to improve public security capabilities,then person re-identification has become a popular technology in the industry.In the natural scene,the person re-identification system contains two important tasks,which are pedestrian detection and person re-identification.Deep learning can simulate human learning mode to a certain extent and solve many problems encountered in machine learning,which makes great progress in artificial intelligence related technology.The recognition model based on deep learning almost achieves the optimal accuracy of person re-identification benchmark.The deep learning method has achieved great success in the field of person re-identification,mainly using supervised learning algorithm,which requires a lot of manual labeled data to achieve good learning effect.However,it is an expensive and time-consuming process to collect and labeled a certain amount of person re-identification datasets.Moreover,the route of pedestrians walking in the camera's field of view varies greatly,such as pedestrians walking at right angles,or the same pedestrian walking in differentcamera's field of view will cause a huge change of the pedestrian's apparent angle of view,resulting in the performance degradation of the recognition model.Therefore,aiming at the above two important problems in person re-identification research,this paper proposed the research on person re-identification based on visual invariance with unsupervised learning.A general unsupervised deep learning recognition model was established for the influence of angle change on the performance of person re-identification model.Under the condition of fine-tuning without label,the visual transformation and visual invariance feature were learned by temporal coherence to realize the design of visual invariance feature descriptor.According to the demand of practical application in natural scene,this paper used BING feature to realize the attention mechanism of pedestrian detection,which quickly located the approximate position of pedestrian in the original monitoring image,accelerated the process of pedestrian detection,and then weigh the speed and accuracy of pedestrian recognition system,and weighed a unsupervised person re-identification system prototype.In order to prove the reliability of unsupervised person re-identification algorithm based on visual invariance,in this paper,the results were verified on the iLIDS-VID,PRID2011 and MARS datasets,and the better performance of 57.5%(R-1)and 73.9%(R-5)were obtained on the iLIDS-VID and MARS datasets respectively.UsingBING + R-CNN as the pedestrian detector,the person re-identification system obtained the computation speed of 0.09 seconds per frame.