Design And Implementation Of Intelligent Service System Based On Unsupervised Person Re-Identification

In modern security and monitoring applications,person reidentification is the key to intelligent applications based on massive videos.However,current systems in the field of intelligent monitoring rely on a large number of labeled data to train a person re-identification model,which increases the cost of system development and iterative training.Due to the great differences in the shooting styles of different cameras in the monitoring scene,it is difficult to extract effective information without labeling information.The style differences of different application scenes affect the model accuracy.In order to reduce the maintenance cost of the system,ensure the accuracy of person re-identification and train the robust model,it is necessary to develop an intelligent service system based on unsupervised person re-identification.The specific work of this paper is summarized as follows:(1)An unsupervised person re-identification method based on pseudo-label noise modeling and correction is proposed,which uses a camera-aware hybrid model to estimate the probability of samples being mislabeled online,dynamically divides the training data into a trusted set and an untrusted set with noisy samples based on the probability,and then uses collaborative correction to filter different types of errors and learn reliable information,thus mitigating the impact of noisy labels on the model.(2)An unsupervised person re-identification method based on curriculum contrastive learning is proposed,and the feature representation of the corresponding class clusters can be steadily reduced by incrementally mining hard samples to make the model more robust by reducing the intra-class variance.A self-paced outlier filtering strategy is also introduced to select outlier points as reliable negative samples for contrastive learning training to further improve the discrimination ability of the model.(3)Based on the above algorithms,an intelligent service system based on unsupervised person re-identification is designed and implemented.The system covers intelligent applications of person re-identification in three scenarios:schools,businesses and communities.The whole system adopts a hierarchical structure,including multiple functional modules:user management module,camera management module,person image detection module,person re-identification module,person tracking module,and intelligent service module.Finally,test each system module and function.Compared with the traditional person re-identification system,the advantage of this system is that it does not need to manually label data,but uses unsupervised person re-identification algorithm to train robust models,which greatly saves labor costs.At the same time,the system is tested,and the results show that the system can meet the user’s requirements for the reliability,real-time,ease of use,scalability and other use of the system in the real scene.