Personal-Reid SystemAcross Multiple Cameras

With the enhancement of computer's ability to process data,especially with the rapid development of deep learning,people have gained the ability to process massive amounts of data beyond any period in history.However,in the face of massive amounts of data in the real world,often due to people's needs,the type of data,and the importance of the task,etc.,although we have the ability to use these massive amounts of information,we have to invest in human and capital resources to do some measurement and trade-offs.Among the massive data,the magnitude of the surveillance image data is absolutely huge.Every day people use cameras to capture massive amounts of image information,such as schools,hospitals,banks,roads,etc.,and then the images collected by these cameras are generally stored in some databases.In the face of these huge amounts of image data,there was actually not a very efficient way to use them before.Most of them are based on human time and other clues to check with the naked eye.Therefore,this thesis hopes to improve the efficiency of the surveillance data being consulted and used based on the deep content understanding ability of deep learning.Finally,let the surveillance data in our system be actively and real-timely displayed to people with valuable information,that is,to construct a real-time cross-camera human recognition and tracking system.The main theoretical core of this thesis is the deep model of pedestrian re-identification While exploring the model to obtain higher accuracy,this thesis is also trying to optimize the work by reducing the size of the model parameters and speeding up the model's oper-ating efficiency.In addition,this thesis builds a set of supporting real-time running frame-works around the pedestrian re-identification model.This will involve data compression,data network transmission,data parallel processing,data storage and data disaster recov-ery.In fact,this part of the work is more heavy,and because there are few projects that can be used for reference,in this part,in the actual work,We have encountered more severe challenges.Finally,this thesis proposes a series of scientific systematic testing and eval-uation methods.These test methods can ensure that in the actual operation of our system,we can make full use of allocated resources,have higher fault tolerance,lower thresholds for use,and thresholds for development and expansion.Finally,this system was deployed and operated in practice.During a very long unat-tended operation time,the system worked well and the system performance met theoreti-cal expectations.These indicate that the proposed pedestrian re-identification monitoring and tracking system based on the use of massive real-time monitoring data is stable and effective.