Research On Key Technologies Of Safe Campus Intelligent Video Surveillance System

As the security system within the campus,the Safe Campus System plays a vital role in maintaining campus order and ensure the property and lives safety of faculty and staff.Video surveillance system,which can obtain intuitive and rich information,has always been the most important and irreplaceable component of the Safe Campus System.The intelligent analysis technology has been developing rapidly in recent years,this make it possible to analyze videos automatically and monitor specific events in video surveillance systems.However,due to the high complexity of intelligent algorithms,achieving full coverage of massive videos by intelligent service will make system costs far exceed the affordability of users.Based on the current technological conditions,we proposed an architecture of the Safe Campus Intelligent Video Surveillance System and studied the corresponding video analysis algorithms.With these works,we can satisfy the actual needs of users and keep a controllable level of costs.To solve the problem that the current video surveillance system architecture can not meet the intelligent analysis of massive videos,we proposed six principles for construction of intelligent video surveillance system,and designed a video surveillance system architecture that can separate the business and data.The proposed architecture consists of two parts,original video surveillance system and intelligent video analysis system.The intelligent video analysis system includes motion object detection layer,data storage and analysis processing layer,service function layer,and management layer.Our design can achieve full coverage of massive videos by using intelligent video analysis without any software or hardware modifications to the original surveillance system,and the cost is affordable to the user.Aim to satisfy the low complexity requirement of detection algorithms for moving object detection,we proposed a lightweight moving object detection algorithm based on differential images.We also designed two mask generation methods for this algorithm.The first one is based on the centroid of the moving background and the second one is based on the statistical characteristics of the moving background.This algorithm is the foundation of intelligent analysis technology.With this algorithm,it become very easy to apply intelligent analysis to videos and the system costs is acceptable to users.Thus,this algorithm is the crucial part of the proposed architecture.We proposed a method to decompose the pedestrian cross domain tracking task into multiple re-recognition tasks’that are retrieved on local data.By this method,we can tackle with the challenge that Re-ID task has the characteristic of high complexity and extreme computational work.At the same time,we proposed an improved re-recognition network that combines spatial-temporal information as well as.This network utilizes the spatiotemporal distribution of pedestrian motion trajectories to reduce the amount of data in Gallery,and changes the scene features learned by the servant branch in the LDS network.In this way,this network achieves better detection results.Based on the above research results,we established a demonstration system for the Safe Campus Intelligent Video Surveillance System in Campus A of a university.There are three intelligent services implemented in this system.To verified the rationality of the intelligent video surveillance system architecture and the effectiveness of two intelligent analysis algorithms,we run the system for three months.The research results of this thesis can guide the implementation of a truly intelligent video surveillance system in campus,and it can be also a reference to different application scenarios.