Investigation Of Moving Object Detection And Tracking Algorithms For Visual Surveillance

Traditional video surveillance system is usually through the staff monitoring and video to achieve security protection,but there is still information missing,and the response is not timely which cannot accurately and efficiently monitor all scenes.In comparison,intelligent analysis and monitoring video can effectively solve the aforementioned problems,through the real-time monitoring screen introduction algorithm analysis,the system can detect and track the target in real-time screen so as to make sudden notification of staff or alarm which could reduce manpower costs.However,the main reason why it is not universally applied is that the current target detection and tracking algorithm has a high error rate and poor robustness.In recent years,with the rapid development of information technology,target detection and tracking technology has become an important subject in the field of computer vision.Based on the framework of video surveillance system,this thesis focuses on the real-time detection and tracking of the target,and focuses on the TLD algorithm.This algorithm has been solved the long-time tracking,scale and illumination changes and partial occlusions problems which has a strong practicality.However its main drawback is that the algorithm is computationally intensive,the program runs very slowly,cannot reach the real-time effect of the target tracking in video surveillance,so it cannot be applied in real life.In view of the above problems,the main work of this thesis includes the following aspects:(1)Had an in-depth research in the several classical algorithms of target detection and tracking.Analyzed their operating principle and made experimental test,made a result of the feasibility of these algorithms applying to the actual video monitoring system according to the experimental results.(2)Made an improvement in the tracking of the original TLD algorithm.Adjusted image resolution dynamically according to the area ratio between object and whole picture frame,through the relevant algorithm to reduce the input of samples when the target area in the entire area occupied by the larger map which can reduce the number of samples overall.As a result,the amount of calculation can be reduced and the scanning speed also can be improved.(3)In the detector part of the algorithm,made a new method which made the detection area generated in the neighborhood,and the input image was calculated by using the optical flow method in the tracker.And then the current approximate position of the target was estimated,so the scanning frame was generated by sliding in the neighborhood which can solve the time-consuming problem of the whole picture.(4)Proposed a new template matching algorithm.In the original algorithm detector,the authors use the traditional normalized product correlation template matching method,and the calculation speed is slow,which is the main reason for the impact of the algorithm.The improved algorithm adopted the new template matching method,which made the program running speed greatly improved.(5)Based on the existing hardware facilities in the laboratory,the real-time video monitoring system platform was built,and the new target detection and tracking algorithm was embedded in the video monitoring system to test the detection and tracking performance of the target under real-time monitoring screen.