Research On Algorithms Of Moving Object Detection And Tracking In Intelligent Video Surveillance

Intelligent video surveillance,which is often used in military,banks,airports,government agencies and other places,is widely applied in all kinds of fields.Moving object detection and tracking are the key and basic technologies of intelligent video surveillance.This thesis focuses on the research on the algorithms of moving object detection and tracking.In the process of moving object detection,traditional Gaussian Mixture Model algorithm is prone to be influenced by "ghost",shadows and background disturbance.So,there will be wrong detection results.To solove the problem,this thesis proposes an improved algorithm of background modeling based on Gaussian Mixture Model.The algorithm combines SUSAN edge detection with Gaussian Mixture Model,chooses the number of gaussian distribution adaptively,selects different learning rate in different times and uses current pixel value instead of the mean of gaussian distribution when the the pixel value changes greatly.Experiments and data analysis which compare the improved algorithm with the original one show that the improved background modeling algorithm has a higher accuracy rate on object detection.The traditional Mean Shift algorithm is prone to get the problem of deviation of tracking rectangle in illumination changes and rotary motion.Aiming at this problem,the improved MS algorithm is proposed.The algorithm introduces SIFT features which is invariant to illumination change and rotation in MS tracking algorithm.By SIFT feature extraction and matching in both object region and MS tracking region,the average offset of SIFT features is calculated.The algorithm uses the degree of similarity to measure the accuracy of results from MS algorithm and SIFT tracking method.Taking it as the weight,the algorithm utilizes weighted method to fuse the two tracking results and then obtains the final tracking result.Experiments show that the improved MS algorithm can track moving object more accurately on the condition of the illumination change and the object rotating.Due to changes in the attitude and scale of moving object,it is prone to get the problem of instability of tracking results during object tracking.Aiming at the problem,this thesis introduces affine transformation model to the improved MS algorithm and concluds the affine MS transformation matrix.This thesis also introduces affine transformation model to the SIFT feature matching and calculates the affine SIFT transformation matrix.Taking the degree of similarity as the weight,the algorithm utilizes weighted method to fuse the two affine transformation matrix and then obtains the final tracking results.Experiments show that the improved MS algorithm with affine transformation can track object more stably in the cases of changes in object's attitude and scale.