Identification And Tracking Of Moving Target From Surveillance Video

Video surveillance is widely used in today's society,playing a huge role in public safety,transportation and other fields.Computer vision technology based intelligent video surveillance has become a hot research direction.Many scholars have done a lot of research on this topic,but some issues such as light changes,occlusion and so on in many complex application scenarios need to be solved in the practical application.Based on the existing classical algorithm,the detection,recognition and tracking technology of moving targets in surveillance video are studied in this thesis.The main contents are described as follows:(1)The three classic moving target detection algorithm include inter-frame difference method,background difference method and optical flow method are studied and analyzed.By comparing and analyzing,an improved algorithm combining mixed Gaussian background modeling and three-frame difference method is proposed,using the advantage of frame differential method being insensitive to light changes solves the problem that the Gaussian mixture model is vulnerable to light interference.Experiments show that the proposed algorithm can detect the target region completely under the condition of slight background jitter and illumination change,and improves the robustness of moving target detection.(2)The commonly target representation methods and classical classifiers are analyzed,and the support vector machine(SVM)classifier based on HOG feature to classify and recognize moving targets is proposed.The positive and negative samples of the training set of vehicle and pedestrian are obtained by using the public data set,and the SVM classifier of vehicle and pedestrian is trained respectively.The bootstrap mode is used to improve the performance of the model.The target area extracted by moving target detection is taken as the HOG feature extraction input to the SVM classifier for classification and recognition.(3)The Kalman filtering tracking algorithm,Mean-Shift algorithm in the generated target tracking and the Tracking-Learning-Detection(TLD)algorithm,Kernel CorrelationFiltering(KCF)algorithm in the discriminant target tracking are studied.The Mean-Shift algorithm integrated with Kalman filtering is adopted for target tracking.Aiming at the drift problem of Mean-Shift tracking algorithm in the case of occlusion,Kalman filtering is applied to Mean-Shift algorithm.The experimental results show that the proposed algorithm can achieve target tracking in the case of short-time occlusion.