Research On Object Detection And Tracking Algorithm Based On Video Surveillance

Moving target detection and tracking is an important part of the intelligent video surveillance system and is widely used in many fields.However,the actual detection and tracking environment is often accompanied by complex situations.The algorithm needs to overcome the challenges of complex scenes to achieve stable and efficient detection and tracking effects,which puts higher requirements on the performance of the algorithm.In the process of moving target detection,this thesis mainly studies the visual background extraction algorithm based on background modeling.In the process of target tracking,this thesis mainly studies the relevant filtering target tracking algorithm.The specific research contents are as follows:(1)The visual background extraction algorithm uses the first frame of the video to initialize the background model is simple and efficient,but this method will introduce ghost areas in a certain range of frames when the moving target appears in the first frame of the video;The Vi Be algorithm is more sensitive to dynamic background disturbances and will detect it as a false foreground spot;Modeling methods do not adapt well to changes in lighting.Based on these three improvements,experiments show that the improved algorithm can effectively meet the challenges of complex scenarios.(2)Based on the kernel correlation filtering algorithm,a multi-strategy fusion correlation tracking target tracking algorithm is proposed.On the decision-making level,adaptively fuse color-named features according to the tracking confidence to achieve the complementary advantages of the features to obtain better tracking results;train a onedimensional scale filter separately to cope with the scale changes of the target;a new adaptive filter update strategy is proposed.Experiments show that the improved algorithm has a greater improvement than the kernel correlation filtering algorithm in the face of motion blur,background confusion,light changes,and slight occlusion.(3)Based on the kernel correlation filtering algorithm,an anti-occlusion long-term target tracking algorithm is proposed.Related filtering algorithms are prone to lose targets when facing occlusion,especially in severe occlusion situations.Its high-risk update strategy makes it difficult to retrieve the targets.Train a one-dimensional scale filter to respond to the target scale change;determine the tracking confidence by combining the maximum response value and the average energy of the relevant peaks,and use the support vector machine to re-detect the target when the tracking confidence is lower than the threshold.Cooperating with each other improves the robustness of the algorithm in the face of occlusion situations.Experiments show that the improved algorithm has a significant improvement over the kernel correlation filtering algorithm in long-term tracking tasks and has occlusion situations.