Research On Real Time Tracking System Of Moving Target Based On Correlation Filtering

With the rapid development of computer vision technology,target tracking as an important part of the visual field has received more attention and research.The core of the target tracking system is the target tracking algorithm,which directly determines the performance of the tracking system.Although there are many mature target tracking algorithms,there are still many shortcomings in practical application scenarios,mainly because they cannot adapt to the complex external environment such as occlusion and illumination changes.Based on correlation filtering algorithm with high speed and tracking precision,in this paper,on the basic framework,in view of the characteristics of fusion and block processing was improved,the improved algorithm is proposed,and on the basis of the improved algorithm to build the moving target real-time tracking system,and the effectiveness of the algorithm is verified by experiment and tracking system.The main research contents are as follows:This paper studies the theory and algorithm flow of traditional correlation filtering algorithm,aiming at the single feature of the algorithm which leads to the insufficient tracking effect,the multi-feature joint method is adopted to improve the adaptability of the algorithm to complex scenes.Based on the reliability experiment analysis of the tracking response graph,a new reliability evaluation index is proposed,which is used for dynamic weight fusion at the response graph level.Aiming at the problem of model drift,the classifier models are dynamically updated according to the reliability index,and a fast and effective scale processing method is proposed to solve the problem of target scale change.To solve the problem of target occlusion,occlusion judgment and redetection are added.When the target is judged as occlusion,the correlation filter model trained by the historical frame is used to carry out the redetection from coarse to fine,and the target is tracked again after it is found.A series of comparative experiments on OTB standard data sets have proved the feasibility of the improved algorithm in this paper.The overall tracking performance is better than most mainstream relevant filtering tracking algorithms.Good tracking effect can be achieved under the interference of complex environment such as target occlusion,illumination change and departure from the field of vision,and the processing speed of 41 FPS can fully meet the real-time processing requirements.The overall framework flow design of the target tracking system is completed.According to the system requirements,the software and hardware platforms were built,and the image acquisition and target detection links were completed.The coordinate information of the target pixel position obtained by the tracking algorithm was calculated as the relative Angle information between the target and the optical axis of the camera.The head of the head was controlled to follow the target according to the Angle information.In the actual scene target tracking experiment,the processing speed was34 FPS,which proved that the system could track the moving target in real time.