Research On Understanding And Diagnosing Visual Tracking System Of Target Tracking Algorithm

Target tracking is an important research direction in the field of computer science.With the progress of information science and technology,target tracking has also been paid more attention to in the direction of artificial intelligence.More and more researchers are beginning to study this challenging technology.However,designing a tracking algorithm that can deal with various complex situations such as background clutter,occlusion,geometric deformation,fast moving is still a tough problem.In this paper,we use the target tracking as a system to understanding and diagnosing visual tracking systems and divide it into five components: motion model,feature extractor,observation model,model updater and ensemble post-processor.For the five components,the corresponding algorithm is proposed.Although these basic algorithms have good accuracy in dealing with some problems such as the appearance of objects which is similar to the target,illumination variation,and it has good real-time performance and high accuracy.Nevertheless,the understanding and diagnosing visual tracking systems has poor tracking ability in cases including background clutter.So in the paper,several improvements proposed for the shortcomings of the understanding and diagnosing visual tracking systems.Firstly,the Haar-Like feature is used in the feature extractor.In the practical application,the most basic Haar features cannot satisfy the recognition of multi-angle faces and the detection of multi-angle grayscale features,which will affect the target tracking detection of the target.For this problem,this paper improves the Haar features,adding multi-angle gray features.Secondly,in order to improve the accuracy of tracking,a simple and fast robust algorithm is introduced to improve the motion model in the system.The method utilizes the context model in visual tracking and establishes the target based on Bayesian framework and its surroundings The fast Fourier transform method is used in the detection,which improves the robustness of the algorithm and makes the tracking more accurate.It also has a good effect in dealing with the problems of occlusion,background clutter.The experimental results show that the proposed algorithm can improve the success rate and accuracy of tracking with a small average frame rate reduction.Finally,robust scale estimation is a challenging problem in visual tracking,especially indealing with a large number of scale variation and background clutter in image sequences.In order to solve this problem,this paper proposes a robust scale estimation method based on the tracking detection framework to improve the understanding and diagnosing visual tracking systems.The proposed method uses a method based on the scale pyramid which represents the learning discriminant related filter,learning the translation filter and scale filter.In addition,this article uses Raw Color to improve HOG for feature extractors.The experimental results show that the proposed algorithm has more obvious improvement in both the success rate of target tracking and the accuracy of target tracking when there are many challenge.