Study On Methods Of Moving Object Detection And Tracking In Dynamic Scene

Moving object detection is an important research topic in the development ofdigital video technology, and has wide range of applications in the fields ofsafety-critical sectors such as video surveillance, and has broad applicationsprospects in many other areas of human society. With the development of thealgorithm of the feature points, using the method of the local feature point match toachieve the target tracking has been widely used. In this paper, the feature pointmatching method was used to achieve the target detection and tracking in dynamicscenes.For the target detection, In order to describe the movement of the camera, weintroduce the rotational parameter model and compute the motion parameters whichthen leads to the implementation of motion compensation and the object detection.In this process we using RANSAC algorithm to eliminate the outliers which isgenerated in the process of feature points matching, after that the least squaremethod can precisely compute the motion parameters which then leads to theimplementation of motion compensation and the object detection. The algorithmhas been tested on several real shot sequences and standard sequences, and theexperiments demonstrate that the method can effectively detect objects whileachieving real-time performance. Our original work can be summarized as:(1) As to the shortcoming of SURF-low matching rate which can’t meet therequirements of real-time target detection, we propose a feature searching andmatching strategy based on position estimation. This can improve the efficiencywhile maintaining performance of original SURF.(2) Update strategy based on the feature points of the feature points of theresidual block SAD values, to ensure that there are enough matching points to solvethe global motion parameters. This update strategy not only makes sure thatfeatures can be updated frame by frame but also increases the running speed.For the target tracking, the feature point matching method was used to trackthe target. The algorithm has been tested on real shot sequences and standard sequences, and the experiments demonstrate that the method can effectively detectobjects while achieving real-time performance. Our original work can besummarized as:As to the shortcoming of SIFT algorithm with the tilt angle change of camera,we propose an improved SIFT algorithm to improve the target trackingperformance.