The Improved Research Of Target Tracking Algorithm Based On TLD Framwork In Video Images

Visual tracking in video images has always been a key research direction in the field of pattern recognition. It has widely application prospect and huge economic value in military and civil field,such as missile guidance, traffic monitoring, security alarming, intelligent navigation and the other areas. The technology of target tracking in video images has been developing rapidly in the past thirty years, but the ways for long time of reliable tracking are currently far from mature. TLD(Tracking Learning Detection) is an algorithm suitable for long time tracking, however, there are still object occlusion, pose variation, illumination and real-time problems. To solve these problems of TLD, we proposed the corresponding improvement measures, mainly in the following aspects:(1)The improved algorithm of TLD and Kalman filer combined is proposed. This algorithm is mainly to solve the single objective occlusion which leads to failure.When the tracking is normal, using TLD to track the state of the next frame as the observation value to update the Kalman filter. While there is occlusion, the similar way is used but the TLD is replaced by the Kalman.(2)The improved algorithm which combines TLD and scale adaptive meanshift tracking is proposed. This algorithm is mainly to solve the single objective shape variation which leads to failure. Using the TLD tracking algorithm to dynamically adjust the scale to adapt the iterative starting point of meanshift algorithm, and then using scale adaptive meanshift algorithm provide updated template for TLD tracking algorithm。(3)An improved TLD multiple target tracking algorithm is proposed. This algorithm improves the detector and tracker in TLD model respectively, The improvement of the detector is mainly reflected in the detection efficiency of the detector and the computation of the algorithm to improve the real-time performance of the tracking. The improvement of the tracker is mainly based on the LK optical flow method which is used to solve the LK optical flow method, which can not be used to track the target and the change of the optical flow field.Through a series of experiments, the results show that the proposed algorithm improves the tracking failure of the single target occlusion or deformation, and theaccuracy and real-time performance of the tracking results basically meet the actual needs.