Object Tracking Method For Space Moving Image Sequences

Moving target tracking has wide application prospect in the field of space exploration, military field, unmanned, robot. It has been an important research topic in computer vision. Marking the moving objects in image sequences or moving object sequences of interest, and using the target tracking algorithm for moving object for long time, uninterrupted accurate tracking are an important research direction for target tracking. In this thesis, we studied target tracking algorithms for the space moving image sequences and developed target tracking tool. Our main contributions are shown as follows.(1) Firstly, we proposed a new enhancement method for low contrast image sequence, which is used to handle brightness image sequence and adjust its brightness, and combined with the inter-frame motion information to luminance brightness compensation. Compared with traditional HE and LIP algorithms, our method gained better effect in subject and object indicators, fulfilled brightness adjustment for moving image sequence.(2) We proposed an advanced TLD (Training Learning Detecting) based on local priority for handled image sequence, which firstly handled video sequence by algorithm proposed, then used TLD algorithm to track moving object. In order to decrease error-tracking-rate, we proposed local priority TLD-based algorithm. Combined with original TLD method, our method had a well promotion in error-tracking-rate, achieved the object tracking system for changes in brightness and shape and made it real-time at the same time.(3) We proposed an advanced Meanshift algorithm based on optimized background to reduce the influence of background on target tracking effect, thus decreasing error-tracking-rate compared with traditional Meanshift method. For object changes in shape, we proposed an advanced method based on self-adjust window, which had a good adaption for shape changes and gained a good tracking result.(4) We developed an object tracking tool based on above research results. It includes three big modules, which are image preprocessing module, object detecting module and object tracking module. We can conveniently set parameters and execute a specified algorithm. It can reduce humanized complicated work and it is a convenient integrated management tool.