Research On Adaptive Tracking Of Moving Objects In Image Sequences

Video surveillance system has been infiltrated into the people's production and life. Facing with massive emergence of video data, many industries' demand for intelligent video processing is increasing. As one of the core of the video processing technology, object tracking has become a research hotspot in the field of computer vision and image processing. Moving target tracking is to timely find interested moving targets in an image sequence, which has been widely applied to various fields, such as intelligent surveillance, visual navigation, intelligent transportation, human-computer interaction, video retrieval, etc. Due to the influence of some distractions such as the light, the target appearance change and possible occlusion, the current many target tracking algorithms still cannot meet the requirements of high precision and good real-time for intelligent video processing.The viewpoint of this thesis is the challenges faced by object tracking system. To design an object tracking algorithm is more effective and robust than those state-of-the-art approaches to appearance variation and heavy occlusion, the method named adaptive randomized ensemble tracking using appearance variation and occlusion estimation and a tracking method based on the feature points location in the framework of TLD are proposed. The main research content includes the following two aspects.1. Adaptive randomized ensemble tracking using appearance variation and occlusion estimation. To solve the problems which RET puts emphasis on estimating the state of the classifier rather than the state of the object and the fixed learning rate parameter was used in the process of model update. The thesis puts forward a method to estimate the state of appearance changes and possible occlusion at each frame. Then we combine the appearance variation and occlusion estimation to adaptively update the classifier and model. A method based on sparse optical flow to estimate the appearance variation of the object and a sparsity-based occlusion estimation method between consecutive frames are proposed. Experiments and evaluations on some challenging video sequences have been done and the results demonstrate that the proposed method is more robust against appearance variation and occlusion than those state-of-the-art approaches.2. A tracking method based on the feature points location in the framework of TLD. TLD cannot handle occlusion well, and the tracker' performance reduces for pedestrians tracking. To solve these problems, a tracking method based on the feature points location in the framework of TLD was proposed. The SURF algorithm was used to extract some feature points at each frame, and then we choose some feature points which are related to the tracking object to assist in locating the location of the object. In the process of tracking, the TLD algorithm was employed to track the object. Once the TLD algorithm fails to track the object, the feature points model was used to find out feature points in current frame and estimate the location of the object by the feature points. The experimental results show that the proposed algorithm is more effective and robust than TLD, VTD, SCM, etc.