Online Object Recognition using MSER Tracking

This thesis presents a robust online learning and recognition system. The basic idea is to exploit information from tracking an object during the recognition and/or learning stage to obtain increased robustness and better recognition results. Object tracking by means of an extended MSER tracker is utilized to extract local features and construct their trajectories. Compact object representations are formed by summarizing the trajectories to corresponding frontal MSERs. All steps are performed online including the MSER extraction, tracking, summarization, SIFT description as well as learning and recognition based on a vocabulary tree. The online learning by tracking approach is evaluated on realistic video sequences which prove the increased performance for robust online recognition. The whole system runs at a frame rate of 9 fps on a standard PC.