Competitive Learning for Genaralized Motion Tracking

Generalized motion tracking algorithm using competitive learning networks is proposed. Two networks are applied in this algorithm with the input for the network is pixel value of the image. The output from this network is then simulated with second network, whose input is split into blocks with size of N x N. Two stages are involved in this project, namely preparation stage and tracking stage. The preparation stage develops both networks and uses them in tracking stage. Histogram threshold is applied to filter the group numbers of the simulated output. The histogram threshold is improved to enhance the performance of the algorithm. The groups of tracking target are initialized based on the position of the target in first frame. Post-processing, such as image filling is involved in the algorithm. The performance of proposed algorithm shows system robustness on orientation change, size and movement. Hence, feasibility of motion tracking algorithm with competitive learning network is verified as the proposed algorithm is able to locate tracking target in any positions.