Multi-resolution based motion estimation for object tracking using genetic algorithm
Harish Bhaskar, R.L. Kingsland, Sameer Kumar Singh · 2006
Object tracking based on motion estimation involves tracking the true motion of objects/features in a motion sequence. A novel mechanism of multi-resolution based motion estimation for object tracking using genetic algorithm is proposed. The proposed method combines variable size block matching using quad-tree decomposition with genetic algorithm based motion estimation. The method iterates between the process of decomposing an image using quad-tree approach and motion estimating the fitness of every region of the quad-tree using genetic algorithms. The key contributions of the work include: a) motion estimation using genetic algorithms, b) accurate tracking of single and multiple objects c) distinction between camera and object motion. The model is validated on several real time datasets and is compared against simple block matching schemes. The role of genetic algorithm based template matching is justified through comparison with 7 other baseline search techniques. Results demonstrate that the proposed method outperforms traditional algorithms.