Abstract:

Aiming at the problem of vehicle tracking in real scene, we proposed an improved particle filter for vehicle tracking algorithm. By using the framework of immune resampling, it reduced particle degeneracy and ensured the effectiveness of particle filter. Meanwhile, a memory base was established which refers to the idea of artificial immune algorithm, so that the algorithm could track targets for a long time. The background weights histogram and subblock identification mechanisms were used to reduce occlusions which caused by offtracking. Moreover, the adaptive learning parameters were added to the movement model and antibody mutation, which could improve the robustness of the algorithm. The experimental results show the algorithm has ability of stable tracking under the different conditions of illumination change, sudden movement or target occlusion, which verifies the validity of the proposed algorithm.

Key words: particle filter, artificial immune algorithm, adaptive learning, vehicle tracking