Nonlinear target tracking method based on optimized wavelet features

For tracking in complicated environment, a nonlinear target tracking method based on optimized wavelet features is proposed. Gabor wavelet network (GWN) is used to describe the features of the object. GWN includes a set of wavelets, and each of their parameters is computed by optimization procedure. The tracking framework is based on optimized particle filter and each particle figures a set of possible motion parameters. L-M optimization is then employed to drive the particles to the local peak values, and tracking with optimized particle filters is robust and efficient as a result of multimodality. The tracking result shows that the algorithm is robust to illumination variation and noise, and it also has the strong ability of tracking under local occlusion. Compared with standard particle filter method, the average tracking error of the proposed algorithm is within 1 pixel, which has been reduced by 50%.