Abstract:

In order to solve the noise information, which the leaf nodes votes in Hough Forest, this paper proposes an target tracking algorithm of the discriminative leaf based on Hough Forest. In the training phase of the original Hough Forest, leaf nodes do not have the ability to classify whether the image patches are target or background. The image patches contain a lot of noise, and aggregating voting information from all image patches extracted at each pixel location is really time-consuming. The experimental results demonstrate that the approach can reduce noise votes and increase tracking efficiency.

Key words: target detection, hough forest, discriminative leaf