Lane recognition and tracking based on muli-features fuzzy fusion and particle filter

In order to overcome the instability of the lane object recognition based on only one feature, the shadiness object is extracted from the lane image, multi-features information including texture and edge are adopted, Eigen vectors composing of texture and edge are transformed with fuzzy theory to subjection degree and fuzzy density. Finally, they are fused with fuzzy rules in the frame of particle filter. In the course of lane image processing, successively the origin image is gray-scale transformed, the gray image is segmented with OTSU method, the segmented edge feature is fitted to straight line with improved Hough method, the texture feature is extracted by means of gray scale co-occurrence matrix, in the end, the extracted edge and texture features are fused fuzzily with the improved particle filter. The experimental results shows that the method based on multi-features fuzzy fusion and improved particle filter can provide strong robustness of the recognition and tracking for structural lane, and also acquire better tracking effect for the unstructured lane.