A lane lines post-processing algorithm: based on foraging search and circulant structure kernel tracking

At present, the method of detecting lane lines using deep convolutional neural networks is very hot. Most of the methods are two-stage, the first stage is image segmentation. The second stage is lane line post-processed. At first stage, many lane line segmentation methods can segment the lane line very well, and the effect is not bad, but if you want to get better results, the lane line post-processing is crucial. Therefore, this paper proposes a new lane post-processing algorithm, foraging search, which is inspired by bee foraging. The method does not need to acquire the internal and external parameters of the camera, and the calculation is simple. The foraging search is performed by using the distance and angle between the pixel points of the lane line, and the lane line feature points are fitted into a quadratic curve in combination with the customized region of interest. The improved csk is also used to optimize and predict the detected lane lines, which improves the robustness of the algorithm. The experimental results show that the proposed algorithm has a good recognition effect on lane lines under different road scenes.