基于多模型结合优化的非结构化道路检测算法

Unstructured road-detection algorithm based on multiple models and optimization

非结构化道路检测一直是道路检测算法中的难点.提出一种基于彩色混合高斯模型与抛物线模型相结合的优化的非结构化道路检测算法.首先采用中值滤波和二次采样法将待处理彩色图像由高分辨率变为低分辨率图像,并对图像进行光照补偿;然后建立基于优化聚类中心的K-means算法的混合高斯模型,通过最小二乘法求解左右道路抛物线模型参数;最后完成对道路边界线的拟合,实现其提取.实验结果表明,该算法对光照不均、阴影等影响的图像处理具有较强的抗干扰性,提高了运算速度,具有一定的鲁棒性和实时性.

Unstructured road-detection has been a difficulty in the road-detection algorithms. An unstructured road-detection approach based on color Gaussian mixture model and parabolic model was presented. First we took the full-resolution color image and produced a low-resolution color image by a combination of averaging filtering and sub-sampling, and had illumination compensa- tion. Then we formulated Gaussian mixture model based on K-means algorithm, which was opti- mized clustering center for the road area and the other areas. Next we solved these parameters of right and left road parabolic models by the least square method. Lastly, we fitted the boundary of the road and achieved the extraction of it. The experimental results show that this approach has robustness against uneven illumination, shadows and reliability improving the processing speed.