针对传统非结构化道路检测算法对各种复杂路面环境通用性不强且计算复杂度高的问题,提出了一种基于交叉视觉皮质模型(ICM)的道路检测算法。ICM具有接近生物视觉信息处理机制的特点,能够根据像素及其邻域的相关性动态区分目标和背景。基于ICM分割算法需要解决的问题是最佳分割阈值和循环迭代次数的确定,提出了采用最小交叉熵判决机制确定最佳分割阈值与循环迭代次数,从而避免了人为干预,提高了分割速度。实验结果表明,该算法不仅能够实现道路图像的精确分割,而且对一些非常规路况的适应性较强。

A lane detection algorithm is proposed using intersecting cortical model (ICM) in view of the weak universality and high complexity of the traditional methods in detecting the unstructured lane whose circumstance is complex and diverse.ICM has the superiority which is much closer to the information processing mechanism of biological vision.The ICM can distinguish objects and background dynamically according to the relevance between pixel and its neighbor pixels.As the best threshold and cyclic iterative times of ICM is artificially given it can not realize segmentation automatically.Therefore the decision mechanism of the minimum cross entropy is introduced to determine the cyclic iterative times and the best threshold automatically.The result of experiments show that the precision of the algorithm is high,and it also has very strong adaptability to some unconventional lanes.