自主移动机器人自适应室外道路检测

目的 为降低室外自主移动机器人视觉导航中遇到的阴影、裂纹及道路边界不规则造成的道路检测算法不鲁棒性,提出一种每帧灰度阈值可调的快速自适应道路检测方法。方法 先采用2维离散小波进行道路图像分解与重构,比较各级小波重构后的近似道路图像,确定出不影响“路-非路”灰度二分类的最佳分辨率等级;在低分辨率尺度空间中,用灰度类间最大方差和类内最小方差共同构造适应度函数,采用改进的遗传算法对各帧道路图像进行阈值自适应分割,找到准确的道路边界,最近两边界中心位置即机器人行驶方向。采用小型陆地自主车作为研究平台,并在卡耐基梅隆大学（CMU）提供的室外移动机器人道路视频中进行算法测试。结果 本文方法能够在具有阴影、裂纹、光照度变化的道路条件下鲁棒分割出道路边界,机器人可以平均30 km/h的速度在有较严重阴影干扰的校园道路上行驶,视觉系统的处理速度平均可达到20 ms/帧。结论 本文方法比传统的灰度直方图分割法表现出更强的环境自适应性,可实现较为鲁棒的室外道路检测,并可作为室外自主移动机器人非结构化道路检测的一种鲁棒性较强的方法加以推广。

Adaptive outdoor road detection for autonomous mobile robot

Objective In process of outdoor autonomous mobile robot visual navigation, shadows, causing cracks and irregular road boundary are encountered, which make the detection algorithms not so robust. The objectives of this paper are to resolve these problems. Method The proposed method in this paper is called a fast adaptive road detection method evolving adjustable gray thresholds per frame. First, a two-dimensional discrete wavelet analysis is used for road image decomposition and reconstruction. After comparing the approximate wavelet reconstruction of the road image in multiple-levels, a best resolution grade is determined which does not affect the "road-non-road" classification. In the best scale space, the grayscale maximum variance between-class and minimum variance within-class are used to create a fitness function, and the improved genetic algorithm is used in road image segmentation with each frame having an adaptive grayscale segmentation threshold. After that, the accurate road boundary is found, and if using the nearest two boundaries to calculate the central position of the road, the robot can know its driving directions. Content of main experiments: In this paper, a small autonomous land vehicle is used as a research platform, and the algorithm is tested by the outdoor path driving video of a mobile robot provided by CMU. Result The experimental results show that this method can detect the boundaries robustly under varying road conditions including shadows, cracks, and illumination changes. The real-time performance of the road detection system is good. The robot with this algorithm can run at a speed of 30 km/h on the school road covered with shadows, and the process rate of the vision system can reach to 20 ms per frame. Comparisons with reviewed researches: This segmentation method showed stronger self-adaptability to the environment than the traditional gray level histogram based segmentation method. Conclusion A robust detection for the outdoor road is realized by the method of this paper. The proposed method in this paper can be seen as a robust method to the outdoor autonomous mobile robot's unstructured road detection, and should been extended.