基于多通道融合和极线约束的道路检测与定位

计算机双目视觉道路检测及定位在实现无人运动平台自主导航中具有重要意义.根据双目摄像机系统模型，提出基于多通道阈值融合的车道线检测方法，融合车道线边缘和色彩信息进行图像阈值分割，采用透视变换和自适应动态滑窗法提取车道线像素，采用最小二乘法拟合道路模型并依据极线约束关系进行定位，投影至SLAM地图中.实验结果表明，算法在光照变化、阴影遮挡等场景中均能精确检测车道线，将车道线信息投影至三维地图可以有效地将车道信息与地图信息进行融合，提高了道路感知能力. 

Road Detection and Location Based on Multi-Channel Fusion and Polar Constraint

The use of computer binocular vision to study road detection and location plays an important role in realizing autonomous navigation of unmanned motion platforms. According to the binocular camera system model, a lane-line detection method based on multi-channel threshold fusion was proposed. The lane threshold and color information were combined to perform image threshold segmentation. The perspective transform and adaptive dynamic sliding window method were used to extract lane line pixels. The least squares method was adopted to fit road model, position according to the polar constraint relationship and project the result into the SLAM map. The experimental results show that the algorithm can accurately detect the lane line in the scenes of illumination change and shadow occlusion. Projecting the lane line information to the three-dimensional map can effectively fuse the lane information with the map information and improve the road perception ability.