【摘要】 目前大多数视觉SLAM系统只能输出相机轨迹与稀疏点云地图,无法适用于机器人进行导航、避障任务。通过单目相机虽然可以构建网格地图,但没有尺度信息,RGBD相机虽然具有尺度信息,但是在室外受环境光影响比较大。该文在ORB-SLAM2算法的基础上,采用双目视觉计算相机位姿,DepthToScan通过立体视觉的深度图像得到深度信息,在此基础上采用逆传感器模型计算占据概率,并在ROS通用平台下结合占据网格地图,重新构建了一种可用于机器人路径规划、导航的2D占据网格地图算法。实验表明,该方案能清晰标识出环境中障碍物的位置,建图精度相比于RTAB-map提高了3%。更多还原

【Abstract】 Currently,most visual SLAM systems can only provide camera trajectories and sparse point cloud maps,which are not suitable for robot navigation and obstacle avoidance tasks. Although monocular cameras can build grid maps,there is no scale information,RGBD cameras have scale information,but it is greatly affected by ambient light outdoors.Based on ORB-SLAM2,a stereo binocular vision camera is used to obtain depth information through a combination of stereo vision,DepthToScan and occupancy grid map on the ROS platform. This system was developed to rebuild a 2 D occupational grid map algorithm that can be applied to robot path planning and navigation. The experiments show that the algorithm could clearly identify the location of obstacles in the environment. Compared with RTAB-map,the mapping accuracy of this algorithm is improved by 3%.更多还原