煤矿巷道机器人管线视觉辅助定位与导航方法研究

Study on pipeline vision-aided positioning and navigation method for coal mine tunnel robot

在煤矿巷道安全巡检以及灾变环境应急救援等工作中通常需要使用到智能机器人。目前智能机器人的自主导航与避障技术已经比较成熟,但是主要应用场景都是有卫星定位(GPS、北斗导航系统等)的室外环境或者特定室内环境。由于煤矿巷道存在环境复杂,光照不均匀,空间狭窄等难点,智能机器人在煤矿巷道的自主导航方法还需要进一步研究。常用的传统惯性导航是使用惯性测量单元(Inertial Measurement Unit,IMU)的加速度与角速度来推算物体在三维空间的位置信息,但是误差累积问题比较严重。煤矿巷道存在大量管线,结构化特征显著,通过机载摄像头获取井下巷道图像,使用机器视觉算法来定位巷道图像中的管线,并且通过解算机器人与管线之间的偏航角来辅助机器人的视觉导航。针对巷道中管线的颜色鲜明、几何形状特征明显的特点,结合颜色与几何特征,采用将原始图像纵向分割成多个独立子图像的方法,减小环境噪声对图像中管线分割带来的影响,然后从每个子图像中获取候选管线轮廓,并判断是否属于同一根管线进行轮廓分组,从每组候选轮廓中根据管线轮廓所拟合直线的平行程度进一步筛选出较为鲁棒和稳定的管线轮廓。结合相机针孔模型和偏航角解算模型,进而获得机器人当前的偏航角度。试验表明:上述方法不仅快速,而且计算的偏航角准确可靠,能够满足煤矿巷道机器人视觉辅助定位与导航需求。

Intelligent robots are usually used in tunnel safety inspection and emergency rescue of coal mine disaster environments.At present,the technology of autonomous navigation and obstacle avoidance of intelligent robot is relatively mature,but the main application scenarios are outdoor environments with satellite positioning(GPS,Beidou navigation system etc.)or special indoor environments.For the complex environment of coal mine tunnel with uneven illumination and narrow underground space,the autonomous navigation method of intelligent robot in coal mine tunnel needs further research.The commonly used traditional inertial navigation applies the acceleration and angular velocity of an inertial measurement unit(IMU)to estimate the position information of the object in 3-D space,but its error accumulation problem is more serious.There are a large number of pipelines arranged in coal mine tunnel,and their structured features are very remarkable.The image of coal mine tunnel is obtained through the onboard camera,then the machine vision algorithm is constructed to locate the pipeline in the tunnel images,consequently,the robot’s vision navigation is assisted by solving the yaw angle between the robot and the pipeline above mentioned.In view of the bright colors and the obvious geometric shape characteristics of the pipeline in coal mine tunnel,this paper combines the color and geometric characteristics and divides the image into multiple independent sub-images longitudinally to reduce the impact of environmental noise on the pipeline segmentation in the image,and then obtain candidate pipeline contours from each sub-image,and the contours are grouped according to whether they belong to the same pipeline.From each group of candidate contours,a more robust and stable pipeline contour is further selected based on the parallelism of the straight line fitted by the pipeline contours.Combined with the camera pinhole model and the yaw angle calculation model in this paper,the current yaw angle of the robot is obtained.Experiments show that the method proposed in this paper was not only fast,but also the calculated yaw angle in an accurate and reliable way,which can meet the needs of vision-aided positioning and navigation of coal mine tunnel robot.