Research On SLAM Technology Of Coal Mine Robot Based On RGB-D Sensor

With the continuous progress of intelligent robot technology and the continuous improvement of coal industry productivity,coal mining is developing towards unmanned and intelligent direction.Among them,autonomous positioning and real-time mapping are important indicators for evaluating the degree of intelligence of mobile ro bots underground in coal mines.The location mapping technology based on visual sensors has gradually become the current research hotspot due to its low cost and rich information.Visual SLAM(Simultaneous Localization and Mapping,SLAM)technology is the key.Compared with monocular and binocular cameras,depth(RGB-D)cameras have the function of active distance measurement,which is more suitable for the special environment of coal mine tunnels.Therefore,this article is based on the depth camera SLAM technology to carry out research on the robot vision positioning mapping in the coal mine environment.(1)The robot vision system would be adversely affected by the complex environment underground in coal mines.A large amount of dust and water mist generated in the underground mining process cause image distortion;artificial lighting in the roadway has insufficient brightness and uneven illumination,making the image dim and the feature points are too concentrated.The poor quality of downhole images affects the accuracy of feature matching,which in turn causes the tracking deviation and loss of the visual sensor.In response to this,this paper designs the image fast defogging algorithm and histogram equalization algorithm in the visual odometry part to improve the image quality,and then uses the quad-tree ORB(Oriented FAST and Rotated BRIEF,ORB)algorithm and the improved grid motion consistency algorithm to extract and match the image features.Obtain a more uniform and accurate matching effect.Experiments show that compared with the traditional ORB algorithm,the uniformity of the feature points of the algorithm in this paper is strikingly increased,and the matching accuracy is increased by about 22.4%.(2)The accumulation of trajectory errors will inevitably occur during the movement of the robot.In response to this,this paper studies the robot trajectory loop detection based on the bag-of-words model and the global optimization of the robot pose based on the pose map,and constructs the point cloud map and the octree map of the coal mine environment.The optimized pose is evaluated through public data set experiments and coal mine environment experiments.Experiments show that the improved robot vision SLAM system performs well in large scenes and data sets with obvious light and dark changes,and the positioning accuracy in the coal mine environment is increased by about 8.9%.