| With the continuous progress of artificial intelligence technology and the continuous improvement of coal industry productivity,coal mining is developing towards the direction of unmanned and intelligent.In the process of using intelligent equipment in coal mine production,it is easy to be constrained by the unknown environment of coal mine roadway,so it is necessary to reconstruct the three-dimensional environment of the inner scene of coal mine roadway through the coal mine detection robot,so as to provide the basic guarantee for the normal operation of intelligent equipment in coal mine.The accuracy of the threedimensional reconstruction of the coal mine roadway completely depends on the positioning accuracy of the coal mine detection and rescue robot.However,due to the characteristics of low illumination,high dust concentration,high humidity and the interference of dynamic objects inside the roadway,the performance of the positioning system of the coal mine detection robot will be seriously affected.For this reason,this paper proposes an image processing algorithm based on combined cost function and a dynamic object removal algorithm combining optical flow method and semantic segmentation for the precise positioning of coal mine detection robot,which can solve the influence of interference factors in coal mine roadway and improve the positioning accuracy and robustness of coal mine detection robot.The specific contents are as follows:In order to meet the needs of precise positioning of coal mine detection robot,firstly,the terrain characteristics of coal mine roadway are studied.Secondly,the research and development of the coal mine detection robot platform which is suitable for the terrain characteristics of the coal mine roadway,and the design of its hardware and software system.In terms of hardware,according to the needs of the actual scene,Xiaomi binocular depth camera is selected to obtain the internal scene information of the coal mine roadway,and inertial navigation equipment is used to assist camera estimation to provide accurate position and attitude estimation for the coal mine detection robot;according to the characteristics of the internal terrain of the coal mine roadway,W-type wheel walking mechanism is used as the chassis walking mechanism of the coal mine detection robot,which not only ensures that Walking speed is fast,and it has the ability to climb over obstacles.In the aspect of software,the wireless communication network is used to communicate between the upper and lower computers,and the ROS robot operating system is used to control the coal mine detection robot.Then,the rviz and gazebo display interfaces are built to meet the real-time display requirements of the normal work of the coal mine detection robot.In view of the influence of sensor parameters on the positioning system of coal mine detection robot,it is necessary to calibrate the sensor.Firstly,the camera imaging model is analyzed.Secondly,the camera is calibrated by kalibr tool and Matlab tool.The experimental analysis shows that the image coordinate errors of the left and right cameras calibrated by kalibr tool are all distributed between ± 1 pixel,the average coordinate errors of the cameras calibrated by Matlab tool are distributed around 0.14 pixel,and the calibration data of Matlab tool is consistent with the factory data of Xiaomi camera.Then,the internal parameters of IMU are calibrated.Through analyzing the noise model of IMU,the data package of IMU is collected.After calibration test,the parameters such as IMU bias and random walk error can be obtained.At last,the camera and IMU are calibrated by off-line joint calibration,and the external parameters of the camera and IMU and the time drift data between the sensors can be obtained by experimental analysis.Aiming at the problems of low illumination,high dust concentration and high humidity in the inner environment of coal mine roadway,an image processing algorithm based on combined cost function is proposed.A coal mine simulation tunnel test platform is developed.By changing the interference factors such as light,dust concentration and humidity in the simulation tunnel,the vision system of the coal mine detection robot is studied.In this paper,the process of image defogging algorithm mainly includes image atmospheric light estimation,image block transmission estimation,combined cost function to avoid image information loss and fine transmission caused by excessive contrast enhancement.Compared with many kinds of image defogging algorithms,the results show that the proposed algorithm can not only ensure the effect of image defogging,but also improve the brightness,contrast and texture details of the image.Aiming at the interference problem of dynamic objects in the scene of coal mine roadway,a dynamic object removal algorithm based on the combination of optical flow method and semantic segmentation method is proposed to eliminate the interference of dynamic objects and improve the accuracy and robustness of the positioning and mapping system of coal mine detection robot.This paper mainly aims at the further improvement of orb-slam2 algorithm framework.By analyzing the advantages and disadvantages of frame difference method and optical flow method,we use the combination of optical flow method and semantic segmentation method to detect and remove the dynamic features of the image,eliminate the interference of dynamic objects in the scene,ensure the dynamic Association of data,meet the positioning accuracy requirements of the coal mine detection robot in the dynamic scene,and improve the accuracy The robustness of the system.Through the experimental analysis of dynamic scene semantic segmentation,the proposed algorithm can effectively segment the characters and computers in the scene.In order to verify the applicability of the dynamic object removal algorithm based on the combination of optical flow method and semantic segmentation proposed in this paper,through a large number of experiments and analysis,the proposed dynamic scene SLAM algorithm can operate normally in the high dynamic environment,and the pose estimation accuracy of the coal mine detection robot is greatly improved,which realizes the highprecision positioning in the dynamic scene,and fully demonstrates the rationality of the algorithm And effectiveness.In view of the image saving form,octomap map based on octree is adopted to improve the map updating efficiency and reduce the memory consumption. |
PDF Full Text Request