Study On Robot Target Recognition Algorithm In Coal Mine Environment

After the occurrence of coal mine disasters,the underground topography will be complicated which will lead secondary disasters at any time.If rescuers rush into the scene of coal mine disaster to rescue trappers and detect the damage of disasters,they will get injured without expectation in the secondary disaster.Then Coal Mine Rescue Robot(abbr.CMR-Robot)is needed to be carried out so that it can replace them to enter the scene of underground mine disaster and detect the damage of the disaster.This paper studies the algorithm and implementation of the object identification of Coal Mine Rescue Robot,with the Coal Mine Rescue Robot's implementing the underground environment object identification in the coal mine at the core,and with the Coal Mine Rescue Robot of CUMT developed by the robot research institution of China University of Mining and Technology as the platform.The working environment of Coal Mine Rescue Robots is characterized by low illuminance,high humidity and high particle density in the coal mine environment,which is very difficult for the robot to achieve object recognition and walk.In order to solve the problems such as object recognition of robots and obstacles avoidance,this paper mainly conducted the following researches:Aiming at the impact of the unclear and noisy in coal mine tunnel video images collected by the Coal Mine Rescue Robot on the robot's visual target recognition,video image analysis and preprocessing methods were used.Studied and analyzed the environmental information of underground coal mine which was collected by the visual collection of the robot,analyzed and preprocessed image intensification,scale transformation and image restoration,proposed the application of Partial Differential Equation(PDE)on the procession of digital information of Coal Mine Rescue Robot visual algorithm,and the algorithm of object identification of the Coal Mine Rescue Robot could conduct the information of objects' features under the coal mine environment through the post-processed visual images.Aiming at the Coal Mine Rescue Robot identifying the coal mine environment feature information,studying the corner feature information and edge contour feature information of the objects in the video image,proposing the extraction method of the feature information of the underground coal mine environment.The information of the corner and edge profile is the critical feature information of object identification.Through the study and analysis of Moravec corner detection algorithm,Harris corner detection algorithm,SUSAN corner detection algorithm,MIC corner detection algorithm and Canny edge detection algorithm,and the improvement of Harris corner detection algorithm and Canny edge detection algorithm,we could achieve strong robustness of the improved corner and edge detection algorithm.Aiming at the robot motion video image processing,the optical flow field analysis method,the inter-frame difference method and the background difference method are proposed and analyzed.For processing the videos and images of robot movements,this PhD thesis proposed the Optic Flow-Field Analysis,Inter-Frame Difference Method and Background Difference Method.Besides,while conducting the underground walking of the Coal Mine Rescue Robot,perceived underground coal mine environment by adopting the real-time videos and images information which was collected by the visual method,detected and identified the objects and obstacles on the marching movement of the robot and set up messages of the environment where the robot worked.This PhD thesis also improved the Inter-Frame Difference Method so that it would be convenient for the robot to implement the information procession.In order to verify the effectiveness and robustness of the target recognition algorithm for the Coal Mine Rescue Robot,several target recognition experiments under different environments were carried out.First,studied and analyzed the stereoscopic standardization of binocular camera and systematic standardization of monocular vision and confirmed the standardization of the left and right camera lens of binocular camera and monocular camera through algorithm.Then,achieved the object identification experiment of the robot in the environment of indoor,underground garage and coal and gas mine experiment under the support of visual identification of obstacles and surroundings,which verified the strong robustness of the object identification algorithm and achieved good results.In order to verify that the Coal Mine Rescue Robot achieves autonomous walking with the support of pure visual target recognition algorithm,achieved the object identification method through the indoor experiment,underground garage and coal and gas experiment base.Based on the pure vision recognition environment information algorithm,the target recognition obstacle recognition and obstacle avoidance in the autonomous walking of coal mine rescue robot are realized.And achieved the automotive walking of the robot based on the visual identification algorithm.Although the robot could walk automatically for only a few minutes each time,the result verified the effectiveness and timeliness of the object identification algorithm,which provides a solid theoretical research on target recognition algorithm for mine rescue robot to realize intelligent sensing environment.This paper has 111 figures,23 tables,and 166 references.