| Aiming at the problems of poor environmental adapting ability,weak topographic perception,path planning based on particle motion and inability to realize the recognition of coal mine equipment classification and matching with known coal mine equipment information,this paper studies the technology of autonomous walking and target recognition of wheeled inspection robot based on machine vision.The particularity of underground environment in mine is analyzed,the platform of autonomous walking robot based on machine vision is built,the mobile mode of the robot is determined.According to the structure of the robot,the kinematics model and the dynamics model are established.Through the simulation experiment,the vibration velocity and acceleration change rules of the robot in the vertical direction are obtained when the robot is driving on the road of different grades and at different speeds.These data provide a basis for ensuring the rationality of mechanical structure design and the determination of safe driving condition parameters.The key problem of robot autonomous walking is studied.A terrain sensing system based on machine vision is used to locate and map the robot,namely SLAM of the robot.By combining Kinect near distance measurement with laser remote measurement,digital elevation map DEM of terrain is obtained,and then topographic information such as slope,slope direction,roughness and relief degree was obtained to achieve the classification of road hazard level.The objective function of path planning is established with path length and road hazard level as indicators.Under the condition of ensuring the safety of the patrol robot’s walking.route,Dijkstra algorithm is used to find the optimal path.The feasibility of path planning method is verified through outdoor and field experiments.Through the SINS/ZigBee combination positioning method,the tracking control of the optimal path is realized.The tracking error is reduced by using the untracked Kalman filter algorithm,and its effectiveness is verified by experiments.The identification and matching model of mine equipment is established,machine-learning algorithm is used to train and test the model.The algorithm of CNN is used to train mine equipment identification model,the identification effect of this model and YOLO3 algorithm is compared and analyzed under the three conditions of bright environment,dark environment and equipment overlap.The results show that the identification accuracy of the two algorithms is high and the difference is small,both algorithms can be applied to the identification of mine equipment.The three-axis position information of the robot relative to the mine system,the three-degree of freedom angle and the angle of the visual camera are taken as the input of the matching model,the serial number of in the camera’s field of view is taken as the output.They form the data set.Two algorithms,BP neural network and support vector machine(SVM)are used to train and test the generalization ability of mine equipment matching model.The comparison results of the two algorithms show that the coal mine equipment matching model based on the conventional SVM can use less training samples to obtain the matching model with certain generalization ability,the overall recognition efficiency and accuracy are better than the BP neural network model.The research results of this paper will provide theoretical and design reference for the research and development of coal mine inspection robot.This paper includes 77 figures,6 tables and 147 references. |
