Research On Real-time Location And Map Construction In Tunnel-type Confined Space

Tunnel confined space refers to the mining roadways,auxiliary transportation roadway of coal mines,and underground tunnels for various purposes.Hazardous gases will be generated during the mining process.In addition,spatial changes in underground space will also cause roof collapse and water inrush problems.Hence,roadway safety inspections and emergency rescue in the event of sudden disasters are very important for safe production of coal mines.However,currently used methods rely heavily on manual and monitoring methods,which are dangerous.Therefore,this thesis designs and develops a drone flight platform.Based on the ORB-SLAM2 framework,and fusion inertial measurement unit(Inertial Measurement Unit,IMU),the research studies the positioning,map construction and navigation algorithms of drones in tunnel scenes,and employs robotics technology in inspection and emergency rescue tasks.The main work is as follows.(1)Research on key technologies for tunnel-type confined spacesThere are very few researches on real-time positioning and map construction algorithms in confined spaces.In view of it,this part mainly studies the visual frontend feature point extraction,homogenization and mismatch elimination algorithms for the tunnel scene.Firstly,time-series images of the tunnel are collected,A variety of classic feature point detection algorithms are employed and evaluation indicators are given to compare the performance of these algorithms.Experimental results show that the ORB algorithm provides the best performance.Secondly,given feature point redundancy,the quadtree method is adopted to homogenize the feature points.Finally,due to the limitations of RANSAC algorithm in terms of accuracy and robustness,correlation calculation is performed on the matching points of the image and the ones that meet the threshold condition are regarded as the candidate points of the interior points.Then the graph cut optimization algorithm is used in the judgment of the inner and outer points in the third step of the RANSAC algorithm.Experimental results show that the performance of the improved RANSAC algorithm has significantly in terms of accuracy and robustness.(2)System hardware platform designCurrently,there are many types of commercial drones.However,there are very few inspection drones dedicated to tunnel scenes,mainly because they must not only meet the safety standards for coal mines,but also achieve accurate positioning and navigation without external signals such as GPS.Therefore,this thesis independently designs the UAV flight platform from the aspects of hardware selection,main technical indicators,frame design and safety protection design.The designed platform offers certain hardware redundancy capabilities,and can be equipped with gas detection modules and computing units,it can meet the experimental requirements after many test flights.(3)Research on the navigation method of UAV in tunnel confined spaceAiming at the UAV navigation problem,the strategy adopted in this thesis is to use the LSD algorithm to detect the intersection between the roadway ground and the walls on both sides.In order to ensure accuracy,the image is divided into regions of interest,and the length threshold is used to eliminate noise line segments and retain all detected straight line segments at the intersection of both sides.Then the intersection point(vanishing point)model is constructed.After the vanishing point being calculated,the yaw angle of the UAV is obtained through the yaw angle solution model,and then the navigation control of the UAV is realized.Experiments verify the effectiveness of the navigation method.(4)Implementation of VI-SLAM system integrating monocular vision and IMUIn order to solve the problem of the unobservable scale of the monocular pure visual SLAM algorithm,the parameter information needed by the VI-SLAM system is obtained by loosely coupling and aligning with the inertial device.Then the back-end optimization is performed through the tightly coupled framework,The tests based on the public data set and the local real scene shows a significant improvement in accuracy and robustness compared to that of the ORB-SLAM2 algorithm.In this thesis,a drone flight platform is designed.Moreover,the location,map construction,scene recovery and navigation algorithms of the drone in the restricted space of the tunnel are studied based on the improved SLAM system.The experiments indicate that the positioning accuracy of SLAM algorithm,which combines vision and inertia,has improved.The thesis includes 68 figures,11 tables and 82 references.