Underground Roadway Location And Mapping Based On Multi-Sensor Fusion

The characteristics of China’s energy endowment determine that coal will still be the main energy structure in the future,and the underground mining of coal mines,as a huge system,includes many links such as coal mining,tunneling,transportation,etc.Among them,roadway is an important channel to ensure safe production.Therefore,constructing a high-precision map of the roadway under coal mines plays an important role in guiding the efficient and safe production of coal mines.In order to adapt to the harsh environment of single feature texture and poor lighting in underground tunnels,this thesis studies the method of underground tunnel location and mapping based on multi-sensor fusion.The main work is as follows:(1)By studying and analyzing the characteristics of the underground roadway environment,the relevant requirements for building a multisensor fusion system in the roadway environment are proposed.Select the system hardware and build corresponding motion and observation models.The system positioning and mapping problem is modeled,and the framework flow of multisensor fusion positioning and mapping methods for underground tunnels is given.(2)On the basis of laser point cloud distortion removal and in view of the characteristics of multiple vertical walls and ground point clouds in underground tunnels,a method for segmentation of wall and ground point clouds based on point cloud grid division is proposed,and the remaining point clouds are projected on the depth map.Based on an improved angle threshold breakpoint detector,the remaining point clouds in the depth map are segmented using breadth-first search.Experimental verification of segmentation effect and speed was conducted on the Kitti dataset and the slope scene of the college underground garage.(3)Based on the error state Kalman filter,a tightly coupled multi sensor positioning and mapping framework for laser,camera,and inertial navigation is proposed.According to the matrix inversion theorem,the calculation of the Kalman gain coefficient is reduced from the observation dimension to the system state quantity dimension,achieving efficient update of the gain coefficient.A priori estimate of the system state quantity is obtained by performing forward integration on the IMU measurement value,and then an observation equation for the residual distance from the laser point to the local plane is constructed to achieve iterative updating of the error state quantity,ultimately outputting the radar inertial navigation odometer.At the same time,point cloud maps are constructed using the efficient dynamic addition and deletion features of ikd-tree.In order to meet the positioning and mapping requirements for the degraded environment of underground roadways,based on the laser inertial navigation odometer,first,the optical flow method is used to determine the inter frame matching points,and rough estimation of the system state quantity is achieved through re projection.Then,based on the camera’s environmental photometric measurement,a photometric observation residual equation is constructed to achieve precise estimation of the system state quantity,and finally,the environmental photometric observation value is projected into the point cloud map through coordinate transformation,Realize the construction of color point cloud maps.Build a simulated roadway and hardware simulation system in the Gazebo simulation environment to verify the effectiveness of multisensor fusion positioning and mapping algorithms.(4)A hardware system for multisensor fusion is built.Calibrate the internal and external parameters of the sensor.Select college corridors,underground garages,and simulated roadway laboratories for positioning and mapping experiments.Finally,the multisensor fusion localization and mapping method proposed in this thesis has better capabilities in pose resolution and map construction compared to pure radar localization and mapping and radar inertial navigation tightly coupled localization and mapping methods.The thesis has 89 figures,6 tables,and 83 references.