Research On Multi-sensor Fusion Mapping And Localization Algorithm For Unmanned Vehicle

In recent years,unmanned technology has made great progress,including automatic driving cars,intelligent unmanned aerial vehicle(UAV),and so on.In this field,simultaneous localization and mapping(SLAM)is the premise of the development of unmanned technology.when the system has a high-precision map and strong robust positioning,it can make the mobile robot do effective path planning.In the research of SLAM technology,a variety of sensors are involved,such as camera,lidar scanner,IMU and wheel speed pulse,etc.The camera has the advantages of low hardware cost and rich acquired information characteristics,but is limited by the working environment and the movement state of the carrier.The lidar scanner has high robustness.The acquired point cloud data contains three-dimensional information of the scene,but it is not suitable for large empty scenes.The output frequencies of IMU and wheel speed pulses are high,but due to the working principle of the sensor itself,it is easily affected by noise,and serious error accumulation occurs over time.Therefore,in this paper,by combining the advantages and disadvantages of various sensors to improve the positioning accuracy and robustness of the unmanned system as the fundamental purpose,a multi-sensor fusion positioning and mapping algorithm is proposed.The main work of this paper is as follows:(1)The mathematical framework and system model of vision and lidar SLAM are analyzed,as well as the components and functional principles of each module in the classical positioning and mapping system.The vision system model applied in this study is discussed in detail based on the front-end odometer,back-end optimization,loopback correction and sparse map modules.the lidar system model applied in this study is also explained in detail based on the front-end odometer,mapping,loopback detection and point cloud map construction modules.(2)The multi-sensor calibration algorithm and related mathematical theory are analyzed.The calibration algorithm includes camera internal parameter calibration,camera and IMU external parameter calibration,camera and lidar scanner external parameter calibration.In order to achieve observability of calibration accuracy,an occupation raster map is proposed and its mathematical model is introduced.The occupation raster map is generated by integrating point cloud data obtained from lidar scanner and trajectory pose obtained from visual positioning system.(3)A multi-modal sensor fusion algorithm is developed.The system framework of the algorithm integrates multiple sensors such as lidar scanner,IMU sensor,camera and wheel speed pulse.It provides a fusion method that meets the requirements of accuracy and real-time at the same time.It can realize robust state estimation without visual features or spatial structure degradation.The fusion algorithm is divided into two modules: mapping and positioning.In the mapping module,the camera and lidar sensor construct the visual map and point cloud map,and correct the corresponding map with the real track.The positioning module applies the basic framework of error kalman filter.Firstly,the predicted value is output by IMU and wheel speed pulse sensor to make the system have high-frequency characteristics.Secondly,the observation value is positioned on the corresponding map through vision and lidar positioning algorithm,which improves the accuracy and robustness of the system and greatly improves the disadvantage of large cumulative error.(4)A mobile robot platform was built and applied.The multi-sensor fusion algorithm studied in this paper was implemented in C++ programming language and relevant experiments were carried out on the platform.Then the experimental results were analyzed,evaluated and compared.The loose-coupled SLAM algorithm combined with IMU,wheel speed pulse,vision and lidar scanner is compared with the traditional visual SLAM algorithm,the traditional lidar SLAM algorithm and the wheel mileage calculation method.The experimental results not only verify the real-time performance and effectiveness of the algorithm,but also show that it has great advantages in precision and robustness.