Research On Localization And Mapping Technology Of Intelligent Grain Transfer Vehicle Based On 3D Lidar

With the increasing production of grain in China,the traditional grain depot transfer method is not only inefficient,but also produces a large amount of spilled dust in the process of grain transfer,which can be harmful to human body and environment.For this reason,modern grain depots are configured with centralized raw grain cleaning centers,and intelligent transfer vehicles can replace traditional transfer vehicle operations to realize the intelligence and greening of grain depots.In this paper,we study the key technology of intelligent grain transfer vehicle-positioning and mapping technology.The construction of a dual-laser SLAM system based on graph optimization is realized,and a 3D point cloud map is established.The repositioning technology is realized by using the 3D point cloud map as a priori map information.The specific work of this paper is as follows:(1)Build a test platform of intelligent grain transfer vehicle.In order to replace the transfer vehicle in the traditional grain depot and realize the grain transfer operation in the depot,the structural form of the chassis of the transfer vehicle is studied,the overall structure of the intelligent grain transfer vehicle is designed,and the hardware platform and software platform of the test vehicle are built.(2)The key technology of laser odometer construction is studied.In view of the problem of "canyon effect" in the grain depot environment,i.e.,the reliable vehicle positioning cannot be achieved by satellite positioning,the method of constructing laser odometer with LIDAR as the sensor is proposed to realize the positioning of the intelligent grain transfer vehicle.Therefore,the point cloud matching algorithm,feature point extraction algorithm and nonlinear optimization method are studied.Firstly,the curvature-based feature extraction method is used as the extraction method of point cloud information according to the characteristics of the actual application scenario.Secondly,the feature matching algorithm with point-to-line and point-to-surface as constrained relationships is constructed by the extracted feature points,which takes into account the local feature attributes of the feature point cloud,reduces the probability of wrong matching due to blind correspondence point finding,and improves the matching accuracy.Finally,the errors arising from the matching process are optimally solved by the nonlinear optimization algorithm of L-M.(3)Graph optimization-based dual-LIDAR information fusion localization is investigated.For a single LIDAR is subject to carrier occlusion,which leads to sparse and less robust environmental information,while too many LIDARs lead to difficulty and cost increase in data fusion.Therefore,according to the working mode and work content of the intelligent grain transfer vehicle,it is proposed to install two LIDARs on the left and right at the front position of the intelligent grain transfer vehicle to enhance the environmental information and overall robustness in the forward direction.GPS time is used to time the LIDAR to complete its time synchronization,and the spatial synchronization of the left and right LIDARs is realized by NDT algorithm.To ensure the accuracy of laser odometry and the realism of incrementally constructed maps,a graph theory-based graph optimization SLAM framework is studied to construct the positional vertices and add loopback constraints and ground constraints to them to correct the errors of positional and maps.By analyzing the cloud data collected from the actual grain storage site,the real reliability of the graph optimization-based method for map construction and localization is verified,and the effectiveness of ground constraint and loopback detection on the estimation of poses and map construction is verified.(4)The repositioning technique based on a priori map is studied.A relocation method is proposed for the grain depot environment which is relatively stable and the structured buildings do not change.Firstly,the data volume of the map is reduced without changing the structure by voxel raster filter for the established 3D point cloud map.Secondly,the 3D point cloud map is converted to 2D occupancy grid map by octomap library,and the initial poses are obtained by AMCL module,and the initial poses,point cloud map and real-time point cloud data are used as the input of NDT algorithm to realize the repositioning of intelligent grain transfer vehicle based on a priori map.For the process of grain transfer,some grain machinery will appear in the environment,therefore,the reliability and effectiveness of the repositioning technology are tested in the pure grain storage environment and the environment with other irregular objects respectively.