Research And Application Of Semantic Map Construction Method For Operating Environment Of Automatic Crane

Unmanned autonomous operation equipment in various industries in society has made large-scale applications,with the progress of technology and changes in demand,cranes are gradually developing in the direction of unmanned and intelligent,which cranes intelligent mainly involves building maps,positioning optimization,autonomous navigation and other technical modules.Crane perception and understanding of the operating scene is an important manifestation of the development of intelligence,and the research and application of semantic maps in crane operations has played an important role in promoting.This topic through the research of domestic and foreign semantic information extraction,multi-sensor fusion,map construction and other related literature,the semantic map,multi-sensor fusion and other technologies and crane applications organic combination,the construction of crane operation scene semantic point cloud map,the main research content is as follows:(1)According to the actual operational requirements of cranes,the main coordinate system of the system is defined,and the calibration methods between multiple sensors in the crane system are studied and completed for calibration and verification,including the external reference calibration between solid-state LIDAR and crane based on Singular Value Decomposition(SVD),monocular camera based on checkerboard calibration board(2)The external calibration between solid-state LIDAR and crane based on Singular Value Decomposition(SVD),monocular camera based on checkerboard calibration board,and external calibration between solid-state LIDAR and monocular camera based on black rectangular calibration board are studied.(2)The map construction and repositioning algorithm of solid-state LIDAR is studied.Firstly,the point cloud motion compensation is completed based on the crane encoder positioning information,and then,considering the dust interference in the operation scene,various point cloud clustering methods are combined to filter outliers to ensure the neatness and clarity of the map construction in the operation scene,and finally,the map construction and update in the dynamic environment is completed based on the octree.In addition,considering that there is an accumulation of errors when the crane actually relies on the encoder positioning,this paper will scan-context algorithm positioning results and encoder positioning information for filtering and fusion,real-time correction of encoder positioning errors to achieve long-term robust repositioning.(3)Due to the different data collection frequencies of sensors in the system,the data timestamp synchronization method between solid-state LIDAR and monocular camera is studied,and the data fusion between point cloud and image is completed to construct a color point cloud of the operation environment.The crane operation environment is simulated and the homemade data set is trained on Center Net++ neural network to realize the detection of common targets in the crane operation environment,including coal,colored stones,model carts,plastic boxes,etc.Finally,the detection results are fused with the 3D point cloud and instantiated semantic segmentation to realize the 3D semantic point cloud map construction under the crane scene.This topic is aimed at intelligent crane operating environment for semantic map construction,the combination of sensor-aware information and deep learning to create a high-level abstraction of the operating environment,based on the semantic map built out,the crane can develop a variety of functions such as real-time sensing of the operating environment,path planning,assisted task decision-making,hazard warning and three-dimensional modeling of the environment,so as to achieve efficient,intelligent and safe crane The crane can be operated efficiently,intelligently and safely,providing effective reference and help for the intelligent and unmanned development of crane field.