Research On Key Technologies Of Cognitive Imaging Lidar Based On Deep Learning

Lidar has the characteristics of extremely high angular resolution,range resolution,speed resolution,etc.,and can obtain the target's angle,distance,speed,reflection intensity and other information.Lidar can not only find targets when scanning a wide range,but also obtain complete geometric information through fine scanning.It also has the advantages of strong anti-interference ability and is not limited by light conditions.It is widely used in fields such as unmanned driving and target detection.However,the traditional lidar detection information exchange direction is simple,which cannot adapt to the needs of multiple working mode switching such as large-scale wide-area search and small-scale fine recognition,and achieve the purpose of cognitive imaging.In response to the above problems of traditional imaging lidar,this article focuses on the development of the cognitive imaging lidar mode based on deep learning,the precise detection of point cloud targets based on deep learning,and the design and implementation of the cognitive imaging lidar module based on deep learning,etc.,the key technology research,the specific content is as follows:(1)Aiming at the problems of fixed working mode of traditional imaging lidar detection,poor environmental adaptability,single direction of information interaction and low data utilization,this paper analyzes and designs a cognitive imaging lidar mode based on deep learning.By integrating the deep learning intelligent algorithm into the lidar imaging parameter control,the cognitive imaging lidar closed-loop feedback control function is realized.Solve the problems of imaging lidar environment adaptability and single direction of information interaction in different scenarios,and improve lidar detection ability and detection accuracy.(2)For deep learning detection algorithm in target detection point cloud problems such as the limited information loss and receptive field,this paper studies and realizes a kind of used to point cloud data based on deformable attention and rasterize the point cloud of accurate monitoring network,through in the voxel point cloud detection network structure to strengthen the global context information,make up for the point cloud voxel information leakage problem;Added attention to deformation characteristics of the study on random sampling point location point cloud,for the most representative characteristics of point cloud subset,based on the characteristics of grid near the point cloud of building blocks,to expand the global context information to a larger area of the point cloud,solve the close and distant targets on the point of the precision of the huge imbalance fell problem,achieved the effect of ascending cloud point target detection precision,by comparing experimental results with typical algorithms,this paper studies the depth of the study on test network in point cloud target detection precision has a better performance.(3)To verify the effectiveness of the design of the cognitive model of imaging laser radar,this paper designed and developed based on the deep learning cognitive scheme and principle of imaging laser radar demonstration module,realized based on the deep learning cognitive imaging laser radar point cloud detection network closed loop control function,achieve the lidar model intelligent switch,promote environmental adaptability and the purpose of improving the detection precision.Outdoor experiments prove that the cognitive imaging lidar module designed in this paper can be applied to the fine detection task of typical targets.