Detection And Recognition Of Road Environment Objects Based On Multi-line Lidar

The Unmanned Ground Vehicle system can be applied to reconnaissance,patrol,transportation and other tasks,which is of revolutionary significance to the mode of production in the civilian field and the mode of operation in the military field.As the cost of lidar sensor is reduced and the computing capacity of computer hardware is greatly improved,the environmental sensing technology based on LIDAR becomes a research hotspot.This paper takes the small Unmanned Ground Vehicle as the experimental platform to study the detection and recognition methods of targets in road scenes.The technology of detection and recognition of common positive obstacles in road scenes has been relatively mature,but the detection and recognition of negative obstacles in unstructured road scenes is still a difficult problem,which have a great threat to the Unmanned Ground Vehicle.At present,the vehicle and pedestrian target detection and recognition method with the highest accuracy is the deep learning method.However,the disorder and irresolution of point cloud data increase the difficulty of data processing,and the 3D convolution operation consumes a lot of time and computing resources.it is difficult to meet the real-time requirements of Unmanned Ground Vehicle.Based on this,this paper proposes the following innovations:1.Aiming at the problem of sparse point cloud data in the method of negative obstacle detection and recognition,a method of negative obstacle detection and recognition using a single Lidar was proposed to solve the problem through point cloud registration and data fusion,which improved the scene adaptability of negative obstacle detection and recognition.2.Aiming at the problem that 3D convolution operation consumes a large amount of computing resources,a deep network based on free of 3D convolution was proposed.It achieved uniform sampling through farthest points sampling.At the same time,local characteristics of the data were better preserved.It realized the detection and recognition of vehicle and pedestrian,reduced the running time of the algorithm,and improved the detection accuracy.The Unmanned Ground Vehicle in this paper was designed according to the four-layer control system of sensing,fusion,planning and control,and the target detection and recognition experiment was carried out on the experimental platform.The experiment showed that the method proposed in this paper can improve the accuracy and robustness of target detection of the Unmanned Ground Vehicle,and at the same time had good scene adaptation performance and real-time performance.