Research On Obstacle Detection Technology In Front Of Unmanned Sweeper Based On Lidar And Vision Fusion

Intelligent driving is one of the global research and development hotspots in the field of vehicle engineering.The active safety of intelligent vehicle is a research hotspot of intelligent driving research institutions at home and abroad.It is also a social issue related to people 's life safety and property.This thesis focuses on the perception of the operation of the unmanned sweeper,using lidar and on-board cameras to collect environmental information around the sweeper in real time.In terms of processing and analyzing the collected data quickly and accurately,a technical solution of obstacle detection based on the fusion of lidar and vision was proposed,achieving the test of obstacle in front,so as to obtain the position information and geometric characteristics of the obstacles in front.This paper is organized as follows:Firstly,according to the working condition of unmanned sweepers and requirements of research,an environmental awareness platform for unmanned sweepers was built;by establishing three kinds of coordinate system of unmanned sweepers,and converting these coordinate systems,the thesis completed the calibration of the camera and lidar,which provided a research basis for the matching of multisensor data in time and space.Secondly,a three-dimensional lidar data de-noising and streamlining method that can adapt to different distances was proposed.After the original point cloud data was rasterized,de-noised and filtered,the nearest neighbor distance clustering method was assigned to determine the adaptive threshold,and the clustering method of fused grid connected labels was assigned to reduce discrete points,improving clustering accuracy,achieving obstacle detection under different distance.Experimental results shown that,compared with other existing clustering algorithms,the new method proposed in this paper can adapt to the obstacle clustering at different distance and improve the recognition ability of road obstacles.Thirdly,the method based on AlexNet convolutional neural network realized the detection and recognition of obstacles in front of the vehicle.The images collected by the car camera were grayed and de-noised,and the road surface was separated by an adaptive threshold.After the hypothesis was generated for the area where the vehicle was located,the target features were extracted through the convolutional neural network to verify the hypothesis area.Finally,a platform was built for obstacle detection algorithm implementation and simulation testing,and analyzed the effect of single sensor detection and multi-sensor fusion detection of obstacles in front.The results shown that the fusion of lidar and visual information reduced the interference of external factors.The features of obstacles were collected more comprehensively and accurately,which accelerating the speed of data processing and improving the safety of autonomous driving of autonomous vehicle.