Research On Environment Perception And Local Path Planning Algorithm Based On Lidar And Vision

With the popularization of intelligent mobile robots,autonomous driving technology has attracted more and more attention.As an important technology in autonomous driving,environment awareness and path planning have been deeply studied by many scholars.Recent studies show that the reliability of using a single sensor to perceive the environment is insufficient.The proposed method can realize all-day and all-weather environmental perception.Path planning is supported by environment-aware information,and the result of environment-aware information determines the result of path planning.The main research contents of this paper include: semantic identification based on visual information,environment perception based on Lidar and vision,and local path planning algorithm based on obstacles and semantic identification.Specific work is as follows:First,traffic light recognition and lane detection are realized through visual information,and obstacle detection is realized through lidar.A functional network N＿Res Net is proposed for traffic light recognition and lane detection.Compared with residual neural network,N＿Res Net network has larger receptive field and can obtain more reliable deep features.The N＿Res Net network was applied to YOLOv5 algorithm to realize traffic light recognition.Compared with the original network,the recall rate of the network proposed in LISA data set increased by 2.66%,and the recall rate of traffic lights in the night environment increased by 11.32%.The N＿Res Net network was applied to the CLRNet algorithm to realize lane detection.The networks proposed in Tu Simple data set obtained 97.01% lane detection accuracy and 97.92% recall rate,respectively,and the networks proposed in CULane data set obtained 79.89% recall rate.Secondly,on the basis of the traditional TEB local path planning algorithm,traffic light information and lane information are added,so that the path planning algorithm can be applied to a variety of environments.According to the status and distance of traffic lights,different accelerations are set to realize the constraints of traffic lights on path planning.According to the category of lane lines,different cost weights are set to realize the constraints of lane lines on path planning.Experiments show that the proposed local path planning algorithm can plan reliable paths according to abundant environmental information.Finally,the proposed algorithm is verified on the simulation platform and hardware platform.The results show that the proposed traffic light recognition algorithm achieves higher recall rate and accuracy.The proposed local path planning algorithm is capable of planning safe and efficient paths based on semantic identification and obstacle information.