Study On Lidar Obstacle Detection And Camera Image Recognition For FSAC Racing Car

Environmental perception is the premise of autonomous driving.Aiming at the environmental perception technology of autonomous formula car,this paper studies the lidar obstacle detection and camera image recognition algorithm,aiming at providing necessary information for the path planning and vehicle control of autonomous formula car,and realizing the safe driving of the car.Based on the general project of national natural science foundation of China(51675257),key research and development plan of Liaoning province(2017106020),major science and technology platform project(JP2017001),innovation talent project of higher education institutions of Liaoning province(LR2016054),and conducts subject research in combination with the research and development of autonomous formula racing.In this paper,an algorithm of target pile bucket detection and color recognition of driverless formula racing cars in the competition environment is studied,and the lidar data and camera data are fused to realize that the perception system of autonomous formula racing cars can quickly locate the target bucket and identify its color.The main work is as follows:For the detection of target buckets,our approach is to limit the axial distance of the Lidar raw data,retain the effective Lidar data around the car,then use the plane segmentation model algorithm to separate target buckets from the ground and filter out the ground,and finally use euclidean cluster to extract target buckets data.For the color recognition,in order to ensure the accuracy of the color recognition,this paper get the Lidar and camera jointly calibrated,then project Lidar data on the corresponding image according to the coordinate conversion principle,and then we calculate the interest field of the target buckets on the image,and finally use HSV color recognition algorithm to identify the color of target buckets.For the data fusion,in order to ensure Lidar data is projected on the image accurately and in real time,this paper design one data fusion system suitable for autonomous racing car based on Robot Operate System.The result of actual vehicle experiment and the competition show that the algorithm of the thesis can quickly detect the target bucket and accurately identify the color of the them on the autonomous formula racing car,which lays a good foundation for the path planning and vehicle control of the car.The autonomous formula racing car with this algorithm won the third place in the 2018 FSAC competition.