In the field of autonomous vehicles, the accurate positioning is vital to the perception and motion planning and control. In order to solve the problem that the Global Positioning System (GPS) cannot function properly due to loss of positioning precision or reliability in certain GPS-denied environment, we propose an accurate positioning method based on GPS/Lidar/Camera in V2V communication environment. First, we utilize the 3D lidar data to generate object-region proposals and map these candidates into the image space, and input the regions of interest (ROI) to a convolutional neural network (CNN) for object recognition. Then, we acquire the GPS information of the neighboring vehicles through the V2V communication, and convert the GPS data from geocentric coordinate system to local Cartesian coordinate system by Universal Transverse Mercator (UTM) projection. After obtaining the relative position and global position of the adjacent vehicle, we can use inverse UTM projection to determine the global position coordinate of the current vehicle. Experiments show that our method can obtain the accurate global positioning result in the tunnel or urban canyon where the GPS signal is obstructed.

CICTP