Advanced Driver Assistance Based On Cooperative Perception And Augmented Reality

Nowadays,intelligent vehicles have been developing rapidly;besides,increasing popularity of Vehicular Ad-hoc Networks(VANets)technology stimulates intelligent vehicles to share information through VANets in real time to enhance their perceptive capability towards themselves and the environment.Moreover,vehicular augmented reality technology progresses steadily and can provide drivers with more and more vivid and intuitive driver assistance information.Combining vehicle cooperation technology and vehicle augmented reality technology,advanced driver assistance based on vehicle cooperative perception and augmented reality is studied,which aims at overcoming the view occlusion problem that any single vehicle can never handle and aims at effectively extending and enriching the driver’s view.This research mainly involves the problem of inter-vehicle cooperative relative localization,and the design and implementation of multi-vehicle cooperative perception and augmented reality system.Relative localization between vehicles is a key technology for multivehicle cooperative perception,which ensures the consistence of perceptive information sharing among vehicles.Cooperative relative localization via environmental information association can estimate the inter-vehicle relative pose effectively,given a proper initial value.On the other hand,initialization for the cooperative relative localization process is difficult,especially in scenarios where satellite signals deteriorate severely.To address this problem,a cooperative relative localization method based on vision-Li DAR fusion is proposed,which exploits the advantage of the visual system in relative pose initialization and the precise positioning ability of the Li DAR.Pure Li DAR based inter-vehicle cooperative relative localization is also studied.To address the initialization problem,spirit of visual methods is borrowed.Li DAR point clouds are mapped into two-dimensional images;relative pose initialization-oriented place recognition and pose estimation are performed in similar way as in visual methods.Neural network technology is adopted for feature extraction to enhance the accuracy of place recognition and pose estimation.The proposed method achieves a real-time and high-precision performance for cooperative relative localization.A system framework of multi-vehicle cooperative perception and augmented reality is proposed.The method is evaluated in Carla simulator which is widely adopted for research purpose in the intelligent vehicle field.In scenarios where view occlusion exists,the proposed method can effectively realize the augmented reality effect of “seeing through”,which would be valuable for driver assistance.