Enhancement of LiDAR Data Association and Fusion Using Imaging Radar Grid-Maps for Advanced Automotive Environment Perception

摘要

The use of grid-maps is a common approach in autonomous driving applications to register information about a vehicle's surroundings. Due to the high randomness of the environment and the uncertainty of the vehicle sensory system, a grid-map should provide broad information about the state of a vehicle's surroundings - an essential task for the autonomous driving decision-making system. In this paper, we present a grid-map which does not merely provide information about the probability of occupancy but includes additional information about the possible dynamic and static regions in the environment. In addition, we show how the dynamic and static region's information enhances the performance of the data association component in our LiDAR-based tracking system and LiDARbased static object representation. In our experiment, a Binary Bayes Filter is used to calculate the probability of occupancy. Also, the dynamics of the cell is modelled as a nonhomogenous Poisson process to estimate the probability of changes in the cell dynamic. This approach was tested and validated using data from an automotive imaging radar and Ibeo LiDAR mounted on an Ibeo test vehicle.