Experiment and Evaluation on Multi-Sensor Fusion Data of Car-Following Scenarios 2020-01-5180
To verify the accuracy, feasibility, and stability of multi-sensor fusion data of the car-following scenarios, a test method, and the field experiment is proposed to verify the performance of sensor fusion data. The testing vehicle loads a high-precision ranging system that combines a base station with RT-range system for communication, which can calculate the relative longitudinal distance and velocity between the target vehicle and hunter vehicle in real-time. By comparing the data of multi-sensor fusion with RT-range system, the test method not only analyzes the fusion sensors data trend, average tolerance, and RMSE value but also further evaluates the performance of the fusion algorithm. The result indicates that the multi-sensor fusion data generated with lidar and camera & radar for the car-following scenarios meet the key performance indicators of relative longitudinal distance and velocity. The relative longitudinal distance and velocity of multi-sensor fusion algorithm data have the nearly identical trend with RT-range data as reference ground-truth value, according to the four groups road tests. Therefore, the fusion algorithm can respond to the trend of distance or velocity in real-time, which verifies the algorithm is feasibility for the car-following scenarios. Besides, in the road test of random car-following, the relative longitudinal distance average tolerance of the four group tests was 0.99%, 1.07%, 0.99%, and 1.12%, respectively, which were floating at 1%. The relative longitudinal velocity average tolerance of the four groups’ tests was 1.68%, 1.99%, 2.30%, and 2.66%, respectively, which were floating at 2%. The above results not only verifies the accuracy of fusion data but also reflects that multi-sensor fusion algorithm is stability for the car-following scenarios. Moreover, for the fusion data, the RMSE value of relative longitudinal velocity is always between lidar and camera & radar, which further reflects that the algorithm is feasibility and stability.