Accurate Perception for Autonomous Driving: Application of Kalman Filter for Sensor Fusion

Object tracking is a foundation task for autonomous driving. An increasing amount of work applies the sensor fusion to facilitate the tracking results. However, sensor fusion alone still cannot reach a desirable accuracy in real road conditions, because of the sensor noises and complexity of the motion dynamics. Bearing this in mind, we propose a method that applies Kalman filter on the LiDAR and radar sensor fusion to improve the accuracy in object tracking for the autonomous driving system. We evaluate our approach on the Udacity dataset. Results demonstrate that the Kalman filter drastically improves the final measurement compared to using sensor measurement alone. The work verifies the effectiveness of employment Kalman filter to facilitate the performance of sensor fusion measurements for the autonomous driving system.