A Novel Method for Ground Vehicle Tracking With Error-State Kalman Filter-Based Visual-LiDAR Odometry (ESKF-VLO)

An autonomous vehicle requires a self-tracking system that can operate both indoors and outdoors using only on-board sensors. In this work, a vehicle self-tracking scheme called Error-State Kalman Filter-Based Visual-LiDAR Odometry (ESKF-VLO) that uses camera, LiDAR, and IMU sensors is proposed. In this scheme, to obtain the 3D vehicle transformation between two consecutive time frames, the corresponding feature points are detected in the captured camera images and the respective 3D location of the feature points is obtained from LiDAR points. The relationship between 3D LiDAR points and camera image pixels is established using the sensors’ LiDAR-to-Camera calibration parameters. In the final step, IMU data is corrected with this transformation matrix using the Error-State Kalman Filter, and the vehicle position is tracked accurately. The proposed method’s efficacy is verified using the KITTI dataset and data obtained through indoor experiments conducted using the P3DX mobile robot. The results show that the proposed method is able to track the vehicle, and the translational and rotational errors are reduced compared to the existing methods.