MSFVT: A Multi-Sensor Fusion System for Vehicle Position Tracking in Unstructured Environments

Accurate position tracking of the leader vehicle is the key to the vehicle-following technology for the unmanned ground vehicle (UGV), but there are still many challenges to implementing it in unstructured environments, especially in complex weather and rough terrain. This paper proposed a multi-sensor fusion system (MSFVT) to solve these problems. This system achieves a real-time and robust leader vehicle detection and tracking based on LiDAR, millimeter-wave Radar (MMW Radar), and inertial navigation unit (IMU)/Global Navigation Satellite System (GNSS) fusion. The implementation of this system is divided into three steps: First, the point cloud from LiDAR is separated into different clusters based on an adaptive threshold clustering method and the clusters are classified using a support vector machine (SVM) to obtain the leader vehicle detection result. The detection result initializes the tracking process and maintains subsequent continuous tracking. Second, the validated detection results from MMW Radar obtained after multiple data filters are used to provide the position information of the leader vehicle during the UGV movement. Finally, the position of the tracking target in each sensor is converted to a global coordinate system using IMU/GNSS information and fed into the Kalman filter to estimate the motion of the tracking target. To verify the effectiveness of our method, we perform it in diverse unstructured environments. The experimental results show that our method has a smaller target position estimation error and an average processing frame rate of more than 40Hz.