Multi-Sensor Fusion Methods for Off-Road Vehicle Guidance System

摘要

The technology of automatic guidance system for off-road vehicle is one of the key technologies of precision agriculture. Accurate and reliable positioning information is the premise for the guidance system. In order to improve the accuracy of positioning and the stability of guidance system, a multi-sensor fusion system was built and the new sensor fusion method was developed. The used navigation sensors included a XW IMU5100 digital micro machined inertial measurement-Attitude and Heading Reference System (AHRS) unit, a Trimble AgGPS332 (RTK-GPS) system and a self-developed GPS (RTD-GPS) receiver. A customized electric motor vehicle was used as the experimental platform. The data collection and management software was developed in mixed environment with VC++ 6.0 and Matlab. The data measured by RTK-GPS were used as the standard values because of high accuracy, and the data from RTD-GPS and IMU were integrated by different sensor fusion methods. Then the performance of multi-sensor fusion models was analyzed and compared. After data alignment, those multi-sensor fusion models, Unscented Kalman Filter (UKF), Particle Filter (PF), and a new method-Unscented Particle Filter (UPF), were established to filter the information of RTD-GPS and IMU based on the kinematics model of off-road vehicle. The third runway of standard 400 meters playground was selected as the navigation path, and three kinds of filter methods with different particles were tested. Experimental results showed that the UPF using shorter time could achieve the same accuracy of UKF and PF. UPF with 50 particles have the same accuracy with the PF with 1000 particles, however the consuming-time was only 21.51% of it. The simulation experiment was repeated 100 times with random re-initialization for each run, and the result indicated that UPF method had stable performance, high precision and real-time, and could fulfill the needs of agricultural operations.