A multi-sensor-based navigation framework for intelligent vehicle
Ratsame Photchara, Surapa Thiemjarus, Toshiaki Kondo · 2010
This paper presents a platform-independent framework for autonomous navigation of an intelligent vehicle. The framework consists of three integrated modules, namely; waypoint navigation, obstacle localization and path planning. Each module has been individually validated based on experiments with a real intelligent vehicle. For waypoint navigation, we propose the use of Google Earth for generation of reference waypoints and a simple bias subtraction method for GPS calibration. Based on the autonomous navigation experiments, this method yields a more stable navigation path compared to the use of GPS-generated waypoints and translational error can be efficiently eliminated. For obstacle localization, we develop a feature-based approach for obstacle detection and map generation based on the use of compactness measure and perspective projections. With an integrated use of a camera, digital compass, and GPS, static obstacles of a known dimension, along with their positions and orientations on the road can be calculated in real-time while the vehicle is travelling. Based on the derived information, an overhead-view obstacle map is generated to provide an internal representation of the road. The experiment on an unmarked road shows that the estimation of an obstacle can be achieved with maximum errors of 1.4 degree, 15 cm and 12 cm in angle, depth and lateral positions, respectively. Based on the obstacle map and the Google Earth waypoints, artificial potential field is adopted for collision-free path generation.