The performance evaluation of a multisensor data fusion system for risk detection designed for an autonomous guidance vehicle (AGV) operating in an outdoor environment has been presented. The vehicle uses a combination of visual, range and ultrasonic sensors and works in an outdoor complex environment, characterized by holes, terrain slopes and trash heaps. The core of the multisensor data fusion system consists of a distributed sensory network organized into a hierarchical tree structure, composed by as many leaf nodes as the number of real sensors employed, e.g., serveral cameras, a laser range finder and some ultrasonic sensors. Nodes at the second level are devoted the extract features from sensor data. The third level of the tree (low-level data fusion) is composed by a set of virtual sensors which integrate the informative content of one or more features to perform different tasks: (a) detection of static obstacles, (b) detection of holes or down slopes, (c) recognition of specific fixed objects (e.g., trash heaps), (d) detection and localization of mobile objects (e.g., other vehicles, pedestrians, etc.). The root node (high-level data fusion) integrates the evidences coming from each virtual sensor in order to obtain a fused information in the general reference system about the number of objects and their 3D pose. Finally, the fused information is used to decide the best action to do, e.g., stop the vehicle, moving the vehicle in a given direction, etc. A global measure of the performances is used to evaluate the efficiency (in terms of gain or loss) of the system when a given sensor or a set of similar sensors is add to the system. The contribution of a single sensor to the system performances is weighted on the basis of the accuracy and reliability of the measure. Experimental results on a real environment demonstrate the effectiveness of the proposed approach.
Performance evaluation of a multisensor system for autonomous guidance vehicles
Leistungsabschätzung eines Multisensorsystems für autonome Fahrzeugführung
2000
10 Seiten, 5 Bilder, 1 Tabelle, 16 Quellen
Conference paper
English
Performance evaluation of a multisensor system for autonomous guidance vehicles
Automotive engineering | 2000
|00ROB009 Performance Evaluation of a Multisensor System for Autonomous Guidance Vehicles
British Library Conference Proceedings | 2000
|