| Résumé : |
(Auteur) Autonomous car navigation systems with on-board positioning sensors and CD-road network require a road accurate positioning relatively to the road network for optimal route planning, for route guidance and for generating driving recommendations. Positioning effected in a certain reference system and its accuracy are not decisive but represent the logical location of the vehicle relatively to the digital map. This task is called map matching.
Modern car navigation systems use for positioning multi-sensor-systems (differential odometers, gyros, magnetic field sensors) in combination with GPS or DGPS. Little availability of satellite methods due to masking and second way effects in urban areas (85% of mobility activities in road traffic occur in urban areas), magnetic disturbances, wheel slips and unfavourable error propagation of dead reckoning lead again and again to loss of position and in principle to total system failures. This can be avoided by map matching which compensates the sensor typical error behaviour of positioning sensors. The systematic error behaviour was examined theoretically for the positioning sensors odometer (wheel sensors, i.e. ABS-sensors), magnetic field sensors, gyros, accelerometers, GPS, and DGPS. Based on these error models five map matching methods with different degrees of freedom were developed, estimated correspondingly to the various ways of presentation of plain curves in curvature pattern, in angular pattern, or with two-dimensional cartesian coordinates.
For orientation or curvature map matching the distance frives (ruve length) and the values of orientation (direction angle or curvatures) are derived from these measurements. This measured function over the curve length is fitted to the corresponding function of the map by estimating one offset and one scale factor in a least square adjustment between the two functions and the two curve lengths.
In case of coordinate procedures the driven distance is divided into line elements of constant length and compared with the line elements of the digital map. For matching onto the digital road network, three approximating transformation translations, plain similarity transformation and affinity transformation are studied. The sensivity of these five methods concerning systematic sensor errors is examined by a precise analysis by means of simulated observations for the sensor systems gyro with odometer and gyro with accelerometers.
Scale factors and bias of gyros can be compensated by curvature matching without any calibration of these errors. Accelerometer errors gradually lead to squared increasing errors regarding the measured distance, and cannot be matched without error compensation to any method on the map.
Based on real observations with GPS, DGPS, wheel sensors, and a strap down inertial system being investigated during test trips under normal traffic conditions the map matching procedures could be tested for various combinations of these sensors. The accuracies reached and the reliability of track identification of the exact track out of various alternative tracks being especially important for vehicle navigation, were analysed and valued.
In urban areas pure GPS measurements are often useless, as every change in satellite constellations leads to disturbances of the vehicle position which cannot be compensated by one of the methods.
DGPS needs a support by wheel sensors to bridge masking effects and to moderate drift effects during extremely slow motion. Measurements by sensor combinations such as gyro with wheel sensors, differential odometer, GPS with differential odometer, DGPS with differential odometer and pure DGPS (without long masking times) can be matched onto the map with all methods (besides translation), if at least two turns are contained within the data material.
The precision of the map matching process to the coordinate pattern decreases with growing distance. The orientation and curvature matching compensate systematic sensor errors considerably better and increase accuracy with growing distance.
The limits of positioning systems and map matching techniques are shown in problematic cases such as masking of the satellite signals, little curving of the track, track identification in nearly parallel alternative tracks.
Autonomous navigation systems based on odometers and gyros reach, by map matching onto the road net, more favourable positioning results at an accuracy of 70cm than pure DGPS measurements so that in urban areas a reliable map matching without DGPS is possible. The integration of DGPS does not show a considerable increase in accuracy, nevertheless the integrity of hybrid systems is increased. |
Leistungsfähigkeit fahrzeugautonomer Ortungsverfahren auf der Basis von Map-Matching-Techniken (2001)
