期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2007
卷号：XXXVI-5/C55
出版社：Copernicus Publications
摘要：Efficient inventory and analysis of highway and road network features requires a reliable and accurate mobile mapping system (MMS). Advances in Geomatics technologies such as the Global Positioning System (GPS), Inertial Navigation Systems (INS), and laser/ultrasonic imaging sensors made it possible to offer such system. The ultimate objective of this research is to suggest a new INS/GPS integration technique that improves the positioning accuracy of the overall system, especially during relatively long GPS outages, which could be experienced by a MMS in urban canyons. If left unaided, INS position errors grow to large values due to the mathematical integration of sensor errors performed during the INS mechanization procedure. In case of long GPS outages, the MMS may suffer from an increasingly position and attitude errors over time. It was also reported that Kalman filtering may not provide reliable estimates of the INS position errors during relatively long GPS outages, especially when tactical and low cost INS is utilized for the MMS. Several artificial intelligence techniques were proposed as replacement for Kalman filtering. However, none of these methods considered the time dependence nature of the INS error, which may also lead to inadequate performance during long GPS outages. Thus, a model capable of establishing time-dependent relationship of the INS errors during long GPS outages is necessary. This research proposes a dynamic neural network model for the INS position and velocity errors utilizing Input Delayed Neural Networks (IDNN). Such network architecture depends not only on the current input to the network but also on few previous inputs and outputs. While the navigation system is relying on INS during GPS outages, IDNN model mimics the patterns of the INS errors and provide reliable prediction of the INS position and velocity errors. The proposed IDNN model is evaluated using real field test INS and GPS data collected during real road tests. The results showed that the IDNN model provided at least 25% enhancements in the positioning accuracy if compared to Kalman filtering and other artificial intelligence models
关键词：Mobile Mapping; INS/GPS; Data Fusion; Dynamic Neural Network
