Research On The Key Technologies Of Dynamic In-car Navigation Systems

The development of mobile Internet makes in-car navigation systems changefrom the static autonomous pattern to the dynamic collaborative pattern. Originalclosed independent architecture is broken and it will be replaced by a more openarchitecture. Meanwhile, large numbers of portable intelligent terminal devices makenavigation systems not only run on in-car terminals, but run on smart phones, tabletPCs and other intelligent terminals supporting Internet access as well. Navigationsystems are gradually evolving into services a user can use anytime, anywhere. Thesechanges will bring navigation systems more technical challenges with regard tosystem architecture design, key technology implementation and application pattern.All these need to be made some appropriate adjustments to solve new problems.This paper analyzes the core needs of the dynamic in-car navigation systems:map display, navigation and route planning, and then uses the Event-B formalmodeling approach to model and analyze systems. In the refinement process, wediscuss the modeling of the related data structures and process. After being refinedfour times, the model achieves all of the core requirements. Importing each refinedmodel into Rodin platform, we find all generated proof obligations provedsuccessfully, which indicates that the model is theoretically correct. Using the modelas a reference, we design the software architecture which can be non-formallydescribed from two perspectives such as the logical view and the process view sothat system developers can design and develop a correct system. Next, we focus onthe map caching technology, map matching technology and intelligent informationprocessing technology of dynamic in-car navigation systems.On the map caching technology, this paper designs and implements a cachesystem based on two-level map tiling for dynamic in-car navigation systems. Thecache system contains index files, data files and some relative programs. We designthe index file structure, the data file structure and the cache structure used to storeSparse Matrix, and propose the effective strategies to manage the cache system on the basis of these data structures. Experimental results show that the cache systemhas good effects on speeding up response time of dynamic navigation system andreducing network traffic data. This paper also studies the cache prefetching strategy,which predicts the vehicle trajectory by analyzing the road network and theintersection in front. This strategy combines the prediction results by the roadnetwork analysis and the results by the heuristic prefetching strategy so that it canobtain more correct prediction results. Experiments show that this prefetchingstrategy can further improve the performance of the heuristic prefetching strategy,effectively reducing the number of needed map tiles.On the map matching technology, this paper analyzes various map-matchingalgorithms, determines the critical factor influencing the map-matching performanceis the junction match, makes a further study of the junction match, and propose ajunction decision domain model which includes the width of the junction-connectedsections, the angle between sessions, GPS accuracy and the road network dataaccuracy. This model is applied to improve the map-matching algorithm based onHidden Makov Model (HMM). Improved map-matching algorithm can effectivelyreduce the junction matching error rate and improve the stability of the navigationsystem. In addition, in order to solve the delay matching problem, this paper analyzesthe vehicle turning characteristics, designs the the vehicle turning recognitionmethod, and uses this method to improve the map-matching algorithm based onjunction decision domain model, shortening the time of making the matching pointstop at the intersection.On the intelligent information processing technology, this paper appliesmachine learning techniques to improve the vehicle turning recognition method, andstudies two learning methods to build vehicle turning recognition model. The firstone applies improved K-means clustering method combined with F-measure method;The second one applies anomaly detection method combined with F-measuremethod. The experimental results show that the models built respectively in suchtwo ways have good generalization ability. Based on the above research achievement,we design and implement a vehicle moving state recognition learning system for dynamic in-car navigation systems and apply this learning system to themap-matching field. The improved map-matching algorithm is tested on a complexurban road network and the experimental results show that the new algorithm cansolve the delay matching problem to some extent.