Enhanced navigation system with real images and real-time information

This paper presents a new visualization system, which integrates multiple kind of information for car drivers. The prototype of this system consists of two key applications. The first one is a car navigation system with real world texture images. In this system, buildings are texture mapped by actual images, which are retrieved by our automatic mapmaking system. The second is a hybrid visualization system, which efficiently visualizes the invisible cars behind the buildings. This system is expected to reduce car accidents that occur around intersections, places where drivers usually have only very narrow fields of view. To realize these systems, we propose an efficient building texture database construction system which, for input, uses the video data captured by a vehicle equipped with multiple video cameras and running along the street. To demonstrate the effective ness of our system, we present some experimental results with actual outdoor environmental data. Introduction With recent progress in virtual reality systems and car sensors (e.g. GPS and gyro-sensor), ITS (intelligent transportation system) applications tend to use a wide variety of information from both real world (e.g., traffic information) and virtual world (e.g., digital map) in conjunction with one another. The fact that every information has its own data structure necessitates the integration of this information into a common structure. Since our proposed system is an enhanced navigation system in which real world texture images and digital map data are merged efficiently, we propose a new matching technique between real-world video and digital maps for data integration. We also propose a driver assistance system, which effectively integrates other car location data into digital maps and visualizes them naturally. This system efficiently makes cars occluded by buildings visible to the driver; and therefore, is expected to reduce the car accidents around intersections, locations where drivers usually have a narrow field of view. Considering the matching between real-world images and digital map, we cannot simply employ the matching techniques for different kinds of data, therefore first we have to make the same kind of data structure for both real-world video data and digital maps. Then subsequently we can employ matching between these two data. To make the same kind of data structure, we use EPI analysis in this research. And for matching, we use DP matching for improving the precision and reliability of the results. In addition, we propose an original configuration of multiple cameras and data capturing system for efficient acquiring of the building images. Once we generate the texture image database, which certainly related to the digital map, we can easily realize the car navigation system with real world texture. Therefore, in the latter half of this paper, we propose the driver assistance system which efficiently visualizes the invisible cars occluded by the buildings, truck or other huge obstacles. This kind of information is very useful for reducing the traffic accident occurred at an intersection where drivers usually have very narrow field of view. This paper is organized as follows. In Section 2, we state the background of our research. Then, in Section 3, we describe the principle of our map-making system. In Section 4, we present the enhanced navigation system with experiments and their results. Finally, we conclude this paper in