Modeling of urban scene by motion analysis

A flexible mechanism for measurement of moving objects is necessary to acquire data from various wide-range real spaces. The authors aim at development of a technique for wide-area data acquisition by measurement of moving objects such as vehicles and aircraft. As part of the research, this paper discusses urban space design using images taken from a vehicle. Urban space modeling is a fundamental element in many fields such as urban engineering, traffic management, education, or entertainment. However, at the moment, data acquisition involves enormous expense while requiring a lot of manual input operation which imposes heavy restrictions on possible applications. Proposed in this study is a method for urban scenery modeling based on analysis of moving images taken from a running vehicle. In this method, depth is measured using moving stereo. Buildings and other objects are represented by distance from the observation path (road) and surface pattern (texture), in the form of a striplike depth map. In experiments with animated images, the depth error was 4.4% through 8.13%. The effectiveness of the proposed method was also demonstrated by experiments with real images taken from a running vehicle. © 2000 Scripta Technica, Syst Comp Jpn, 31(2): 1–9, 2000