1992 Fiscal Year Final Research Report Summary
Intelligent Navigation System Which Autonomously Selects Landmarks along Route
Grant-in-Aid for Developmental Scientific Research (B)
|Allocation Type||Single-year Grants |
|Research Institution||Osaka University |
TSUJI Saburo Osaka Univ., Faculty of Engineering Science, Professor -> 大阪大学, 基礎工学部, 教授 (60029527)
YAGI Yasushi Osaka Univ., Faculty of Engineering Science, Research Associate, 基礎工学部, 助手 (60231643)
YAMADA Seiji Osaka Univ., Institute of Sci. & Ind. Research, Assistant Professor, 産業科学研究所, 講師 (50220380)
IMAI Masakazu Osaka Univ., Faculty of Engineering Science, Assistant Professor, 基礎工学部, 講師 (60193653)
YACHIDA Masahiko Osaka Univ., Faculty of Engineering Science, Professor, 基礎工学部, 教授 (20029531)
ASADA Minoru Osaka Univ., Faculty of Engineering, Associate Professor, 工学部, 助教授 (60151031)
|Project Period (FY)
1990 – 1992
|Keywords||Autonomous Robot / Dynamic Image / Range Image / Representation of Environment / Cognitive Map / Panoramic Representation|
A human can memorize selected patterns of route scenes as landmarks which guide him/her to move around the environment. This project builds an intelligent robot with such intelligence in order to develop the technology for understanding of the environment by computer.
[1 Panoramic Representation
A robot continuously takes images by a TV camera while it moves along a route and arranges the slit image of consecutive frames into a panoramic view of scenes along the route. Ranges to feature points are estimated from their image velocities and we integrate the panoramic view with the ranges into a panoramic representation, a 2(1/2)D representation of the scenes along the robot route.
[2 Selection of Landmarks
In order to find suitable landmarks for guiding the navigation, the structures of the route scenes are found by analyzing the panoramic representation. Feature points above the ground are separated from those on or below the ground from 3D information, and then, they are segmented into each object by a simple cluster analysis. By integrating the color patches in the panoramic view with these range patches, we determine each object in the scenes. The results are represented by a structure map which provides the object arrangement and object features. A method for selecting the suitable landmarks is proposed which uses distances of the object from the path and differences of features as a measure. Experiments results show the landmarks are useful to guide the navigation.
[3 Building of Robot and Experimental Results.
A robot with multiple cameras is built to take the panoramic view by one camera while following the road edges by other cameras. Experimental results show the proposed method is robust for guiding the robot from changes in the actual robot path and illumination conditions.
Research Products (25 results)