1986 Fiscal Year Final Research Report Summary
Modeling of Environment by a Mobile Robot Utilizing Time-Varying Stereo Imagery
| Project/Area Number |
60460134
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
計算機工学
|
|---|
| Research Institution | Osaka University |
|---|
Principal Investigator |
TSUJI Saburo Faculty of Engineering Science, Osaka University, 基礎工学部, 教授 (60029527)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
MINORU Asada Faculty of Engineering Science, Osaka University, 基礎工学部, 助手 (60151031)
ABE Norihiro Faculty of Engineering Science, Osaka University, 基礎工学部, 助手 (00029571)
YACHIDA Masahiko Faculty of Engineering Science, Osaka University, 基礎工学部, 助教授 (20029531)
|
|---|
| Project Period (FY) |
1985 – 1986
|
| Keywords | Mobile Robot / Computer Vision / Stereo / Time Varying Imagery / センサ融合 / 世界モデル |
|---|
| Research Abstract |
Conventional vision-guided mobile robots need to stop frequently to take images. The objective of this research is to establish a method for building the world model from the imagery taken while the robot is moving. A small mobile robot HEROl on which two TV cameras are mounted is used for the experiments. The images of the cameras are converted into digital picutres and analyzed in real time by an image processor TOSPIX-2.
1. Analysis of velocity field using stereo disparity
Analysis of time-varying stereo imagery needs to establish not only correspondences between stereo pairs but also those between frames. Rather than to analyze each frame independently, we find the correspondences using prediction of the disparities in the next stereo pair from the 3-D model obtained already. Since the location and orientation of the robot estimated from its internal sensors are inaccurate, a hill climbing method is applied to estimate more accurate camera parameters. The experimetal results show the method is useful for modelling the world.
2. Finding of moving objects and building of the world model
A method has been developed for finding moving objects in the pathway from images taken by moving cameras by analyzing velocity field in the imagery and applying a clustering technique.
A new stereo matcher is proposed which can find patterns on the floor by utilizing constraints valid in man-made environments. Using the results, a world model is built for planning collision free paths. Experiments of real-time-vision guided navigation to sub-goals are successful.
In conclusion, the basis of stereo-vision guided mobile robots are established.
|
