| Budget Amount *help |
¥3,200,000 (Direct Cost: ¥3,200,000)
Fiscal Year 2006: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 2005: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 2004: ¥1,800,000 (Direct Cost: ¥1,800,000)
|
|---|
| Research Abstract |
This research proposes a new method in order to resolve a stereo matching problem with high-precision and high-speed, and applies to automatic production of rough 3 D city area map. The fruitful results are summarized as follows.
1) New stereo matching method
A main issue of stereo matching problem is to obtain correspondence of characteristic points in stereo images including occlusion and reversal position with high-precision and high-speed. We propose a new stereo matching algorithm, which has the performance of high-precision and high-speed, based synergetics, which is a mathematical model for an autonomous complex system.
2) Automatic production of rough 3 D city area map employing binocular stereo
Digital map has been developed from 2 dimensions to 3 dimensions with requirement for advanced navigation. The main issue of the production of 3-dimensional digital map is to reduce its high production cost. Then, we investigated the automatic production method of rough 3D city area map tha
… More
t employs binocular stereo images for city area. The method is constructed by the followings: (1)region segmentation of subjects, building, street, sky, road side tree etc., from a pair of stereo images, (2)obtaining 3 dimensional coordinates of each subject by stereo matching, (3)identifying a polygon model for each subject, and (4)pasting texture for each subject. Here, the useful way, which is not complete yet, for steps 1),2), and 3) is proposed.
3) Object recognition from binocular stereo
The automatic production of 3 D city area map needs to estimate the shapes of subjects from their 3 dimensional coordinates. This is naturally extended to recognize objects from binocular stereo images. Then we propose a new method, which consists of K-means clustering and fuzzy c-varieties with noise clustering for 3 dimensional coordinates including much noise, to recognize the number of subjects and the shapes of subjects. Here, the proposed method effectively works under the condition that the number of subjects is less than 4 and the subjects do not contact with each other. Less
|
