3-D Object Recognition by Computer Vision for Autonomous Working Robot
Grant-in-Aid for General Scientific Research (C)
|Allocation Type||Single-year Grants|
|Research Institution||Nagoya University|
NOMURA Yoshihiko Nagoya Univ. Fac. of Eng. Asso. Prof., 工学部, 助教授 (00228371)
FUJII Seizo Nagoya Univ. Fac. of Eng. Prof., 工学部, 教授 (20023038)
|Project Period (FY)
1992 – 1993
Completed(Fiscal Year 1993)
|Budget Amount *help
¥2,100,000 (Direct Cost : ¥2,100,000)
Fiscal Year 1993 : ¥400,000 (Direct Cost : ¥400,000)
Fiscal Year 1992 : ¥1,700,000 (Direct Cost : ¥1,700,000)
|Keywords||Object recognition / Model based vision / Vision / Pattern matching / Non-linear optimization / Pose estimation / Potential field / 3-D object / 非線形最小二乗法 / 位置・姿勢推定|
We have studied some unknown parameter estimation techniques, i.e., matching algorithms between a model object and an input one under this grant.
1. Matching based on observed 3-D shape data
We proposed a shape matching algorithm :
Step 1 : Create potential fields for both shape data of model object and input one.
Step 2 : Match two potential fields using optimization techniques such as the non-linear least squares method and dynamic programming. 3-D pose, i.e., the positoin and orientation of the input object is estimated as a result of matching.
Then, we showed its effectiveness by various simulations. The results are summarized as follows.
(1)Fundamental dynamic characteristics of estimation are clarified under various conditions using a simple rectangular parallelepiped object.
(2)A simple articulatcd model object is transformed into a completely differint shape of input object.
2. Matching based on observed 2-D image data
We proposed an iconic matching algorithm :
Step 1 : Create estimated images assuming pose and shape of a model object.
Step 2 : Match the estimated and input images using optimization tecuniques such as the non-linear least squares method and dynamic programming. 3-D pose and shape of the input object is estimated as a resuit of matching.
Furthermore, we found that the model object can be transformed into the greatly changed input image by considering the degree of coincidence of both estimated and input images.
Research Output (18results)