2000 Fiscal Year Final Research Report Summary
Pose Estimation of Three Dimensional Objects Using Genetic Algorithms
| Project/Area Number |
11680376
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | Utsunomiya University |
|---|
Principal Investigator |
SHOJI Kenji Faculty of Engineering, Associate Professor, 工学部, 助教授 (70143188)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
TOYAMA Fubito Faculty of Engineering, Utsunomiya University Assistant Professor, 工学部, 助手 (60323317)
|
|---|
| Project Period (FY) |
1999 – 2000
|
| Keywords | pose estimation / rigid body / articulated object / edge image / silhouette / genetic algorithm |
|---|
| Research Abstract |
We treat the problem to estimate the pose of model objects from its single images as visual input.
The objects include rigid and articulated ones. The most previous approaches detect the geometric characters(points, vertices, etc.) from the input images, and then match the model ones with them.
These methods, however, would be trustless in the case where the images contain noise. In order to improve the robustness against the noise, we use an image matching method instead of the feature matching ones. The image matching is carried out by fitting the projection of a model object to the edge or silhouette image aS visual input. The best fit pose is the solution of the problem.
The pose estimation of rigid objects is to search the maximal value of the evaluation function which calculates the overlapping degrees between the projection of a model object and the input image in the 6-dimensional parameter space, We use genetic algorithms, GA in short, which are said to be excellent in the global searching ability. The purpose of this work is to improve the efficiency of searching an optimal pose by considerations of GA.The main result of this work is as follows.
1. In the pose estimation of rigid objects, we propose the use of the similarity ordered pose tablet which is made in advance from the similarities between the model views of every directions. The proposed algorithm refers the table for searching more fit pose efficiently. The simulation experiments show the effectiveness of the table.
2. In the pose estimation of articulated objects, we propose the parameter space reduction method using morphological operation erosion. The method puts the model into the input silhouette in pose generation and crossover of GA.The simulation experiments for two-dimensional articulated objects show the effectiveness of the method.
|
Research Products
(2 results)
