部分特徴を有する3次元モデルを用いた実画像からの物体の検出と位置・姿勢推定

抄録

In this paper, we propose a new method to detect object and estimate the position and the orientation from an image using a three-dementional (3-D) model having feature points. In the proposed system, a 3-D model is employed in order to represent a 3-D object, and the system can generate the 2-D models from various viewpoints. Then, the system compares the model that projected onto 2-D with the object in the input image. In the comparison, the system gives attention to some partial feature points of the object, and computes the fitness value of the 2-D model in the input image. The system can estimate the position and the orientation of the object by maximizing the fitness of the 2-D model. In the proposed system, a genetic algorithm (GA) is employed for this optimization of the fitness. In the experiment, we took up detection and position and orientation estimation of a car. We have tested various kinds of images and obtained excellent results.