Posicionamento de múltiplos objetos a partir de visão estéreo

Abstract

The following study is about the estimation of three-dimensional position of multiple objects captured in images in a stereo vision setting. The mathematical formalism begins with the representation of motion in Euclidean space. After that, the projection of the images is performed in an ideal camera perspective using homogeneous coordinates. This results in the intrinsic and extrinsic parameters, which are part of the camera calibration. From that, stereo vision is used to obtain three-dimensional position of objects that are captured in the images from two cameras. Thus, depths of various objects are obtained to reconstruct their positions. The validation of the proposed algorithms are made through an experiment built for this purpose. This experiment uses two cameras with semi-parallel views to the center. In the center of the environment there is a gimbal platform, which performs a uniform motion. Above, there is a plan with three white dots, simulating the objects position. This motion captures two images, which are processed by the proposed algorithm. The validation consists in making a comparison between the orientation planes of the gimbal servomotors and reconstruction objects.