Panoramic Stereovision Sensor

Abstract

Panoramic sensors with sequential acquisition are mainly based on rotating cameras, several approaches can be found in [311, 234, 192, 136, 22, 24]. Before we conceive any panoramic sensor we have to think of a geometry that will decide of all the treatments that will be applied in the image side. The architecture of the panoramic stereovision sensor has been developed to symplifiy to the maximum the calibration stage that is 2D instead of the usual 3D and ease the features matching with an implicit epipolar geometry that becomes 1D instead of the usual 2D. When we talk about panoramic stereovision we mean observing and reconstructing an unknown scene all around the sensor. The panoramic image that describes this observation is the set of all equidistant points the optical center. The projection is then central, it can be cylindrical or spherical. In what follows the retained solution is the cylindrical one. The equivalent sensor can be looked at as a cylinder on which will be projected points of the 3D scene, as shown by Figure 9.1. Once we have defined the elementary panoramic sensor, we can establish the architecture of the panoramic stereovision sensor. The omnidirectional stereovision sensors have not been developed as a unique entity. The main drawback of these sensors is that usually they are an empirical assembly of standard elements putting all the difficulties on computation. The architecture of the system built is based on two linear CCD cameras fixed vertically one on the top of the other. They are specially designed as a part of a whole mechanical entity and accurately mounted parallel. They are rotated around a vertical axis, the distance between the two cameras is adjustable between 10 cm and 40 cm, with a 5 cm step. The linear sensors we use are Thomson 1024 pixels TH7804A, and the lenses are of 12.5 mm focal length. The motion of the system is conducted by a stepper motor and the whole system is controlled by computer (Figure 9.2).