Next: 3D–3D Odometry Up: Multi-Camera Vision Previous: Corresponding points in alternative
Visual Odometry aims to estimate the relative motion of a camera, or of a stereo pair, by analyzing a sequence of images.
In the case of a single camera, the visual-odometry problem can be addressed by computing the Essential Matrix and subsequently decomposing it. In this case, as noted previously, the scale of the motion cannot be determined from monocular observations alone, but the different relative motions can be related to one another. In the case of a stereo pair, knowledge of the calibration and baseline also makes it possible to determine the scale of the motion.
There are two main approaches to visual odometry. The first, referred to as 3D-to-3D, estimates motion from pairs of three-dimensional points reconstructed at successive instants. The second, referred to as 3D-to-2D, assumes that the previously observed three-dimensional points are known and estimates the current pose by minimizing the corresponding reprojection error.