Multi-camera rectification using linearized trifocal tensor

Multi-camera systems such as linear camera arrays are commonly used to capture content for multi-baseline stereo estimation, view generation for auto-stereoscopic displays, or similar tasks. However, even after a careful mechanical alignment, residual vertical disparities and horizontal disparity offsets impair further processing steps. In consequence, the multicamera content needs to be rectified on a common baseline. The trifocal tensor represents the geometry between three cameras and hence is a helpful tool to calibrate a multi-camera system, and to derive rectifying homographies. Against this background we propose a new method for a robust estimation of the trifocal tensor specialized for linear camera arrays and subsequent rectifying homography computation based on feature point triplets