Line Based Trinocular Stereo

An approach to solving the stereo correspondence problem in trinocular stereo vision is described. It is based on geometric matching constraints relating the orientation of lines extracted in three images taken from different viewpoints. These novel constraints are termed unary orientation and binary orientation constraints. Matching is achieved within an optimisation framework in which the constraints are encoded into a cost function that is optimised using the simulated annealing method. Results are demonstrated and the characteristics of the approach are explored on both synthetic and real trinocular images.