Robust Computation of Trifocal Tensor Based on Hybrid Particle Swarm Optimization

Abstract

In this paper, we present a novel method to calculate trifocal tensor based on hybrid particle swarm optimization. This method takes pole coordinates in three views as particles and the fitness function is to minimize geometric error. The proposed method is evaluated both in synthetic and real data. Experiments show that our method is more robust and accuracy than other typical methods. Rotation matrices and translation vectors estimated by the proposed method have high precision compared with ground truth data.