Underwater 3D Reconstruction Based On Binocular Stereo Vision

Underwater 3D reconstruction is widely used in the area of underwater AUV (Autonomous Underwater Vehicle) navigation, marine exploration, marine environmental monitoring as well as underwater target recognition, which has important academic meaning and application value. This paper aims at extension of land-based stereo vision to the underwater environment, developing fast algorithm for underwater 3D reconstruction, and providing assistance for AUV navigation.In this paper, due to the effect of light absorption and scattering in water, more attention is paid to image restoration while image preprocessing is studied. Inverse filter and Wiener filter are discussed respectively, and four different degradation models are selected to represent actual degradation. For underwater refraction, we presented a new underwater calibration method, that is converting underwater image to air image first and then calibration.To meet the requirement of real-time, robustness and accuracy of the underwater 3D reconstruction approach, local matching, semi-global matching and feature based Bayesian estimation matching algorithms are studied respectively. Experimental results show that Bayesian approach has the best performance, while Sobel-based Bayesian estimation matching algorithm has the minimum computational complexity. Multi-resolution matching method can be used to further reduce matching time and improve the algorithm in real time.