A dense stereo matching method based on optimized direction-information images for the real underwater measurement environment

Abstract

The refraction phenomenon of light at the interface of different media causes the epipolar to bend (the traditional camera model is invalid), and the stereo matching can’t be performed under the row matching. Therefore, we proposed a new model based on light field, in which an underwater image is expressed with the direction-information image and position image. The underwater stereo matching is conducted in the direction-information image instead of the original underwater image because the curve constraint of the correspondence becomes the row constraint in the direction-information image. In addition, the essence of stereo matching is similarity detection, but the phenomenon of absorption and scattering in water causes problems such as blurred outlines and loss of a large number of details in real underwater images, which will cause great interference in the similarity detection process. Therefore, we propose an image optimization method that fuses color correction and Dark channel Prior to improve the quality of the directional-information image. Finally, the underwater stereo matching process is divided into three steps. In order to verify the feasibility and effectiveness of the proposed method, we carry out more experiments. The experimental results show that our method and equipment can be used for underwater 3D measurement with high accuracy.