Just as humans use two eyes to obtain 2D and 3D visual information of the scene to move and work freely in the objective world, robots also need RGB 2D color images and 3D point cloud information of the scene and target when carrying out target recognition and fine work. At present, Microsoft Kinect, Intel R200, R461, and other RGB-D cameras on the market can meet the needs of robot applications in the air. Underwater robots also need RGB-D vision sensors to complete underwater target recognition and fine operation tasks, but so far there are no reports of commercial products of underwater RGB-D cameras. At present, the 3D measurement technology based on binocular stereo vision technology in the air has been relatively mature. However, the particularity of the underwater environment brings difficulties to the development of underwater RGB-D cameras. Firstly, due to the absorption and scattering of water medium and suspended particles in water, the captured underwater image has more serious problems than the image in the air, such as color distortion, low contrast, fuzzy details, which affects the matching accuracy of corresponding points in the stereo matching process and the authenticity of RGB image color. Therefore, for underwater RGB-D camera, underwater image enhancement processing is particularly important. The image enhancement technology studied by many scholars provides high-quality image guarantee for underwater stereo vision. In addition, when the industrial camera is applied to the underwater scene, the camera must be encapsulated in a waterproof shell. The propagation of light needs to pass through three media of "water-glass-air" in turn, and finally image in the camera image plane. However, due to their different refractive indices, light is refracted at the interface of different media, so that the object-image relationship of underwater camera imaging does not meet the imaging principle in air. Therefore, the traditional camera model is invalid. Researchers around the world have proposed that building an underwater refraction model is the fundamental solution to the problem of underwater 3D reconstruction. Although the research on underwater RGB-D technology is rare, the above research has paved the way for underwater RGB-D camera. In order to meet the needs of underwater target recognition and fine operation of an underwater robot, an RGB-D underwater camera engineering prototype based on binocular stereo vision integrating 3D point cloud and 2D RGB image are developed, and the related technologies are studied. It mainly includes: Firstly, considering the underwater multi-layer refraction effect, the underwater camera imaging and binocular stereo imaging models are established to convert the underwater image into the image in the air to eliminate the influence of refraction on the reconstruction accuracy. Then the corresponding points are matched to obtain the 3D point cloud; Secondly, according to the principle of underwater image imaging and Retinex theory, a color correction algorithm based on the attenuation correction coefficient of the underwater image is proposed to correct the color of the underwater image, and then the underwater uniform light component estimation method is improved based on the dark primary color prior theory to enhance the definition of the underwater image. Several common image enhancement algorithms are selected as a comparison, and the information entropy, the number of SIFT feature points, and canny feature points are used as evaluation indexes to evaluate the effect of different algorithms; Finally, the RGB values of the three channels of the matching points are extracted, and the standard RGB values are calculated by using the RGB information of the left and right matching points. The alignment superposition model is established, and the 3D point cloud and 2D color data are superimposed and fused to obtain underwater RGB-D image data. The experimental results show that the proposed underwater image enhancement algorithm can not only correct the color distortion of underwater image, but also improve the effect of underwater 3D reconstruction, which lays a foundation for obtaining higher quality RGB-D data; the established point cloud and RGB color registration model can register the color corrected color data with 3D point cloud data to generate high-quality underwater RGB-D data; the underwater RGB-D engineering prototype has good 3D measurement accuracy, and the system reconstruction error at 3M is 2.6 mm, which has a good actual underwater 3D reconstruction effect.