Underwater Range-Gated Binocular Vision 3D Imaging based on RG-DeNoiseNet

This dataset is derived from the experimental research on the integrated underwater 3D imaging scheme combining range-gated imaging (RGI) technology, binocular vision, and deep learning denoising. It aims to provide reliable image data support for the research and application of medium-to-long-range underwater high-precision 3D imaging technology.The dataset mainly includes four categories of images, covering key links of the underwater range-gated binocular vision imaging system. The first category is the original left-right sub-field images captured by the hardware system, which is composed of a single gated-ICCD camera and a four-mirror optical beam splitter. These images are obtained under different water turbidity conditions (with the one-way attenuation length of turbid water reaching 7AL’s) and different target reflectivity levels, recording the real underwater imaging characteristics affected by backscattering and absorption. The second category is the synthetic training image pairs constructed for solving the scarcity of underwater paired data. The pairs are generated by fusing clear target images collected in clear water (as ground truth) and pure noise images acquired under different water attenuation conditions, which are dedicated to the training of the RG-DeNoiseNet deep denoising network. The third category is the denoised images processed by the RG-DeNoiseNet network. Compared with the original images, these denoised images show significantly reduced backscattering noise, improved signal-to-noise ratio and feature clarity, with the average PSNR increased by more than 16 dB and SSIM increased by about 0.45. The fourth category includes the disparity maps and depth maps calculated by the optimized semi-global matching (SGBM) algorithm based on the denoised images, as well as the valid disparity point statistics corresponding to different experimental conditions. Among them, the valid disparity points of the depth maps in turbid water are doubled compared with those without denoising, and the depth calculation accuracy reaches the centimeter level.In addition, the dataset also contains auxiliary images such as calibration images of the binocular vision system (20 checkerboard calibration board images with different poses and distances) and their calibration result records, gradient maps of synthetic and real images, and grayscale histogram comparison charts. These auxiliary images provide verification basis for the reliability of the hardware system calibration and the effectiveness of the synthetic image data.This dataset is applicable to the research fields such as underwater image denoising algorithm optimization, binocular vision stereo matching technology improvement, and performance verification of integrated underwater 3D imaging systems. It can help researchers deeply understand the imaging mechanism of underwater range-gated binocular vision systems, promote the development of related algorithms and technologies, and provide data support for the practical application of underwater autonomous detection and dynamic target tracking.