The method of underwater image enhancement and dense mapping based on ORBSLAM3

To address the challenges of feature point extraction and tracking in underwater visual SLAM caused by uneven lighting, blurriness, and low contrast, we suggest a sophisticated image enhancement technique that integrates the Retinex model with wavelet transform. Additionally, we introduce a dense mapping thread into the ORBSLAM3 algorithm to facilitate the construction of dense maps. Initially, we employ guided filtering to replace the Gaussian filtering in the Retinex algorithm, effectively eliminating blurriness in underwater images. Subsequently, the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm is utilized to enhance the global contrast of the images. Following this, wavelet transform is applied to merge the reflection component map and the enhanced image, thereby suppressing the noise introduced by CLAHE and producing higher quality, clearer underwater images. Finally, a dense point cloud map is constructed based on keyframes and their corresponding poses and depth maps. The proposed image enhancement algorithm is applied to underwater datasets and evaluated within the improved ORB-SLAM3 algorithm. Experimental results demonstrate that the enhanced datasets significantly improve both the detection of feature points and the number of map matching points, thereby substantially enhancing the performance and stability of the visual SLAM algorithm in underwater environments. This approach facilitates the construction of dense underwater maps, thereby expanding the algorithm's application scenarios.