Refraction Correction Algorithm for Visual Slam Based on Rgb-D Camera

The visual SLAM algorithm is widely adopted in the air because of its superiority in simultaneous localization and mapping. However, applying the visual SLAM algorithm to environments in other media can lead to refractive errors in the generated images caused by light passing through different media, resulting in algorithm failure. To enhance the environmental adaptability of the visual SLAM algorithm and achieve simultaneous localization and mapping of specialized robots in special media, we construct a visual SLAM experimental device using a Raspberry Pi and an RGB-D camera. Specifically, we propose a refraction correction algorithm based on an RGBD camera, obtaining equivalent pixel images of distorted pixels in the air. We integrate the refraction correction algorithm into the front end of the previously proposed point-line visual SLAM algorithm. Through validation in a real experimentalenvironment, the refraction correction algorithm successfully draws semi-dense pointcloud maps underwater. Compared with not using the refraction correction model, our proposed algorithm better restores the structural information of the underwater environment and objects in the water.