A proposal for monocular depth estimation in non-ideal viewing conditions for point cloud extraction

LiDAR sensors, while highly effective, have a significantly higher cost compared to cameras and also suffer from distortions under non-ideal visibility conditions such as fog and rain. In recent years, image and video processing has made significant advances in reducing non-ideal visibility conditions and extracting monocular depth information thanks to artificial intelligence, which can be used for point cloud extraction. In this study, an environment was developed for identifying and processing environmental conditions of fog and rain in images, offering a more cost-effective and accurate alternative to LiDAR sensors. This approach improves distortion reduction in images, allowing for more accurate depth maps under these adverse conditions. The implementation focused on algorithms for image processing in fog and rain. Although the algorithms require further improvement to reduce processing errors, this initial approach demonstrates significant potential for developing a depth map extraction system under non-ideal visibility conditions. The key feature of this method is that it allows for background estimation and subsequent point cloud extraction in a much more economical way than with a LiDAR sensor.