Semantic Segmentation in Obstacle Detection for Rover Vehicles

This paper studies semantic segmentation methods and their application in the problem of obstacle detection during rover navigation. With the advancement of space exploration and the increased use of autonomous vehicles to explore remote areas, accurate obstacle detection is critical in ensuring safe and effective missions. As part of the study, we analyze various semantic segmentation methods, including the use of convolutional neural networks, sensor systems, and other image processing techniques. Particular attention is paid to the implementation of the semantic segmentation method based on the adapted UNet architecture and transfer learning. This approach is especially valuable in the context of limited computing resources of the rover and limited availability of training data from extraterrestrial surfaces. Our experiments using the open-source Artificial Lunar Landscape Dataset demonstrate that this approach can achieve an MSE of 0.0577, indicating the effectiveness and potential of the proposed method for future space missions.