Fusion of Color, Depth, and Thermal Modalities for Rover Navigation in Unstructured Environments

This multimodal dataset consists of labeled images combining color (RGB), depth, and thermal data. The imagery was generated using a RealSense D435i stereo camera for RGB and depth, and an Optris PI-640i thermal camera for thermal imaging. The dataset was created during MaRTA rover traverses at the Bardenas Reales semi-desert, located in Cabanillas, Navarra, Spain. The labeled segmentation masks are single-channel images of the same resolution as the captured data (1280 x 720 pixels), with each class represented by a unique integer label. RGB images are provided in PNG format, while depth and radiometric thermal images are stored in CSV format. The dataset includes eight semantic classes, each assigned a unique label and corresponding RGB color code: Label ID Class Name RGB Color Code 0 Void (0, 0, 0) 1 Compact (234, 182, 118) 2 Grass (91, 159, 74) 3 Bedrock (139, 128, 0) 4 Sandy (192, 64, 0) 5 Pebble (115, 147, 179) 6 Rock (229, 228, 226) 7 Vegetation (116, 216, 16) For additional unlabeled data from the same campaign, please refer to the BASEPROD paper and dataset repository. Version 1.0.1: Fixed zip extraction issues with 'unzip' and the default macOS 'Archive Utility'. The original upload still works using, e.g., 7zip and Windows 11's default file extraction. Dataset contents remain unchanged.