LunarSafe AI

LunarSafe AI is an explainable computer-vision decision-support prototype for analyzing lunar terrain imagery and identifying candidate landing zones. A user uploads a lunar surface image and selects mission parameters such as spacecraft size and safety margin. The image is then processed by a FastAPI backend using Python and OpenCV. The analysis combines several classical computer-vision signals, including Sobel gradients, local texture variation, deep-shadow detection, and conservative detection of compact small-scale terrain features. These signals are used to generate a potential-hazard mask. LunarSafe AI then expands detected hazards by the selected safety margin and applies an OpenCV distance transform to measure image-space clearance from unsafe regions. Using this information, the system identifies three distinct landing candidates: Site A, Site B, and Site C. Each candidate receives a relative suitability score based primarily on clearance from detected hazards and local terrain safety. The interface also provides LOW, MODERATE, or HIGH risk classifications. The frontend provides four explainable visualization modes: the original image, Hazard Map, Distance Map, and annotated Landing Zones. Users can also download the final annotated result as an image and export the analysis data as JSON. The frontend was created with Native.builder using React and TypeScript, while the backend uses FastAPI and OpenCV and is deployed on Render. LunarSafe AI is an experimental prototype and not a flight-certified spacecraft guidance system. It does not claim scientifically precise crater, obstacle, slope, or physical-distance classification. Its goal is to demonstrate how interpretable computer vision can transform lunar imagery into an understandable landing-zone analysis workflow.