Planetary Landing Site Selection Using Multi-Modal Information Fusion

This paper introduces an autonomous landing site selection methodology for planetary exploration through multimodal information fusion. To address the limitations of single-modal perception in complex extraterrestrial environments, a cross-modal fusion framework is developed that synergistically analyses depth and texture features. Initially, the least median squares (LMedSq) estimation method is employed at the technical implementation level to solve the terrain plane equation and quantify surface roughness. Next, a multi-scale texture feature description model is constructed by integrating an improved Bandeira edge detection operator. Subsequently, two modalities are fused using logical operations, and a sliding window filtering algorithm based on morphological constraints is designed to identify hazardous areas and integrate multi-modal features. Experimental results demonstrate that, compared to traditional single-modal recognition schemes, the proposed multi-modal information fusion approach can identify suitable landing sites more safely and effectively.