Block Fall Detection Algorithm for Water Conveyance Tunnels Based on Sonar Point Cloud Data

The identification of fallen blocks in water transport tunnels is crucial for ensuring long-term operational stability. This paper presents a sonar point cloud-based algorithm for detecting fallen blocks, which uses ring-scanning sonar to collect 3D point cloud data. After downsampling and geometry-based denoising, fallen blocks are separated from the tunnel bottom via sample consistency estimation based on elevation differences. An improved Alpha Shape algorithm optimizes the rolling circle radius using a genetic algorithm's natural selection mechanism, enabling accurate boundary extraction and complete point cloud model reconstruction. Experimental results show 100% detection rate for large blocks, 92% for medium-sized, and 88% for small ones, with no false positives. The method precisely identifies tunnel bottom block boundaries and captures their spatial distribution and morphological features, providing data support for tunnel safety inspections.