Video Stream Processing for an Autonomous Tunnel Drainage Rover

rainage pipes in tunnels are very complicated to service but still need rather frequent inspections in order to detect the initiation of deposited scales stemming from carbonate that tends to precipitate when the pH-value of the mountain water rises with the release of gaseous carbon dioxide CO2due to a decreasing water pressure. Drainages allow to design tunnels without resistance to ground water pressure. We present a camera-based optical sensor, developed for an autonomous operating tunnel drainage rover, that is able to detect and quantify deposited calcite constricting more or less severely the free cross section the drainage rover needs to safely navigate the pipe. The pipe's cross section is imaged via an area-scan camera at a frame rate adopted to the rover's speed so that at least every 50 mm of movement a frame is acquired. The presented image processing software is segmenting and classifying each frame into the pipe's wall, the pipe's free lumen, eventually existing water, and calcite