Virtual reality (VR) simulations involving highly detailed, realistic 3D models of underground mines offer significant cost and safety benefits in many application areas, including equipment prototyping, operator training, rescue simulations, and education. However, the generation of high-detail models of real mines is a time-consuming task. Conventional approaches involving total stations and/or laser scanners require extensive manual post-processing while offering only limited texture detail. In contrast, an approach for mine virtualization based on video data is proposed. A mobile robot carries a sensor bar, composed of a stereo camera system and lighting unit. While traveling the mine, the robot captures stereo images several times per second. During post-processing, algorithms from the field of computer vision are used to extract a photorealistic 3D model of the mine. Experiments have been conducted at the local Research and Educational Mine ‘Reiche Zeche’ in Freiberg, Germany. The virtual mine model can be easily imported into common 3D software tools, such as CAD systems, air flow simulations, games engines, and immersive VR applications.