Abstract: With the rapid development of intelligent mining technologies, computer vision has become an essential technology for operational monitoring, personnel behavior recognition, equipment condition assessment, and intelligent control of mining machinery in underground coal mines. However, the complex underground mining environment, characterized by low illumination, dust and water mist interference, confined spaces, and equipment occlusion, poses significant challenges to the accuracy and robustness of computer vision systems. This study summarizes the major challenges faced by computer vision technology in complex underground mining environments, including environmental challenges, data challenges, algorithm and model challenges, as well as system and application challenges. The current applications of computer vision in underground complex environments are systematically reviewed, covering image and video enhancement, image stitching and panoramic video generation, target detection and intelligent recognition of key objects (e.g., conveyors, personnel, and coal-gangue materials), as well as intelligent perception and control of underground mining equipment. To promote the further development of computer vision technology under complex mining conditions, this study proposes several key research directions, including multimodal perception and fusion, lightweight systems and edge computing, the development of domain-specific models incorporating coal mining knowledge, and the establishment of system-level safety frameworks and standardization protocols.