High-locality and hidden landslides, due to its significant characteristics of being difficult to access, identify, and monitor, have strong suddenness and destructiveness when they occur. Continuous monitoring and risk assessment of these landslides are of great significance. Traditional artificial ground survey methods and ground monitoring equipment have the characteristics of high risk, low efficiency, easy damage to equipment, and frequent offline false alarms. Thus, based on unmanned aerial vehicle (UAV) tilt photogrammetry, this study attempts to provide a digital twin method to characterize high-locality and hidden landslides by monitoring and analyzing the deformation and spatiotemporal evolution of geological disasters. This study uses UAV tilt photogrammetry technology to obtain 10 periods of aerial survey data of the Baige landslide on the Jinsha River in Tibet as the research area from April 2019 to September 2021. A multi-temporal digital twin landslide body is constructed, and high-precision quantitative monitoring of multi-dimensional factors, such as the overall sliding characteristics, local micro deformation, and collapse volume of the Baige landslide, is achieved, which are applied to the monitoring and warning of Baige landslide. The results show that there are signs of continuous deformation in the Baige landslide during the monitoring period from 2019 to 2021, and strong deformation mainly occurs at both sides and rear edges of the landslide, gradually expanding, and posing a risk of collapse and river blockage. The multi-temporal digital twin method and application of landslide deformation monitoring on qualitative and quantitative characteristics description and risk assessment of geological disasters are further analyzed. The method in this study has the advantages of fast and flexible, comprehensive coverage, and not limited by complex and dangerous terrain conditions, which could provide information for the large gradient deformation monitoring and engineering practice of slope disasters, such as high-locality and hidden landslides.