Soil erosion represents a major global threat, necessitating a detailed understanding of its spatial and temporal dynamics. Advanced geospatial technologies such as time-lapse structure-from-motion (SfM) photogrammetry provide high-resolution monitoring of surface changes. This study presents a novel event-driven approach for near-continuous monitoring of hillslope surface dynamics over a multi-annual period. The system employed synchronized digital single-lens reflex cameras at three slope stations, triggered by a rain gauge and a daily timer. Ground control points (GCPs) were surveyed with millimeter accuracy to ensure precise georeferencing. An automated Python-based workflow was developed to synchronize images, detect GCPs using a convolutional neural network, generate daily digital 3D surface models via SfM, and compute 3D surface models of difference. The absolute accuracy of SfM point clouds ranged between 8 and 12 mm on average, primarily due to registration errors, with lower deviations (<5 mm) in central areas after height adjustment. Relative accuracy decreased concentrically with distance from the cameras, with level of detection values between 5 and 25 mm depending on distance and location. Time series analysis revealed surface changes driven by rainfall, snowmelt, and agricultural activity. The most significant changes often occurred shortly after tillage, even with minimal rainfall, indicating both erosional and non-erosional processes. A strong negative correlation between rainfall and elevation loss was especially evident within the first seven days following tillage. Seasonal surface lowering of 3-5 cm during winter and occasional positive changes due to frost or vegetation growth were also observed. The monitoring system and workflow are transferable, and the resulting high-resolution datasets are expected to be valuable for analyzing erosion dynamics and testing process-oriented soil erosion models.