Lightweight vision architecture deployed in the terminal for safety monitoring and early warning of transmission lines.

Intrusion of hazardous objects into critical transmission lines can directly cause severe accidents, including large-area power outages and even electric shock casualties. However, current monitoring systems remain constrained by single-sensing modalities, such as monocular/binocular vision or Light Detection and Ranging. They fail to achieve reliable real-time three-dimensional perception and lack correlative analysis between external hazard intrusions and the safe clearance distances of transmission lines across complex multi-terrain scene. This study proposes a lightweight vision framework. By fusing pose estimation, visual detection, and depth transformation, this framework achieves high-precision ranging and early warning of external hazards within transmission corridors across all terrains. The architecture employs the lightweight transmission line hazard detection model, by introducing a positive-negative sample dynamic balancing mechanism, thereby improving the detection performance of the model. Furthermore, an improved pose estimation algorithm is proposed to achieve high-precision spatial mapping. Further combined with the depth transformation and point cloud reconstruction, this enables refined ranging for hazards relative to transmission lines under arbitrary terrain conditions. The proposed method has been validated on terminal AI platforms and deployed on on-site camera terminals along transmission lines, showing excellent inference performance and deployment adaptability on terminal devices.