Escalator monitoring and early warning method and system combining event driving and visual large model
By combining event-driven and visual big data model-based escalator monitoring and early warning methods, dynamically adjusting the image acquisition frequency and constructing an edge inference cluster, the problems of resource consumption and recognition robustness are solved, and efficient risk identification for high-altitude operations is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA DATANG GRP DIGITAL TECH CO LTD
- Filing Date
- 2026-01-12
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies suffer from huge resource consumption and rigid edge computing scheduling mechanisms due to the use of full-time high-frequency continuous analysis of video streams. This results in insufficient understanding of key actions such as climbing and escalator movements, and the multimodal fusion method is fragile, making it impossible to achieve accurate risk identification of escalator behavior during high-altitude operations.
An escalator monitoring and early warning method combining event-driven and visual big data models is proposed. By using low-frequency frame acquisition, edge inference clusters, and visual multimodal big data models, the image acquisition frequency is dynamically adjusted. A lightweight visual model instance resource pool and multi-core load balancing scheduling are constructed to perform spatial topology and behavior analysis, and generate escalator compliance judgments and risk warnings.
Achieve low resource consumption and high-precision visual semantic understanding in large-scale, multi-channel video surveillance scenarios, improve the accuracy of action recognition in complex scenes, and enhance the robustness of risk monitoring for high-altitude operations in industrial environments.
Smart Images

Figure CN122157139A_ABST