Video Tracking and Auxiliary Obstacle Avoidance of Power Plant Inspection Robot based on Deep Learning and 3D Digital Twin System

The highly intricate landscape of power plant equipment and personnel demands sophisticated solutions, rendering robot-based automation control and the digitalization of production information indispensable for the realization of plant intelligence and smart operational paradigms. Traditional inspection robots, despite their proficiency in gathering information and executing predefined tasks, confront substantial hurdles in terms of adaptability and robustness within the confines of complex environments. The limitations of onboard sensors, such as the narrow fields of view offered by chassis-mounted devices and head-mounted cameras, result in the creation of numerous blind spots that make the robots vulnerable to collisions. Additionally, these robots are frequently perceived as standalone entities, rather than as integral components of the intricate power plant ecosystem. In response to these challenges, this article introduces an innovative control method that leverages the refined Unicorn video tracking algorithm. This algorithm provides real-time tracking, obstacle warnings, and collision pre-alerts for inspection robots in edge cases and unforeseen situations, thereby significantly enhancing their adaptability and preprocessing capabilities in complex environments. Furthermore, the process of collecting and annotating virtual datasets has been greatly simplified through the utilization of a 3D digital twin system, which facilitates the preparation of training datasets with minimal effort.