EMMA-SURVEILLANCE: Enhancing Substation Security with AI-Powered Visual Detection

EMMA-SURVEILLANCE: Enhancing Substation Security with AI-Powered Visual Detection

Within the context of the R2D2 project, an innovative tool known as EMMA-SURVEILLANCE is currently in development to bolster the security of critical facilities located in electrical substation transformer centers. This solution integrates an artificial vision algorithm that has been honed by retraining the well-known YOLO (You Only Look Once) model. This algorithm is equipped with the ability to efficiently detect fires, smoke, and the presence of animals in the vicinity of the substation. The purpose of identifying fires and smoke is to promptly alert personnel to potential emergencies, ensuring a swift and effective response. Furthermore, recognizing animals is of utmost importance, as many of them tend to come into contact with the substation structures, posing the risk of electrocution and causing significant disruptions to the electrical system. This model will be deployed in a stationary camera situated within the corresponding pilot substation.

In this initial phase, the algorithm has exhibited remarkable precision, achieving an F1 score of 0.84. This achievement is particularly noteworthy, especially when considering the project’s early stages. As the next steps, the plan involves expanding the dataset, with the objective of collecting more images of fires and smoke to enhance the model’s accuracy. Concurrently, the load testing phase will be initiated, assessing the model’s inference capacity when operating in a real camera and continuously processing real-time video streams. This process is critical to ensure that the algorithm can perform effectively without significant delays, thereby guaranteeing its practicality for real-time monitoring scenarios.

Some examples of recorded images of the AI-powered visual detection.

Further information:

Ugo Stecchi (Project coordinator)

This project has received funding from the  European Union’s Horizon Europe research and innovation programme under grant agreement No 101075714.