Abstract:
Escalators are essential components of modern infrastructure, facilitating vertical transportation
in public spaces. However, escalator-related accidents pose a persistent safety challenge. This
project introduces an escalator safety system that utilizes computer vision and deep learning to
enhance hazard detection and response mechanisms.
The proposed system, built on an ESP32 and Raspberry Pi, employs a pre-trained deep learning
action recognition mode, namely the LRCN to analyze video footage for identifying risks, such
as people falling on the escalator or becoming jammed at key points.
The results showed that the system has high accuracy of nearly 94.99% . The results also
showed that the LRCN model required high computational time. Although the simulation on a
Personal computer was done in real time, we did not reach a real time response on the Raspberry
Pi microcontroller