An Intelligent Elevator Assistance System Based on Res-YOLOv5

To address the various issues related to elevator safety, convenience, and intelligence in many residential communities, we propose an intelligent elevator assistance system based on the Res-YOLOv5 object detection algorithm. This system can recognize the facial and eye features of passengers in different types of elevators, extract the identified eye images, estimate the direction of the passenger's gaze, and, combined with machine vision algorithms, estimate the gaze point. This enables the system to automatically and intelligently perceive the floor that the passenger intends to go to. Experimental results show that the YOLOv5 object detection network model, trained for this task, achieves 99.44 % detection accuracy, 99.54 % recall rate, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$99.48 \% \text{mAP} {@} 0.5$</tex>, and 68.95 % [email protected]:0.95 for recognizing the eye features of elevator passengers. The model can effectively identify the eye features of different passengers entering the elevator, extract the eye region, estimate the gaze direction, and predict the gaze point. Verification in real elevator scenarios demonstrates that the intelligent elevator assistance system can quickly and accurately detect the eyes of passengers entering the elevator from different angles and distances, and predict the gaze point. The average error in gaze point prediction remains within 1.3 cm, proving that the proposed intelligent elevator assistance system is practical and reliable.