EEIE-YOLO: Enhanced Edge Information Extraction YOLO for Electric Bicycle Detection in Elevator

In response to the challenges of low detection accuracy caused by obstructed and overlapping targets in elevator detection of electric vehicles, recent research efforts have proposed various methods to address these issues. Examples include incorporating Transformer-Based VIT attention mechanisms or altering the structure of feature fusion pyramids to enhance network performance. However, these improvements often come with a significant increase in computational costs, thereby reducing the feasibility of deploying real-time object detection algorithms in subsequent stages. Therefore, this paper presents an approach that tackles the aforementioned challenges with an improved YOLOv8 model, which integrates Efficient Multiscale Attention module (EMA)and Involution block indicates to enhance the networks’ ability to extract objects’ edge information , termed EEIE-YOLOv8. Validation on a custom-built dataset of elevator electric vehicle scenarios demonstrates that the EEIE-YOLOv8 algorithm achieves an average precision improvement of 1.1% and a 7.3% increase in precision compared to the original YOLOv8 algorithm. These experimental results affirm that the proposed enhancements effectively meet the detection requirements for electric vehicles, bicycles, and passengers inside elevators under real-world conditions.