SMA-YOLO: A Novel Approach to Real-Time Vehicle Detection on Edge Devices.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent traffic management and autonomous driving rely heavily on accurate vehicle detection.
• Current deep learning models struggle with speed, computational cost, and accuracy in complex environments, hindering edge deployment.

Purpose of the Study:

• To develop a lightweight and efficient vehicle detection model for real-time applications on edge devices.
• To improve the detection speed and accuracy of existing YOLOv7 models.

Main Methods:

• Proposed SMA-YOLO, integrating MobileNetV3 backbone, SimAM attention, ACON activation, and SIoU loss with YOLOv7.
• Implemented model modifications to reduce computational load and enhance feature extraction.

Main Results:

• SMA-YOLO achieved significant reductions in model size, computational requirements, and parameters.
• Demonstrated improved detection speed and maintained high accuracy on the UA-DETRAC dataset.

Conclusions:

• SMA-YOLO offers a feasible solution for deploying efficient vehicle detection on embedded systems.
• The proposed model balances performance and resource efficiency for real-time intelligent transportation systems.