Research on an Image Recognition Automatic Counting System Based on Improved YOLOv8

• Published in: IEEE International Conference on Power, Intelligent Computing and Systems (Online) pp. 1417 - 1421
• Main Authors: Chen, Jiaqi, Zhao, Jiaxi, Zhang, Deyong, Ye, Zhimin, Liu, Jianyou, Huang, Wen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.07.2024
• Subjects: Accuracy; Automatic counting; Industrial automation; Logistics; Object detection; Production; Real-time systems; Robustness; Target detection; Target tracking; Testing; Training; Warehousing
• ISSN: 2834-8567
• DOI: 10.1109/ICPICS62053.2024.10795929

This paper proposes a deep learning-based automatic target counting system for industrial environments. The system features enhanced YOLOv8 for small target detection using attention mechanisms and optimized training strategies. It also customizes the DeepSort algorithm for improved robustness in industrial multi-object tracking. Utilizing GPU acceleration, the system achieves real-time response within 0.2 seconds. Testing with simulated and real data shows a detection mAP of 88.7%, a tracking MOTA of 91.5%, and a counting accuracy of 96.8%. This system meets industrial requirements for automation, real-time performance, and reliability, offering promising application prospects.