Real-Time Object Detection for Forklift Automated Guided Vehicles Using Deep Learning

Automated Guided Vehicles (AGVs) are increasingly used in industrial and logistics operations for material handling, offering benefits such as reduced human error, improved efficiency, and lower operational costs. This study presents the design and implementation of a real-time intelligent management system for Forklift AGVs based on deep learning techniques. The core of the system is an optimized version of YOLOv3, termed YOLOX, enhanced with Adaptive Spatial Feature Fusion (ASFF) and advanced data augmentation strategies. The ASFF module employs spatially adaptive weights (α, β, γ) to dynamically integrate multi-scale features across the Feature Pyramid Network, improving the detection of small, occluded, and overlapping objects. The system is trained on a combined Pascal VOC dataset using mix-up and label smoothing to enhance generalization and model robustness. It is deployed on embedded hardware, including Raspberry Pi 4, enabling real-time processing of visual data and sensor inputs under various lighting and environmental conditions. Evaluation results indicate that the model achieves a high mean Average Precision (mAP) of 94.17%, with real-time confidence scores reaching 98.1% in natural lighting and 94.3% in dim conditions. The system effectively detects and classifies a wide range of objects—including static, dynamic, small, distant, and partially occluded—in complex scenes. The proposed solution demonstrates robust real-time performance and adaptability, making it suitable for deployment in resource-constrained environments. It offers a scalable and intelligent framework for autonomous AGV navigation, contributing to safer and more efficient material transportation in real-world applications.