Edge-Aware Multi-Sensor YOLOV12: Lightweight Real-Time Object Detection for Smart Transportation

Real-time object detection in IoT-based intelligent transportation systems presents challenges of multi-sensor integration, limited edge computational resources, and constrained communication bandwidth. This paper proposes a lightweight multi-modal object detection framework based on YOLOV12-Tiny, designed for edge–cloud cooperative deployment. A mid-level feature fusion module combines heterogeneous sensor inputs—including RGB, infrared, and radar—using channel attention and lightweight convolution, improving robustness under adverse environmental conditions. The YOLOV12 backbone is optimized through structured pruning, knowledge distillation, and INT8 quantization-aware training. Experiments on nuScenes, FLIR Thermal, and KAIST Multispectral datasets demonstrate that the proposed method achieves detection accuracy comparable to full-scale YOLOV12 while reducing model size, latency, and energy consumption. An efficient edge–cloud communication strategy transmits only detection metadata or selective keyframes, further minimizing bandwidth usage without compromising semantic information. The results indicate that the proposed framework provides a practical, scalable, and robust solution for next-generation IoT-enabled smart transportation systems.