Vision-Based Traffic Signal Light Detection for Intelligent Signal Timing Systems

The monitoring and control of traffic signals have become increasingly important in intelligent transportation systems, where manual observation and fixed-timing control often fail to respond effectively to changing road conditions. This project presents Vision-Based Traffic Signal Light Detection for Intelligent Signal Timing Systems, a deep learning framework designed to detect traffic signal light states from road-camera images. The system uses two advanced YOLO-based models: High-Frequency Texture Detail YOLO (HFTD-YOLO) and Lightweight Multi-Scale Context YOLO (LMSC-YOLO). HFTD-YOLO focuses on extracting fine texture, spatial, and small-object features for accurate signal-light detection, while LMSC-YOLO provides lightweight multi-scale feature learning for faster and efficient inference. Both models classify traffic signal lights into three classes: red, green, and yellow. The models are developed using Python, PyTorch, OpenCV, and the Ultralytics YOLO framework. By using an image-based detection approach, the system can identify small traffic-light objects under varied road scenes and generate bounding boxes, class labels, and confidence scores. This automated detection framework supports intelligent signal timing decisions, reduces manual monitoring effort, and improves the reliability of traffic signal state recognition.

Keywords: Traffic Signal Light Detection, Intelligent Signal Timing, High-Frequency Texture Detail YOLO (HFTD-YOLO), Lightweight Multi-Scale Context YOLO (LMSC-YOLO), YOLOv11, Computer Vision, Object Detection, Traffic Management.