A Photonics-Enabled Framework for Intelligent Variable Traffic Sign Recognition and Anomaly Detection

Variable Message Signs (VMS) and Variable Speed Limits (VSL) are two of the most important parts on evolving Intelligent Transportation Systems (ITS) for supply drivers with dynamical traffic information in real time nowadays. Having accurate sign detections is crucial to road safety regulations. But such dynamic interfaces can be corrupted by occasional hardware failure, environmental conditions and data transmission errors, which, if not detected in time, can make serious trouble for the driver. The changes in the environment i.e. day and night lighting, unfavourable weather, change in the camera distance or quality make it quite a challenging task to identify anomalies in these road signs in real time. This paper introduces the photonics-enhanced deep learning system which is reliable in detecting anomalies in Variable Traffic Signs. The methodology relies on the CCTV videos collected over 2.5 years, a diversity of conditions including lighting, distance, and the design of signs, and annotated with thousands of samples. The suggested pipeline transfers the images through the CMOS-built sensors and optical transmission links, integrates YOLOv8 to detect the objects, OpenCV to pre-process the images, and Optical Character Recognition (OCR) to detect and analyze Variable Traffic Signs. Anomalies are identified by comparing the received text and numeric values to logical consistency, format and semantic completeness. Intended to be used as real-time roadside VMS and VSL applications, the system is proven in experiments that the system is capable of providing a lightweight yet scalable solution to automated sign surveillance, which reinforces the management of road safety.

Перевод пока недоступен