Intelligent Vehicle for Urban Road Traffic Recognition System Using Deep Learning Algorithm

In urban road autonomous driving, accurately identifying various objects on the road is crucial for ensuring safety and reliability. The Faster R-CNN model, known for its real-time object detection capabilities, is a key component in urban road traffic monitoring systems. It uses a two-stage approach involving a region proposal network (RPN) for generating object bounding boxes and a convolutional neural network (CNN) for refining these proposals to classify and localize objects accurately. Enhanced by improvements like refined anchor box prediction, feature pyramid networks, and advanced backbone architecture, Faster R-CNN offers better accuracy and efficiency, although it can be computationally intensive and slower for real-time applications. Urban road traffic monitoring faces challenges such as high computational demands, the need for real-time processing, and complex urban environments. Traditional models like Faster R-CNN, while accurate, can be too slow for real-time applications. The DETR (DEtection TRansformer) model, using a transformer-based architecture, provides robust feature extraction but may struggle with speed. YOLO models, on the other hand, are fast but may sacrifice some accuracy for speed. To address these issues, the proposed DETR-YOLO v8n Fusion model combines the strengths of DETR's robust feature extraction and YOLO v8n's real-time detection capabilities. The fusion leverages attention mechanisms to improve accuracy without compromising speed, providing a balanced solution for urban traffic monitoring. The study highlights improvements in YOLO v8, including refined anchor box prediction, feature pyramid networks for multi-scale detection, and an advanced backbone architecture. These enhancements enable the model to identify objects with greater accuracy, especially in complex urban traffic scenarios. The model introduced in this paper, YOLO v8n, is faster and more accurate in detecting and monitoring vehicles on urban roads, with an accuracy of 0.860 mAP@50, a precision of 85.0%, and a recall of 85.8%. This research contributes to the field of intelligent transportation by showcasing advancements in object detection for autonomous vehicles and setting new benchmarks in urban road traffic monitoring systems. The integration of YOLO v8 and DETR continues to refine object detection mechanisms, enhancing the safety and efficiency of intelligent vehicles navigating complex urban environments.