Developing AI-Powered Solutions For Traffic Analysis Using Yolov8 For Object Detection And Textual Descriptions Generation

Abstract

There is a need for creative traffic monitoring due to the rise in traffic jams and accidents. Conventional approaches are insufficient because they lack precision and speed in real-time and typically involve human interaction. YOLOv8, the most sophisticated iteration of the "You Only Look Once" object identification model, will be used in this work to develop a machine-learning model that can identify traffic. Using traffic images, the model can identify and categorize traffic items, produce textual descriptions, and provide real-time insights for improved traffic control. There were 17 distinct object classes in the custom traffic dataset, including vehicles, buses, and people. Additionally, data augmentation methods such as blurring and grayscale transformations were used to increase the model's resilience under various circumstances. To achieve remarkable performance metrics at mAP50 of 0.385 and mAP50-95 of 0.236, YOLOv8 was further optimized for over 30 epochs. While analyzing photos at a real-time speed of 30 to 50 milliseconds, the system demonstrated a high degree of accuracy in identifying and categorizing important traffic aspects. The model effectively detects objects like cars, buses, and trucks with processing times between 59.8 ms and 103.3 ms per image, which makes it suitable for real-time use. It also generates descriptive text from detected classes, enabling automated traffic analysis. By providing useful insights, this skill improves safety management, urban planning, and traffic monitoring. The suggested method demonstrates that deep learning and computer vision can be used to create automated, scalable traffic systems that address contemporary urban issues.