Traffic signal recognition using YOLO, a comparative analysis

The identification of traffic signs is a major challenge for intelligent automobiles. Recognition of traffic signs gives useful data, such as alerts and directions, for cooperative intelligent transport systems (CITS) and advanced driver assistance systems (ADAS). In realistic autonomous driving scenarios, traffic signs can be challenging to identify even with a very precise real-time method. As a result, drivers are calling for more secure and trustworthy traffic sign recognition technology, and current automakers are also working on developing one. All images captured by the actual driving car are blurry and distorted. The study of traffic sign recognition systems faces several challenges due to external, unmanageable components, such as the impact of unfavorable conditions, and the practical use is still in its infancy. This study provides the YOLO model for traffic sign identification as a solution for the issues mentioned above. The first step is to generally arrange the traffic signs into various categories. Next, the prepossessing process then follows the characteristics of the different kinds of signs. The improved convolution neural network receives the altered images and breaks down the categories into more specific subcategories. The YOLOv3-tiny and YOLOv4-tiny algorithms are examined by using the CCTSDB dataset and the Tsinghua-Tecent 100k(TTK100) dataset and YOLOv2,YOLOv3 andYOLOv5 are examined using the CCTSDB and HRRSD(High Resolution Remote Sensing Detection) datasets depending on the various categories. The findings indicate that YOLOv4-tiny algorithm achieves good recognition performance on both TTK100 and CCTSDB datasets and YOLOv5 on both CCTSDB and HRRSD dataset are most suited for use in traffic sign recognition systems and has a high classification accuracy.