Use of Deep Learning using the YOLOv5 and YOLOv8 models to estimate traffic sign recognition efficiency
Castro Casas Alexis Del Piero, Surichaqui Alvarez Sebastian Amulek · 2023
The study was based on a dataset of 1046 images, addressing regulatory, preventive and informative signals. An experimental design comprising labeling and training of two previously mentioned neural network models was employed. After the training process, it was observed that YOLOv5 achieved an average accuracy of 88.6%, in contrast to YOLOv8, which achieved only 64.1%. These findings are supported by confusion matrices and post-training measurements of both models. Furthermore, in the evaluation of YOLOv5 in traffic sign detection, an average accuracy of 73% was achieved in 7 scenarios evaluated. In summary, the use of Deep Learning using YOLOv5 and YOLOv8 proved to be effective to a certain extent, since the accuracy in the detection of moving images and real time is influenced by several factors such as frame rate, illumination, return speed, among others.