A Fusion Model with YOLOv3 and ShuffleNetV2 Networks for Traffic Sign Recognition
Paper Key : IRJ************581
Author: Caipeng Wu,Xinghuan Wu,Yueyun Jia
Date Published: 07 Dec 2022
The rapid and accurate recognition of traffic signs helps to improve the performance of advanced driver assistance systems and provides important safety guarantees for unmanned driving. Aiming at the existing object detection algorithms with the low accuracy of traffic sign recognition, weak generalization ability, and difficult detection for small targets, which cannot be well applied to practical applications, a lightweight deep network model for fast and accurate recognition of traffic signs is established by introducing the lightweight deep learning network ShuffleNetV2 and improving the regression-based object detection network model YOLOv3. Five data augmentation techniques were used to amplify the public dataset and the network parameters were trained using the transfer learning strategy, indicating that the new network was less computationally intensive and could achieve an accuracy of 96%.