Comparison of VGG-16 and InceptionV3 for Traffic Sign Detection and Classification

Engineering notes

Experimental results showed that VGG-16 outperforms InceptionV3 with 97% accuracy on one dataset, but suffer from overfitting on the other.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

In the future, autonomous vehicles need to detect and identify traffic signs, and figure out their locations. Traffic sign recognition systems may play a crucial role in self-driving cars, artificial driver assistance, traffic surveillance as well as traffic safety. Accordingly, traffic sign recognition is an important area of research that triggered many research studies. Nowadays, more and more object recognition tasks are solved using Convolutional Neural Networks (CNN). Due to their high recognition rate and fast execution, CNNs have enhanced most computer vision tasks, both those that are already in place and those that are new. Many researchers exploited CNNs (among other machine learning and deep learning models) for traffic sign recognition and experimented with numerous datasets. Nevertheless, this is still an open research area in which many other models and datasets can be explored. In this paper, we compare VGG-16 and InceptionV3 CNN models for this purpose and experiment using two different datasets. Experimental results showed that VGG-16 outperforms InceptionV3 with 97% accuracy on one dataset, but suffer from overfitting on the other. This will be handled in future work.

Links and sources

Comments