Traffic Sign Classification Based on Prototypes
Hao Fu, Hongjun Wang · 2021 16th International Conference on Intelligent Systems and Knowledge Engineering (ISKE) · 2021
Traffic signs classification is an indispensable task for intelligent vehicles. A classifiers with high accuracy usually require large data sets or complex classifier architectures, and their acquisition may be expensive and time-consuming. In order to solve this problem, a new method was proposed. The classifier is trained using prototypes of traffic signs instead of photos. First is the prototype of the task object, and then combines the background photos captured from the search engine to build the prototype data set. Then, after pre-training ConvNets on the existing database TSRD, the classifier was fine-tuned on the prototype data set. The classifier was tested on GTSRB-dataset and achieved an accuracy of 90.13%.