Research on application of meta-learning in traffic light recognition

Abstract

Traffic light recognition is essential to intelligent vehicle information perception, and its accuracy directly affects traffic safety. Based on sorting out and analyzing the existing related research on traditional machine learning and deep learning in traffic light recognition, the basic principles of two standard algorithms in meta-learning are introduced in detail. Compared with traditional machine learning, using the MAML algorithm and Reptile algorithms in meta-learning to recognize traffic lights by selecting the WPI dataset and programming in PYTHON. The simulation results show that the recognition accuracy based on the meta-learning algorithm is higher than that based on the traditional machine learning algorithm; In the meta-learning algorithm, the Reptile algorithm outperforms the MAML algorithm.