Open AccessReal-Time Vehicle Taillight Recognition Based on Siamese Recurrent Neural Network
Author(s) -
Ming Liu,
Bingyan Liao,
Chenye Wang,
Yayun Wang,
Yaog Wang
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1673/1/012056
Subject(s) - computer science , artificial intelligence , pattern recognition (psychology) , block (permutation group theory) , artificial neural network , recall , mathematics , linguistics , philosophy , geometry
Vehicle taillight recognition has been well investigated recent years. However, few methods can be applied in practical because of the time consumption or less recognition accuracy. In this paper, we design a light-weight framework for vehicle taillight recognition in real-time, which is divided into detection and recognition stage. To achieve this purpose, Finer Detection Block and dense anchor regression are utilized to improve the robust of detection stage. In addition, a Siamese CNN-GRU network is proposed, which not only captures the short-time difference but also learns the long-time flash features. We conduct ablation experiments to verify our proposed method, which obtains a mean recall of 94.34% with 103 FPS on 1080Ti.