License plate detection plays an important role in intelligent transportation system. However, it is still a challenging task due to plenty of complex scenes. Recent studies show that deep learning approaches achieve prominent results on general object detection. Therefore, in this paper, we propose a deep cascaded convolutional neural network for improving license plate detection in complex scenes. Firstly, we utilize convolutional features to generate candidate vehicles proposals. Then a network is used to detect a license from each vehicle proposal by analyzing the correlation between vehicles and licenses. Finally, we enhance detection performance by processing license boundary. Experimental results on a large dataset demonstrate that our method works effectively in a variety of complex scenes.
展开▼
