Traffic sign recognition based on PCANet
Guoqing Le, Xue Yuan, Jing Zhang, HanSong Li · 2016
The local features or the inner information of the signs are used in the most traffic sign recognition(TSR) systems for recognition. Such as histograms of oriented gradients (HOG), local binary pattern histogram (LBP), scale-invariant feature transform (SIFT), and so forth. However these features in the treatment of the rotation, strongly illumination, and deformation situations still has limitations. In this paper, we have made use of PCANet, as known as a simple unsupervised convolutional deep learning network for TSR system. The contributions of this paper are as follows: 1) we utilize the PCANet for TSR on the CTSRB database; 2) we analyze the impact of various parameters in PCANet. Experimental results on validating the effectiveness of the PCANet with traffic sign recognition system are satisfying, even for containing rotation, deformation, or images taken under different illumination situation. PCANet is very effectively in extracting useful information for classification of traffic sign images.