Traffic Sign Recognition Algorithm Based on Improved Matrix Capsule Neural Network
Traffic sign recognition is one of the most important functions of driving assistant system.However,the motion blur in the traffic sign images has great impact on the traffic sign recognition accuracy.A traffic sign recognition algorithm based on im-proved matrix capsule neural network is proposed to solve this problem.Matrix capsule neural network is used to mix the capsules in lower capsule layer and generate ones in higher layer.It can build the correlations among features and generate higher level features.Traffic sign category could be inferred with the help of activations of classification capsules.The encoder used to generate the feature in primary capsule will be pre-trained by siamese neural network.This process can make the feature code of traffic signs more dis-criminative.The pre-trained encoder can be used to generate the feature of primary capsule.The experiment show that with the pro-posed method the convergence difficulty of matrix capsule neural network can be relieved and improve the motion-blurred traffic sign recognition accuracy by building feature correlation.
国家科技期刊平台
NSTL
万方数据
