Temporal Classification Error Compensation of Convolutional Neural Network for Traffic Sign Recognition

2017 International Conference on Control Engineering and Artificial Intelligence

计算机科学

Yoon, Seungjong^1 ; Kim, Eungtae^1

Dept. Electronics Engineering, Korea Polytechnic University, Sangidaehak-ro, Siheung-si, Gyeonggi-do

237, Korea, Republic of^1

关键词: Convolutional neural network;    Convolutional Neural Networks (CNN);    Error rate;    Temporal classification;    Temporal correlations;    Traffic sign recognition;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/806/1/012007/pdf
DOI  :  10.1088/1742-6596/806/1/012007

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

In this paper, we propose the method that classifies the traffic signs by using Convolutional Neural Network(CNN) and compensates the error rate of CNN using the temporal correlation between adjacent successive frames. Instead of applying a conventional CNN architecture with more layers, Temporal Classification Error Compensation(TCEC) is proposed to improve the error rate in the architecture which has less nodes and layers than a conventional CNN. Experimental results show that the complexity of the proposed method could be reduced by 50% compared with that of the conventional CNN with same layers, and the error rate could be improved by about 3%.