Automotive Radar Signal Classification Using Bypass Recurrent Convolutional Networks

Abstract

In this paper, we propose recurrent convolutional neural networks for V2X communication, which classify moving target in automotive radar system. Moving target classification is of importance in protecting the vulnerable road users. The proposed neural network is computationally efficient and has improved performance. We introduce bypass connection used in recurrent convolutional neural networks. For the systems like radar systems, in which memory is restricted and high reliability is required, we show that the proposed neural network outperforms the conventional approaches by comparing classification accuracy using radar data measured in realistic scenario.