A convolutional feature map-based deep network targeted towards traffic detection and classification

作者：

Highlights：

• Covariance matrix based adaptive learning rate(LR). (83%).

• Multimodal CNN RGB-optical flow by fusing the convolution features. (84%)

• Different structured NoCs trained and best used for classification. (83%)

• Compared results of NoC trained and tested using blurred and normal sets (74.6%).

• 1C3fc with adaptive LR gives better performance in blur, optical and normal sets.

摘要

•Covariance matrix based adaptive learning rate(LR). (83%).•Multimodal CNN RGB-optical flow by fusing the convolution features. (84%)•Different structured NoCs trained and best used for classification. (83%)•Compared results of NoC trained and tested using blurred and normal sets (74.6%).•1C3fc with adaptive LR gives better performance in blur, optical and normal sets.

论文关键词：Deep learning,Traffic detection and classification,Convolutional neural network (CNN),Optical flow (OF),Adaptive learning

论文评审过程：Received 23 April 2018, Revised 21 October 2018, Accepted 4 January 2019, Available online 22 January 2019, Version of Record 26 January 2019.

论文官网地址：https://doi.org/10.1016/j.eswa.2019.01.014