12th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2019, 29th - 30th Jun 2019, Weihai, China

Research Article

A Novel Abnormal Driving Detection Method via Deep Learning in Wireless Sensor Network

Download658 downloads
  • @INPROCEEDINGS{10.4108/eai.29-6-2019.2282840,
        author={Xi  Liu and Mingyuan  Luo and Wei  Wang and Wei  Huang},
        title={A Novel Abnormal Driving Detection Method via Deep Learning in Wireless Sensor Network},
        proceedings={12th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2019, 29th - 30th Jun 2019, Weihai, China},
        publisher={EAI},
        proceedings_a={MOBIMEDIA},
        year={2019},
        month={6},
        keywords={abnormal driving detection wide group densely network wide group residual densely network},
        doi={10.4108/eai.29-6-2019.2282840}
    }
    
  • Xi Liu
    Mingyuan Luo
    Wei Wang
    Wei Huang
    Year: 2019
    A Novel Abnormal Driving Detection Method via Deep Learning in Wireless Sensor Network
    MOBIMEDIA
    EAI
    DOI: 10.4108/eai.29-6-2019.2282840
Xi Liu1, Mingyuan Luo1, Wei Wang2, Wei Huang1,*
  • 1: Nanchang University
  • 2: Chang’an University
*Contact email: n060101@e.ntu.edu.sg

Abstract

In this study, the abnormal driving detection in the current research hotspot wireless sensor network (WSN) is emphatically discussed, and three improved fusion models based on Densely Connected Convolutional Network (DenseNet), which is named Wide Group Densely Network (WGD), Wide Group Residual Densely Network 1 (WGRD1), and Wide Group Residual Densely Network 2 (WGRD2) respectively, are proposed for the first time. WGD introduces two deep learning network indicators, width and cardinality, into DenseNet. WGRD1 and WGRD2, on the basis of WGD, use two different methods to introduce the important idea of ResNet into DenseNet, which is residual-block output and direct-connected streams are added by elements. These three models use end-to-end learning for training. The experimental analysis based on the abnormal driving image data set shows that the performance of our improved model for abnormal driving detection in the wireless sensor network is better than several excellent deep learning models and traditional deep learning models.