pdf

Real-time driver monitoring system via modal and viewpoint analysis

  • 2024-09-04
  • 5.61MB
  • Points it Requires : 1

通过模态和视角分析的实时驾驶员监控系统

Driver distractions are known to be the dominant cause

of road accidents. While monitoring systems can detect non-driving-related activities and facilitate reducing

the risks, they must be accurate and efficient to be applicable. Unfortunately, state-of-the-art methods prioritize

accuracy while ignoring latency because they leverage

cross-view and multimodal videos in which consecutive

frames are highly similar. Thus, in this paper, we pursue

time-effective detection models by neglecting the temporal relation between video frames and investigate the importance of each sensing modality in detecting drives’ activities. Experiments demonstrate that 1) our proposed

algorithms are real-time and can achieve similar performances (97.5% AUC-PR) with significantly reduced computation compared with video-based models; 2) the top

view with the infrared channel is more informative than

any other single modality. Furthermore, we enhance the

DAD dataset by manually annotating its test set to enable

multiclassification. We also thoroughly analyze the influence of visual sensor types and their placements on the

prediction of each class. The code and the new labels will

be released

unfold

You Might Like

Uploader
MartinFowler
 

Recommended ContentMore

Popular Components

Just Take a LookMore

EEWorld
subscription
account

EEWorld
service
account

Automotive
development
circle

About Us Customer Service Contact Information Datasheet Sitemap LatestNews


Room 1530, 15th Floor, Building B, No.18 Zhongguancun Street, Haidian District, Beijing, Postal Code: 100190 China Telephone: 008610 8235 0740

Copyright © 2005-2024 EEWORLD.com.cn, Inc. All rights reserved 京ICP证060456号 京ICP备10001474号-1 电信业务审批[2006]字第258号函 京公网安备 11010802033920号
×