TY - GEN
T1 - WiDrive
T2 - 39th IEEE International Conference on Distributed Computing Systems, ICDCS 2019
AU - Bai, Yunhao
AU - Wang, Zejiang
AU - Zheng, Kuangyu
AU - Wang, Xiaorui
AU - Wang, Junmin
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/7
Y1 - 2019/7
N2 - Autonomous vehicles often need human driver to take over in some complicated conditions. Such a sudden takeover could jeopardize the vehicle's safety and stability if not han-dled properly. Hence, if the driver's takeover intention can be recognized as early as possible, the vehicle can have sufficient time to make important takeover preparation. The existing in-car monitoring systems are mostly based on camera, which have several key limitations, such as brightness condition and motion obscurity. On the other hand, WiFi-based wireless sensing has recently shown a great promise in human activity recognition, but mainly for large-scale movements performed in the room environment. In this paper, we propose WiDrive, a real-time in-car driver activity recognition system based on Channel State Information (CSI) changes of WiFi signals. WiDrive consists of three major components: A novel algorithm to extract small-scale in-car human activity features, a real-time recognition system based on Hidden Markov Model (HMM), and an online adaptation algo-rithm to adapt for different drivers and vehicles. We implement WiDrive with commercial WiFi devices and evaluate it in real cars. Our results show that WiDrive has an average recognition accuracy of 91.3% and improves the takeover safety.
AB - Autonomous vehicles often need human driver to take over in some complicated conditions. Such a sudden takeover could jeopardize the vehicle's safety and stability if not han-dled properly. Hence, if the driver's takeover intention can be recognized as early as possible, the vehicle can have sufficient time to make important takeover preparation. The existing in-car monitoring systems are mostly based on camera, which have several key limitations, such as brightness condition and motion obscurity. On the other hand, WiFi-based wireless sensing has recently shown a great promise in human activity recognition, but mainly for large-scale movements performed in the room environment. In this paper, we propose WiDrive, a real-time in-car driver activity recognition system based on Channel State Information (CSI) changes of WiFi signals. WiDrive consists of three major components: A novel algorithm to extract small-scale in-car human activity features, a real-time recognition system based on Hidden Markov Model (HMM), and an online adaptation algo-rithm to adapt for different drivers and vehicles. We implement WiDrive with commercial WiFi devices and evaluate it in real cars. Our results show that WiDrive has an average recognition accuracy of 91.3% and improves the takeover safety.
KW - Advanced Driving Assistance System
KW - WiFi sensing
KW - Wireless networks
UR - https://www.scopus.com/pages/publications/85074869645
U2 - 10.1109/ICDCS.2019.00094
DO - 10.1109/ICDCS.2019.00094
M3 - 会议稿件
AN - SCOPUS:85074869645
T3 - Proceedings - International Conference on Distributed Computing Systems
SP - 901
EP - 911
BT - Proceedings - 2019 39th IEEE International Conference on Distributed Computing Systems, ICDCS 2019
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 7 July 2019 through 9 July 2019
ER -