TY - GEN
T1 - A multimodal detection and tracking system based on deep-learning for traffic monitoring
AU - Li, Xinxu
AU - Yu, Guizhen
AU - Wand, Yunpeng
AU - Wu, Xinkai
AU - Ma, Yalong
N1 - Publisher Copyright:
© ASCE.
PY - 2018
Y1 - 2018
N2 - Traffic surveillance videos can provide data of multimodal (car, bus, pedestrian, bicyclist, et al.) activities for transportation research in general. However, automated data extraction remains an issue due to illumination variation, size scaling, and angle changes. To tackle these challenges, this paper proposed a deep-learning based multimodal detection and tracking system. The proposed system consists of two modules: 1) a multimodal detection module which is based on the object detection framework of Faster R-CNN, and 2) a multiple object tracking module which is based on the KCF tracking algorithm. A comprehensive experiment based on traffic videos captured in different scenarios is conducted to evaluate the performance of the proposed system. Testing results demonstrate that the proposed system can achieve better performance for multimodal detection and tracking compared with existing methods. The proposed system is tested to be robust to illumination variation, size scaling, and angle changes.
AB - Traffic surveillance videos can provide data of multimodal (car, bus, pedestrian, bicyclist, et al.) activities for transportation research in general. However, automated data extraction remains an issue due to illumination variation, size scaling, and angle changes. To tackle these challenges, this paper proposed a deep-learning based multimodal detection and tracking system. The proposed system consists of two modules: 1) a multimodal detection module which is based on the object detection framework of Faster R-CNN, and 2) a multiple object tracking module which is based on the KCF tracking algorithm. A comprehensive experiment based on traffic videos captured in different scenarios is conducted to evaluate the performance of the proposed system. Testing results demonstrate that the proposed system can achieve better performance for multimodal detection and tracking compared with existing methods. The proposed system is tested to be robust to illumination variation, size scaling, and angle changes.
UR - https://www.scopus.com/pages/publications/85044221592
U2 - 10.1061/9780784480915.059
DO - 10.1061/9780784480915.059
M3 - 会议稿件
AN - SCOPUS:85044221592
T3 - CICTP 2017: Transportation Reform and Change - Equity, Inclusiveness, Sharing, and Innovation - Proceedings of the 17th COTA International Conference of Transportation Professionals
SP - 572
EP - 582
BT - CICTP 2017
A2 - Wang, Haizhong
A2 - Sun, Jian
A2 - Lu, Jian
A2 - Zhang, Lei
A2 - Zhang, Yu
A2 - Fang, ShouEn
PB - American Society of Civil Engineers (ASCE)
T2 - 17th COTA International Conference of Transportation Professionals: Transportation Reform and Change - Equity, Inclusiveness, Sharing, and Innovation, CICTP 2017
Y2 - 7 July 2017 through 9 July 2017
ER -