期刊
IEEE SENSORS JOURNAL
卷 22, 期 5, 页码 4697-4708出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/JSEN.2022.3140791
关键词
Trajectory; Servers; Location awareness; Surveillance; Feature extraction; Sensor systems; Streaming media; Indoor localization; fusion positioning; PDR; video-based pedestrian tracking
资金
- National Natural Science Foundation of China [61675040, U1633129]
This paper proposes a pedestrian positioning scheme that combines IMU sensors of smartphones and surveillance video. The scheme utilizes PDR information to mark the pedestrians to be located and uses surveillance video to track all pedestrians in the field of view. Video pedestrian tracking is improved by a modified Markov model to reduce trajectory merging and interruption. Fusion of video trajectories is achieved through single-step feature weighting and multi-step DTW to improve matching accuracy. Experimental results show an average positioning error of less than 1.5 m for the proposed fusion scheme.
Existing pedestrian positioning technologies have difficulty balancing multiple aspects such as positioning accuracy, system maintenance, equipment and deployment costs. We propose a pedestrian positioning scheme employing inertial measurement unit (IMU) sensors of smartphones and surveillance video. We apply pedestrian dead reckoning (PDR) information to mark the pedestrians to be located and use surveillance video to track all pedestrians in the field of view. For video pedestrian tracking, we use a modified Markov model to reduce the probability of trajectory merging and interruption. In the fusion of video trajectories, we use single-step feature weighting and multi-step dynamic time warping (DTW) to improve the matching accuracy of trajectories, respectively. The experimental results show that the average positioning error of our proposed fusion scheme is less than 1.5 m.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据