TY - GEN
T1 - Pedestrian positioning in surveillance video using anthropometric properties for effective communication
AU - Sato, Toshio
AU - Qi, Xin
AU - Yu, Keping
AU - Wen, Zheng
AU - Myint, San Hlaing
AU - Katsuyama, Yutaka
AU - Tokuda, Kiyohito
AU - Sato, Takuro
N1 - Funding Information:
ACKNOWLEDGMENT This research has been supported by research grant for expanding radio wave resources (JPJ000254) of Ministry of Internal Affairs and Communications under contract for “Research and development of radar fundamental technology for advanced recognition of moving objects for security enhancement”
Publisher Copyright:
© 2020 IEEE.
PY - 2020/10/19
Y1 - 2020/10/19
N2 - Positioning of pedestrians or persons is an important technique for video-based systems. For network surveillance systems, positioning can be applied to reduce data volume for storage devices and communication traffic. In this paper, we propose a simple positioning method using anthropometric properties such as a face length. A foot point in an image is estimated based on face detection results and anthropometric properties, then perspective transformation converts the foot point into a position on the floor plane. We improve the anthropometric model to reduce estimation errors of positioning. Moreover, as an application of pedestrian positioning, we implement data reduction functions of video data for surveillance systems. Experiments using a 4K video indicate that the average positioning error improves to 0.5 m. In terms of data reduction, we found that combination of tracking, selection of key frames, cropping, resizing, and JPEG compression reduce the 35.6 MB video data to 70 kB. These experiments induce that our approach realize simple and precise positioning and data reduction for effective communication for video surveillance systems.
AB - Positioning of pedestrians or persons is an important technique for video-based systems. For network surveillance systems, positioning can be applied to reduce data volume for storage devices and communication traffic. In this paper, we propose a simple positioning method using anthropometric properties such as a face length. A foot point in an image is estimated based on face detection results and anthropometric properties, then perspective transformation converts the foot point into a position on the floor plane. We improve the anthropometric model to reduce estimation errors of positioning. Moreover, as an application of pedestrian positioning, we implement data reduction functions of video data for surveillance systems. Experiments using a 4K video indicate that the average positioning error improves to 0.5 m. In terms of data reduction, we found that combination of tracking, selection of key frames, cropping, resizing, and JPEG compression reduce the 35.6 MB video data to 70 kB. These experiments induce that our approach realize simple and precise positioning and data reduction for effective communication for video surveillance systems.
KW - Data reduction
KW - Efficient communication
KW - Face detection
KW - LPWA
KW - Positioning
KW - Surveillance video
KW - Tracking
UR - http://www.scopus.com/inward/record.url?scp=85099572138&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85099572138&partnerID=8YFLogxK
U2 - 10.1109/WPMC50192.2020.9309520
DO - 10.1109/WPMC50192.2020.9309520
M3 - Conference contribution
AN - SCOPUS:85099572138
T3 - International Symposium on Wireless Personal Multimedia Communications, WPMC
BT - WPMC 2020 - 23rd International Symposium on Wireless Personal Multimedia Communications
PB - IEEE Computer Society
T2 - 23rd International Symposium on Wireless Personal Multimedia Communications, WPMC 2020
Y2 - 19 October 2020 through 26 October 2020
ER -