Bit @ TRECVID 2013: Surveillance event detection

Yicheng Zhao; Binjun Gan; Shuo Tang; Jing Liu; Xiaoyu Li; Yulong Li; Qianqian Qu; Xuemeng Yang; Longfei Zhang

Bit @ TRECVID 2013: Surveillance event detection

Yicheng Zhao, Binjun Gan, Shuo Tang, Jing Liu, Xiaoyu Li, Yulong Li, Qianqian Qu, Xuemeng Yang, Longfei Zhang

School of Computer Science and Technology

Beijing Institute of Technology

Research output: Contribution to conference › Paper › peer-review

Abstract

In this paper, we present an event detection system evaluated in TRECVid 2013. We investigated a generic statistical approach with spatio-temporal features applied to the event “Object Put”. This approach is based on local spatial constrained spatio-temporal descriptors, named as SC-MoSIFT. We extended the spatio-temporal features, MoSIFT, by relative position to camera. We also statistic the frequency and the location of each event occurs, and use selective hot region to construct spatial Bag-of-Feature. Non-linear SVM is exploited to train classifier for each event on each camera. In the limited experimental time, we adopted experiments and got comparable results to show the effectiveness of our approach.

Original language	English
Publication status	Published - 2013
Event	2013 TREC Video Retrieval Evaluation, TRECVID 2013 - Gaithersburg, United States Duration: 20 Nov 2013 → 22 Nov 2013

Conference

Conference	2013 TREC Video Retrieval Evaluation, TRECVID 2013
Country/Territory	United States
City	Gaithersburg
Period	20/11/13 → 22/11/13

Cite this

@conference{4a72788ce0c945b8bd1af3234f185bfd,

title = "Bit @ TRECVID 2013: Surveillance event detection",

abstract = "In this paper, we present an event detection system evaluated in TRECVid 2013. We investigated a generic statistical approach with spatio-temporal features applied to the event “Object Put”. This approach is based on local spatial constrained spatio-temporal descriptors, named as SC-MoSIFT. We extended the spatio-temporal features, MoSIFT, by relative position to camera. We also statistic the frequency and the location of each event occurs, and use selective hot region to construct spatial Bag-of-Feature. Non-linear SVM is exploited to train classifier for each event on each camera. In the limited experimental time, we adopted experiments and got comparable results to show the effectiveness of our approach.",

author = "Yicheng Zhao and Binjun Gan and Shuo Tang and Jing Liu and Xiaoyu Li and Yulong Li and Qianqian Qu and Xuemeng Yang and Longfei Zhang",

note = "Publisher Copyright: {\textcopyright} 2013 TREC Video Retrieval Evaluation, TRECVID 2013. All rights reserved.; 2013 TREC Video Retrieval Evaluation, TRECVID 2013 ; Conference date: 20-11-2013 Through 22-11-2013",

year = "2013",

language = "English",

}

TY - CONF

T1 - Bit @ TRECVID 2013

T2 - 2013 TREC Video Retrieval Evaluation, TRECVID 2013

AU - Zhao, Yicheng

AU - Gan, Binjun

AU - Tang, Shuo

AU - Liu, Jing

AU - Li, Xiaoyu

AU - Li, Yulong

AU - Qu, Qianqian

AU - Yang, Xuemeng

AU - Zhang, Longfei

PY - 2013

Y1 - 2013

N2 - In this paper, we present an event detection system evaluated in TRECVid 2013. We investigated a generic statistical approach with spatio-temporal features applied to the event “Object Put”. This approach is based on local spatial constrained spatio-temporal descriptors, named as SC-MoSIFT. We extended the spatio-temporal features, MoSIFT, by relative position to camera. We also statistic the frequency and the location of each event occurs, and use selective hot region to construct spatial Bag-of-Feature. Non-linear SVM is exploited to train classifier for each event on each camera. In the limited experimental time, we adopted experiments and got comparable results to show the effectiveness of our approach.

AB - In this paper, we present an event detection system evaluated in TRECVid 2013. We investigated a generic statistical approach with spatio-temporal features applied to the event “Object Put”. This approach is based on local spatial constrained spatio-temporal descriptors, named as SC-MoSIFT. We extended the spatio-temporal features, MoSIFT, by relative position to camera. We also statistic the frequency and the location of each event occurs, and use selective hot region to construct spatial Bag-of-Feature. Non-linear SVM is exploited to train classifier for each event on each camera. In the limited experimental time, we adopted experiments and got comparable results to show the effectiveness of our approach.

UR - http://www.scopus.com/inward/record.url?scp=85086288067&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85086288067

Y2 - 20 November 2013 through 22 November 2013

ER -

Bit @ TRECVID 2013: Surveillance event detection

Abstract

Conference

Other files and links

Fingerprint

Cite this