6 CONCLUSIONS
We have demonstrated the apparent potential in tailor-
ing approximate counting to effectively detect distrac-
tions in unbound surveillance videos, with little hu-
man counseling. Our highly compact bag-of-counts
representation is scalable and provides learning of
many windows that outline a deeper past, to facilitate
the expansion of the more conventional, point similar-
ity approach. Running on an unoptimized prototype
version, the obtained sustainable throughput of about
half the input frame-rate on average, consistently on
all of our experimental videos and the full extent of
model parameter settings, provides for real-time ap-
plication compliance.
A natural progression of our work is to reduce the
absolute counting error by introducing more smaller
buckets for constructing the time-based histogram
over a window, and improve overall detection rate.
Rather than a step function that characterizes window
transitions in the current implementation, the use of a
tapered sliding window offers overlapping windows,
as frames gradually enter and exit the window, to fur-
ther assist and lower distraction misses.
ACKNOWLEDGEMENTS
We would like to thank the anonymous reviewers for
their insightful suggestions and feedback.
REFERENCES
Arasu, A. and Manku, G. S. (2004). Approximate counts
and quantiles over sliding windows. In Principles
of Database Systems (PODS), pages 286–296, Paris,
France.
Baeza-Yates, R. and Ribeiro-Neto, B., editors (1999). Mod-
ern Information Retrieval. ACM Press Series/Addison
Wesley, Essex, UK.
Chen, X. J., Zhan, Y. Z., Ke, J., and Chen, X. B. (2015).
Complex video event detection via pairwise fusion of
trajectory and multi-label hypergraphs. Multimedia
Tools and Applications, pages 1–22.
Datar, M., Gionis, A., Indyk, P., and Motwani, R. (2002).
Maintaining stream statistics over sliding windows.
SIAM Journal of Computing, 31(6):1794–1813.
Efrat, A., Fan, Q., and Venkatasubramanian, S. (2007).
Curve matching, time warping, and light fields: New
algorithms for computing similarity between curves.
Mathematical Imaging and Vision, 27(3):203–216.
Jaechul, K. and Grauman, K. (2009). Observe locally, in-
fer globally: A space-time MRF for detecting abnor-
mal activities with incremental updates. In Computer
Vision and Pattern Recognition (CVPR), pages 2921–
2928, Miami, FL.
Lee, L. K. and Ting, H. F. (2006). Maintaining significant
stream statistics over sliding windows. In Discrete Al-
gorithm (SODA), pages 724–732, Miami, FL.
Lee, S. C. and Nevatia, R. (2014). Hierarchical abnormal
event detection by real time and semi-real time multi-
tasking video surveillance system. Machine Vision
and Applications, 25(1):133–143.
Leskovec, J., Rajaraman, A., and Ullman, J. D., editors
(2014). Mining of Massive Datasets. Cambridge Uni-
versity Press, New York, NY.
Li, H., Zhang, Y., Yang, M., Men, Y., and Chao, H. (2014).
A rapid abnormal event detection method for surveil-
lance video based on a novel feature in compressed
domain of HEVC. In Multimedia and Expo (ICME),
pages 1–6, Chengdu, China.
Salton, G., Wong, A., and Yang, C. S. (1975). A vector
space model for automatic indexing. Communications
of the ACM, 18(11):613–620.
Shimada, A., Nagahara, H., and Taniguchi, R. I. (2015).
Change detection on light field for active video
surveillance. In Advanced Video and Signal Based
Surveillance (AVSS), pages 1–6, Karlsruhe, Germany.
Vezanni, R. and Cucchiara, R. (2010). Video surveil-
lance online repository (ViSOR): an integrated frame-
work. Journal of Multimedia Tools and Applications,
50(2):359–380.
ViSOR (2010). Video surveillance online repository.
http://www.openvisor.org .
Xiang, T. and Gong, S. (2008). Video behavior profiling
for anomaly detection. Pattern Analysis and Machine
Intelligence, 30(5):893–908.
Zhong, H., Shi, J., and Visontai, M. (2004). Detecting un-
usual activity in video. In Computer Vision and Pat-
tern Recognition (CVPR), pages 819–826, Washing-
ton, DC.
APPENDIX
Our meta-data for hand annotated time intervals that
capture distraction instances from each of our ViSOR
sequences are listed in Table 9.
KDIR 2016 - 8th International Conference on Knowledge Discovery and Information Retrieval
334