Instrumentation & Measurement Magazine 23-5 - 61

GDMD is applied to localize abnormal events for each frame in
the videos. The performance of the proposed method is verified
using UCSD datasets, which clearly show a higher efficiency of
our method, in comparison with other state-of-the-art methods.

References
[1] J. Kim and K. Grauman, "Observe locally, infer globally: a spacetime mrf for detecting abnormal activities with incremental
updates," in Proc. IEEE Conf. Computer Vision and Pattern
Recognition (CVPR 2009), pp. 2921-2928, 2009.
[2] R. Mehran, A. Oyama, and M. Shah. "Abnormal crowd behavior
detection using social force model," in Proc. IEEE Conf. Computer

Fig. 3. Comparison of frame-level detection using the proposed approach
with traditional methods.

The kNN-DEWMA scheme with nonparametric threshold was applied on the features generated by CNN, for the 12
scenarios in the USCD Ped2 dataset. In our study, the smoothing parameter 'ν' was set to 0.25. The monitoring results are
summarized in Table 1 and highlight the superiority of the
DEWMA scheme with a nonparametric threshold, by the fact
it achieved an AUC of 0.94 and had a lower EER of 0.05 (Frame
level).
Fig. 2 illustrates the detection of abnormal events (red
color) in the UCSD Ped2 dataset, with "INPUT" representing the input videos, "Detection" representing the detection of
anomalies detection in video frames, and "Localization" representing the localization of anomalies in each frame.
Fig. 3 shows that better performances are obtained with
the CNN-based DEWMA-kNN approach, in comparison
with some state-of-the-art machine learning procedures using PED2 dataset. Our method performs better than most of
the state-of-the-art methods at the frame level. Additionally,
results show a higher efficacy of the nonparametric DEWMAkNN algorithm, compared with the parametric one. This is
mainly due to the fact that the kNN-kNN incorporates all existing information from previous and actual observations in
the decision, which extends its detection performance.

human fall detection via shape features and improved extreme
learning machine," IEEE J. Biomedical and Health Informatics, vol.
18, no. 6, pp. 1915-1922, 2014.
[4] A. Adam, E. Rivlin, I. Shimshoni, and D. Reinitz. "Robust
real-time unusual event detection using multiple fixed-location
monitors," IEEE Trans. Pattern Analysis and Machine Intell., vol. 30,
no. 3, pp. 555-560, 2008.
[5] M. Bertini, A. Del Bimbo, and L. Seidenari, "Multi-scale and
real-time non-parametric approach for anomaly detection and
localization," Computer Vision and Image Understanding, vol. 116,
no. 3, pp. 320-329, 2012.
[6] S. Zhou, W. Shen, D. Zeng, M. Fang, Y. Wei, and Z. Zhang,
"Spatial-temporal convolutional neural networks for anomaly
detection and localization in crowded scenes," Signal Processing:
Image Commun., vol. 47, pp. 358-368, 2016.
[7] M. Sabokrou, M. Fayyaz, M. Fathy, Z. Moayedd, and R. Klette,
"Fully convolutional neural network for fast anomaly detection
in crowded scenes," 2016.
[8] S. Bouindour, M. M. Hittawe, S. Mahfouz, and H. Snoussi,
"Abnormal event detection using convolutional neural networks
and 1-Class SVM classifier anomaly detection in crowded
scenes," IET, ICDP 2017.
[9] S. Shirmohammadi and A. Ferrero, "Camera as the instrument:
the rising trend of vision based measurement," IEEE Instrum. and
Meas. Mag., vol. 17, no. 3, pp. 41-47, 2014.

Conclusion

[10] M. M. Hittawe, D. Sidibé, and F. Mériaudeau, "A machine vision

Crowded urban areas are becoming more vulnerable to terrorist
threats. Thus, the need for vision-based monitoring systems that
can detect and localize abnormal objects in crowded areas is becoming essential. Here, we propose an innovative vision-based
detection and localization method for the detection of atypical events in crowded areas. Our method uses an integrated
approach merging the benefits of the CNN model, the kNN algorithm and the DEWMA monitoring scheme. CNN is used to
extract complex and pertinent features from the captured videos from the supervised area. kNN is then applied to compute
the dissimilarity between the actual CNN features and the CNN
features from anomaly-free videos. The DEWMA statistical
monitoring scheme is subsequently used for sensing abnormal
changes based on kNN distances, followed by the computation
of a nonparametric threshold for DEWMA, using KDE to extend its flexibility and sensitivity to small changes. Lastly, the
August 2020

Vision and Pattern Recognition (CVPR 2009), pp. 935-942, 2009.
[3] X. Ma, H. Wang, B. Xue, M. Zhou, B. Ji, and Y. Li, "Depth-based

based approach for timber knots detection," in Proc. 12th Int.
Conf. Quality Control by Artificial Vision, Int. Soc. for Optics and
Photonics, Apr. 2015.
[11] O. Beya, M. M. Hittawe, D. Sidibé, D. and F. Meriaudeau,
"Automatic detection and tracking of animal sperm cells in
microscopy images," in Proc. 11th IEEE Int. Conf on Signal-Image
Technol. and Internet-Based Syst. (SITIS), pp. 155-159, Nov. 2015.
[12] F. Gao, J. Lin, H. Liu and S. Lu, "A novel VBM framework of fiber
recognition based on image segmentation and DCNN," IEEE
Trans. Instrum. Meas., vol. 69, no. 4, pp. 963-973, Apr. 2020.
[13] C. Nuzzi, S. Pasinetti, M. Lancini, F. Docchio, and G. Sansoni,
"Deep learning-based hand gesture recognition for collaborative
robots," IEEE Instrum. Meas. Mag., vol. 22, no. 2, pp. 44-51, 2019.
[14] S. E. Shamma and A. K. Shamma, "Development and evaluation
of control charts using double exponentially weighted moving
averages," Int. J. Quality and Reliability Manage., vol. 9, no. 6, 1992.

IEEE Instrumentation & Measurement Magazine 61

Instrumentation & Measurement Magazine 23-5

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-5