Instrumentation & Measurement Magazine 23-5 - 57

Malicious Attacks Detection
in Crowded Areas Using Deep
Learning-Based Approach
Fouzi Harrou, Mohamad Mazen Hittawe, Ying Sun, and Ouadi Beya

ith the increasing need to ensure people's safety
in crowded areas, the development of a systematic anomaly detection mechanism is becoming
indispensable. Here are a few examples of recent malicious
attacks targeting crowded areas in big cities: in 2016, a truck
driver attacked and killed 84 persons walking in the promenade in Nice, France; and on 19 December, 2016, a truck was
deliberately driven into the Christmas market, in Berlin,
Germany, killing 12 people and injuring 56 others. These attacks demonstrate the need for efficient monitoring systems
to avoid such devastating attacks. To do so, early detection
and prevention abilities are vital. Detecting and localizing
abnormal events in crowded scenes is important and has significant implications in video surveillance applications. Video
surveillance can be challenging, as abnormal events can be unpredictable and changing, based on the context. Accurately
detecting and localizing anomalies in videos is a powerful tool
that can help to improve security and understand the behavior
of anomalies. In this paper, we present an automated visionbased monitoring scheme specifically designed for atypical
event-detection and localization in crowded areas.

Abnormal Event Detection
To achieve the reliable detection of abnormal events based on
videos, a substantial amount of research effort has been invested over the last two decades [1]. Several methods have
been designed using trajectories analysis to uncover atypical
events [1], [2]. However, these techniques usually need precise
tracking solutions and are highly sensitive to occlusion [3]. Alternatively, other techniques use spatio-temporal features for
representing the events in the video and do not require trajectories analysis [4]. Techniques falling into this category use
low-level local visual features (e.g., motion or texture) for background modeling and construction of template behavior [4]. In
[5], a multi-scale and non-parametric approach is proposed to
detect and locate occurred atypical events in real-time. In [6],

at first the optical flow is applied to select video volumes of interest, and then a convolutional neural network (CNN) is used
to extract relevant features. Many studies have been done to
extract spatiotemporal features based on the idea of Bag of
Video words (BOV). Essentially, the backbone idea of this concept to detect abnormal events consists of using local video
volumes based on dense sampling or by selecting points of interest. Unfortunately, this approach ignores the relationship
between video volumes. Recently, alternative solutions have
been proposed using the contextual information, but high
computational capacity is needed for the implementation of
these methods, which makes them unsuited in real-time purposes. In [7], the authors present fully CNNs, combining a
pre-trained CNN and another convolutional layer trained using sparse auto-encoder. In [8], Bouindour et al. propose to use
CNNs and a 1-Class support vector machine algorithm to detect anomalies in video datasets. Note that the major challenge
is to extract relevant descriptors and design detection schemes
able to uncover unusual and atypical behaviors with a high detection rate and a minimum of false alarms [9]-[11].
Recently, with the rapid advancements in deep learning
and computational technologies, representations of data are
accomplished in a sophisticated way and learned by end-toend neural networks. This study was motivated by the strong
capacity of the deep learning CNN to extract relevant features
from images. Particularly, this paper proposes an efficient vision-based approach for attack detection and localization in
crowded areas. This approach merges the desirable properties
of a CNN to learn relevant features from videos, the flexibility of the k-Nearest Neighbor (kNN) algorithm, and the
sensitivity of double Exponentially Weighted Moving Average
(DEWMA) to sense small changes in time series data.
In this paper, we treat the problem of malicious attacks
in crowded areas as an anomaly detection problem based on
features extracted from the CNN model. The design of a CNNbased detection approach is performed in two phases. The

This work was supported by King Abdullah University of Science and Technology,
Office of Sponsored Research, under Award no: OSR-2019-CRG7-3800.
August 2020

IEEE Instrumentation & Measurement Magazine 57
1094-6969/20/$25.00©2020IEEE

Instrumentation & Measurement Magazine 23-5

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