Flow Measurement tool for Crowd Management Systems

Zakaria Alkabbab
Electrical Engineering Department
KAIST, Daejeon, Korea
[email protected]
ABSTRACT we discuss experimental results in Section 4 and we
Public security is a worldwide major concern that is conclude the paper in Section 5.
continuously confronting serious challenges especially in
2. RELATED WORK
crowd management area. Events, concerts, demonstrations
Most of the current research work in this area focuses on
and all kinds of gatherings can easily lead to uncontrollable
monitoring crowd as a management tool and carrying out
situations. Although, significant progress has been achieved
simulations which help in understanding crowd behaviors.
to develop and enhance new tools to deal with these issues,
Works like Marana et al. (1997) and Wu et al. (2006) [6],
human presence is still mandatory to ensure the
proposes crowd estimation systems which monitor and
effectiveness of common used security surveillance tools.
extract specific crowd information. In Boghossian and
In this paper, we present a novel user-independent
Velastin (1999) [2], a motion-based machine vision
measurement tool for event and crowd management to
technique is used to capture crowd flows and display visual
prevent overflows.
information which assists CCTV operators in understanding
Author Keywords the dynamics of the crowd under observation. The
Object detection, measurement tool, overflow prevention, monitoring devices act as management information systems
crowd management, grouping. and help in managing the crowd but they are not always
1. INTRODUCTION effective in reducing the overcrowding problem. Some
Crowding happens due to gathering of large number of studies also focus on empirical or experimental studies of
people in an area. Gathering of large crowd in places like pedestrian flow by means of analyzing video [5] [7].
stadiums or concerts is positive and desirable most of the In addition to monitoring systems, computational and
time. For example, a larger crowd might produce more mathematical models of crowd can help in increasing the
profit from commercial point of view. However, excessive awareness and knowledge on crowd behaviors. Building
crowding and poor crowd management can lead to and structure designs can be tested to see how they perform
unwanted incidents like crushing, stampedes, injuries and in overcrowding situations. Several models for crowd
even loss of life. A lot of efforts have been put in movements and behaviors have been proposed. Physics
understanding crowd features and behaviors. Although based models like particle, gas and fluid dynamics methods
numerous methods, solutions and tools have been used to use physical laws to model movements of pedestrians [4]
solve or improve the overcrowding problems but there is while cellular automata and other matrix-based [1] [3] [8]
space for improvement in all aspects especially in managing divide the simulation area into cells and simulate the crowd
crowd. The present research work addresses security movements based on the transition of pedestrians among
systems effectiveness in large scale events and provides these cells.
accurate measurement and census tools through a Our research is somewhat different from previous works
straightforward application based on video surveillance and because it basically consists of a user-independent interface
monitoring systems which are widely used in public places, that provides key information in crowd management such
indeed the use of these resources allow a large scale as individual speed, real-time density and real-time flow
deployment. The accuracy of this application depends on rate.
the image sharpness and quality as well as the position of 3. ARCHITECTURE
the camera. This paper is organized as follows. We discuss Briefly, this application collects video information from a
related works in Section 2. An overall architecture and main fixed camera providing sharp images, ideally with a bird’s
functions of the application is introduced in Section 3. In eye view. A first step resides in storing the background
Section 4, we present our implementation on Matlab. Then, image which will be updated every minute if no movement
appears; this approach is to cope with brightness change in
images over time. By continuously calculating the
Permission to make digital or hard copies of all or part of this work for difference between new frames and the background image,
personal or classroom use is granted without fee provided that copies are we can detect moving persons and make groups/macro
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, blocks of pixels corresponding to each person. In the
or republish, to post on servers or to redistribute to lists, requires prior meantime, we compute motion estimation on images to
specific permission and/or a fee. keep track of each person and study his movement. The
figure below summarizes the overall architecture of the block. There are various cost functions, of which the most
application: popular and less computationally expensive is Sum of
Absolute Differences (SAD) given by this equation:
Tracking ∑ ∑ | |

Where minI, minJ, maxI and maxJ are boundary indexes of

Grouping Motion Estimation each group of pixels that corresponds to a person or a
moving object that doesn’t belong to the background image.
yes Cij and Rij are the pixels being compared in current macro
no block and reference macro block, respectively.
Movement Detection

New frame Update

background
image

Video source

Fig.1: Information Flow

Motion estimation is invoked to find displacement of Fig.2: Block matching

moving groups of pixels (persons)
The Full Search algorithm is the most computationally
3. IMPLEMENTATION
expensive block matching algorithm of all. This algorithm
The implementation of the presented application was made
calculates the cost function at each possible location in the
on Matlab; it is structured into several functions. Each one
search window. As a result of which it finds the best
is saved in a separated M file. Following is a short
possible match and gives the highest PSNR amongst any
description of the m files in this zip
block matching algorithm. It is important to note that this
method is used only to find motion vectors of moving
Script to execute the application, plot
finalProjMain.m groups using boundary indexes calculated by the grouping
figures and show results
Motion Estimation (using Full Search function.
GmotionEstFS.m
method) Function Grouping
grouping.m Grouping Function The purpose of this function is to group macroblocks
tracking.m Tracking Function corresponding to moving objects/persons and find their
costFuncSAD.m Sum of Absolute Differences Function boundary indexes. Fig.3 illustrates this operation:

Motion Estimation using Full Search Method

The underlying supposition behind motion estimation is
that the patterns corresponding to objects and background
in a frame of video sequence move within the frame to form
corresponding objects on the subsequent frame. The idea
behind block matching is to divide the current frame into a
matrix of ‘macro blocks’ that are then compared with (1) Background image (2) New frame
corresponding block and its adjacent neighbors in the
previous frame to create a vector that stipulates the
movement of a macro block from one location to another in
the previous frame. This movement calculated for all the
macro blocks comprising a frame, constitutes the motion
estimated in the current frame. The search area for a good
macro block match is constrained up to ‘sr’ pixels on all (2)-(1) Movement detection Grouping
fours sides of the corresponding macro block in previous
frame. This ‘sr’ is called as the search range. Larger Fig.3: Grouping Process
motions require a larger ‘sr’, and the larger the search range
the more computationally expensive the process of motion Tracking
estimation becomes. The idea is represented in Fig 2. The This function compares motion vectors provided by
matching of one macro block with another is based on the computing motion estimation on the previous frame and
output of a cost function. The macro block that results in results of grouping function of the current frame to (1)
the least cost is the one that matches the closest to current analyze speed and trajectory of each group, (2) detect new
groups and (3) count groups that just left the frame. baseline framework for more general applications such
Moreover, tracking function also calculates real-time those discussed in future works.
density and flow rate.
ACKNOWLEDGMENTS
4. EVALUATION We thank Professor Munchurl Kim for supporting this
The grouping function is the core of the presented research work and providing helpful comments. Some of
application. In fact, accuracy measurements totally depend the references cited in this paper are included for illustrative
on that function. Therefore, we will focus on the constraints purposes only.
that can affect the exactitude of grouping results by
REFERENCES
studying the effect of proximity between persons on the 1.Blue, V.J., Adler, J.L., 2000. Cellular automata micro-
accuracy. simulation of bi-directional pedestrian flows. Journal of
the Transportation Research, 135–141.
100
2.Boghossian, B.A., Velastin, S.A., 1999, 5–8 September.
80 Motion-based machine vision techniques for the
Accuracy (%)

60
management of large crowds. In: Paper Presented at the
6th IEEE International Conference On Electronics,
40 Circuits and Systems, ICECS ‘99.
20 3.Burstedde, C., Klauck, K., Schadschneider, A., Zittartz,
0 J., 2001. Simulation of pedestrian dynamics using a two
0 20 50 >50 dimensional cellular automaton. Physica A 295, 507–525.
Proximity b/w persons (cm) 4.Helbing, D., Molnar, P., Farkas, I.J., Bolay, K., 2001.
Fig.4: Accuracy vs. proximity b/w persons Self-organizing pedestrian movement. Environment and
Planning B: Planning and Design 28, 361–383.
The graph shows that the accuracy of the grouping function 5.Hoogendoorn, S.P., Daamen, W., Bovy, P.H.L., 2003.
decreases dramatically to 50% when people are too close to Extracting microscopic pedestrian characteristics from
each other. However, this situation cannot occur because video data. In: Paper Presented at the Transportation
the application will detect earlier that density per frame is Research Board Annual Meeting 2003, National
very high. Academy,Washington DC, Press.
5. FUTURE WORKS 6.Marana, A.N., Velastin, S.A., Costa, L.F., Lotufo, R.A.,
Future works may address various higher level applications. 1997. Estimation of crowd density using image
For instance, by combining the actual system with a facial processing. In: Paper Presented at the IEE Colloquium on
or object recognition subsystem, we can conceive an Image Processing for Security Applications.
Automatic attendance check system in classrooms or
Census application in buildings, trains or airplanes. 7.Teknomo, K., 2002. PhD thesis. Tohoku University,
Sendai, Japan.
6. CONCLUSION
In this paper, we presented a novel application based on 8.Thompson, P., Marchant, E.W., 1995. A computer model
video surveillance and monitoring systems to provide for the evacuation of large building populations. Fire
effective measurement tools for detecting overflow in Safety Journal 24, 131–148.
public places. We believe our work came up with a major 9.Still, G.K., August 2000. Crowd Dynamics. PhD thesis,
contribution to crowd management field and can serve as a University of Warwick.