Video surveillance at banks, examples of situations to detect automatically

 

MSc dissertation proposal 2009/2010

 

Observit - Video Surveillance for Banks

 

 

Introduction:

 

"For good reason, banks are often seen as the most secure institutions on the planet. We entrust them with our money, and it's their job to keep it safe. In order to fulfill this commitment, top-notch video surveillance systems are a necessity. With recent innovations in digital technology and IP surveillance, many banks are looking to increase the effectiveness of their security systems by investing in this new technology." Excerpt from http://www.videosurveillance.com/banks.asp.

 

Nowadays, most of the bank facilities have numerous video surveillance cameras but too few people looking at them. Hence the chance to detect suspicious behaviours is actually very small. In fact most of the installed systems lack automatic components in order to detect unexpected events that may be configuring dangerous situations such as bank robberies. In this work, the main objective is to study the automatic detection of a number of unexpected events that may configure bank robbery situations.

 

This work will be conducted in conjunction with the company Observit, www.observit.pt, under a protocol with IST/DEEC.

 

Supervision: Prof. José Gaspar, Prof. Alexandre Bernardino, Eng. Bernardo Motta (Observit)

 

 

Objectives:

 

The objectives of this work are twofold: (i) detecting persons, and (ii) detecting unexpected events such as faces hidden under the clothes or people lying on the floor. At the end of the work it is expected to demonstrate a video processing tool able to signal the unexpected events in a video sequence.

 

 

Detailed description:

 

Video surveillance for banking applications is still a very active research-and-development topic mainly because of the large variability of the installations, fixed/mobile cameras, indoor/outdoor scenarios, etc, and because of the huge variety of the principal actors, namely people of various heights, ethnic cultures, ways of dressing, physical impairment, etc, and their actions.

 

The main objective of this work is the automatic identification of events that may indicate a bank robbery is undergoing. In this work, the events to study will be restricted to two important cases: detecting people entering the bank with their faces hidden, and detecting people lying on the floor. Note that while there are nowadays very efficient face detectors [Viola04, OpenCV], the case of hidden faces challenges the utilization of those detectors. It is considered an important part of the project selecting ways to circumvent this problem. One suggested way starts by the detection of the persons (see [Dollar09]), and then estimating the face location from the silhouette of each person. In this case, the face detectors could be once more useful just to determine the low evidence of visible faces.

 

The work is therefore organized in the following main steps:

1) Detecting people, using e.g. background subtraction. One specific is of importance here, namely whether the detection can include outdoor contents (distracters) such as the imaging provided by windows making the street visible. In this case discriminating people from mobile objects (e.g. cars) can benefit from the strategies described in [Figueira09].

2) Detecting the pose of a person, in order to estimate the location of the face in the image.

3) Verifying that the face is, or is not, visible.

 

The collaboration with the Observit company, besides bringing to the work references on commercial systems, is also a contact point with potential clients (banks) for the system.

 

 

References:

 

[Viola04] "Robust Real-Time Face Detection", Paul Viola and Michael J. Jones, International Journal of Computer Vision, Vol. 57, No. 2. (1 May 2004), pp. 137-154

 

 [Figueira09] "Optical flow based detection in mixed human robot environments", D. Figueira,P. Moreno, A. Bernardino, J. Gaspar, J. Santos-Victor. In Proc. of ISVC 2009 - 5th International Symposium on Visual Computing, 2009

 

[Dollar09] "Pedestrian detection: A benchmark", P. Dollar, C. Wojek, B. Schiele, P. Perona, IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2009, 20-25 June 2009 Page(s):304 - 311

 

[OpenCV] Open Computer Vision Library, http://sourceforge.net/projects/opencv/

 

 

Expected results:

 

At the end of the work the students will have enriched their knowledge in:

* Computer vision

* Video Surveillance

 

Examples of expected demonstrations in simulated and/or real environments:

* Detections of persons

* Detecting visible / not visible faces in the persons found

 

 

Observations:

--

 

 

More MSc dissertation proposals on Computer and Robot Vision in:

 

http://omni.isr.ist.utl.pt/~jag