Network surveillance cameras (Axis), People Tracking (Caviar dataset)

 

MSc dissertation proposal 2013/2014

 

Vision Based Multi-Target Detection and Tracking

 

 

Introduction:

 

The increasing need of surveillance in public spaces and the recent technological advances on embedded video compression and communications made camera networks ubiquitous. There are however missing methodologies for watching so much data captured by so many cameras having few staff. In this project the main objective is to design methodologies for the automatic detection and tracking of multiple targets using multiple cameras.

 

 

Objectives:

 

The main objectives of this project are: (i) calibrating the cameras composing the camera network, and (ii) detecting novel targets and maintaining their trajectories.

 

 

Detailed description:

 

This project proposal is focused on detecting and tracking moving objects, people or vehicles, which cross the fields of view of a network of cameras. Detection of moving objects is usually performed by background subtraction or by tracking visual features. Tracking involves motion modelling, filtering and data association.

 

Background subtraction usually involves modelling the background. Scene or illumination changes usually imply background changes that do not correspond to moving objects. One way to overcome these changes is to model the background as one, or multiple, gaussians per pixel.

 

Detecting moving objects by feature tracking is an interesting alternative when the objects occupy many pixels. In this case the texture of the objects usually allows detecting and tracking image features (e.g. SIFT or SURF). The similarities of the trajectories of the tracked features allow clustering independently moving objects.

 

Tracking moving objects can be helped by considering motion models. For instance, when the moving objects are cars or trucks, they cannot move sideways, and thus one has priors to help filtering. In this case, usually one uses Extended Kalman Filters (EKF). Considering multiple objects is handled by using multiple EKFs. Alternative methodologies include using Particle Filtering or Multiple Hypothesis Tracking.

 

In summary, this project proposal involves:

1) Calibrating the cameras forming the network of cameras

2) Detecting moving objects by background subtraction or feature tracking

3) Assigning a tracking filter to a moving object

4) Tracking multiple objects by using multiple filters

 

 

References:

 

[Hartley00] R. I. Hartley and I. Zisserman. Multiple view geometry in computer vision. pages 150–152, 2000. 

 

[LoweWWW] David G. Lowe. Demo software: Sift keypoint detector. http://www.cs.ubc.ca/˜lowe/keypoints/.

 

[Axis] IP camera manufacturer actively involved in the definition of the ONVIF standard http://www.axis.com/pt/onvif/index.htm

 

[ONVIF-www] open industry forum for the development of a global standard for the interface of IP-based physical security products http://www.onvif.org/

 

[Micheloni10] "Video Analysis in Pan-Tilt-Zoom Camera Networks: From master-slave to cooperative smart cameras", Christian Micheloni, Bernhard Rinner, and Gian Luca Foresti, IEEE Signal Processing Magazine September 2010

 

[Starzyk11] "Multi-tasking Smart Cameras for Intelligent Video Surveillance Systems", Wiktor Starzyk, Faisal Z. Qureshi, 8th IEEE International Conference on Advanced Video and Signal-Based Surveillance, 2011.

 

 

Requirements (grades, required courses, etc):

-

 

Expected results:

 

At the end of the work, the students will have enriched their experience in computer vision applied to surveillance camera setups.

 

 

Place for conducting the work-proposal:

ISR / IST

 

 

More MSc dissertation proposals on Computer and Robot Vision in:

 

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