Unimore logo AImageLab
Back to the project list

Multiple Camera Cooperation for Group Detection and zooming

The project it's a software that can automatically steer and control one or more PTZ cameras basing on information gained from a static camera. This system can show in detail suspect groups of people to help the surveillance and security of particular zones. The information by the fixed camera are extract with a particular algorithm of Crowd Detection.


comunecrowd

The use of surveillance cameras is gaining more and more impact on crime analysis and prevention. One of the problems that arises when the surveillance is applied in a crowded environment is that of recognizing people and maybe follow them making use of multiple cameras. The purpose of this project is that of making multiple cameras cooperate to automatically accomplish tasks that would be otherwise done manually by an expert operator.

While it's easy for a human mind finding a correspondence between the same object in different images, this is not the case for an automatic system, thus a number of mathematical transformation are necessary to be applied over the obtained images to find such correspondences.

In particular, the environment is set up with a static camera, used to make detection of people, with a crowd detection algorithm, in a scene and one or more PTZ cameras, used to get finer details such as zoomed views on the faces of the detected persons. The detector running on images from the static camera outputs a list of coordinates where the controllable cameras should be steered.

The designed solution implies translating such coordinates in the corresponding ones in the coordinate system of the moving cameras and than, after an initial calibration needed to model a conversion function to translate distances in the x,y coordinate system into the corresponding values of Pan, Tilt and Zoom, those values are computed for every single detected point to be used to automatically command the camera to move in the corresponding position.

Source code: 

Source code for Crowd Detection 

Datasets:

Training data 
Testing sequences 
Testing sequences on ViSOR 

Publications

1 Manfredi, Marco; Vezzani, Roberto; Calderara, Simone; Cucchiara, Rita "Detection of static groups and crowds gathered in open spaces by texture classification" Pattern Recognition Letters, PATTERN RECOGNITION LETTERS, vol. 44, pp. 39 -48 , 2014 | DOI: 10.1016/j.patrec.2013.11.001 Journal

Video Demo

Project Info

comunecrowd

Staff:

Duration:

01/01/2012 - 30/06/2013

Funded by:

Comune di Modena