Salons-itis

• Project Name: 3D tracker for serveral people using evolutive algorithm

• Authors: Sara Marugán Alonso (smarugan [at] gsyc [point] es) and Jose María Cañas Plaza (jmplaza [at] gsyc [point] es)

• Academic Year: 2007-2008

• Degree: Undergrad

• Jde Version: jde-4.2.1

• SVN Repository: source code (restricted)

• Tags: computer vision, fall detection, tracking

• Technology: c, jde

• State: Finished

• Source License: not defined

• Abstract:

This application tracks 3D position of several people walking inside a room in real time and learns their clothes color using the images from four cameras. To resolve the localization an evolutive algorithm has been designed. The algorithm uses two types of population with different purposes: searching for movement (explorers) and tracking of people (exploders). Information of color and movement is extracted from images and it's the support for population evolution.

• Videos:

• Images:

Motion detection is the first step to the algorithm. It is based on difference between two consecutive frames and background learned image. You can see the result in this image (a- non-filtered image, b- filtered image).

Through explorer population the algorithm detects movement. If this movement is significant, a explode population called race is generated in the detected movement zone. An example of explorer population is shown in this image.

A race is a point cloud that extract color and movement information from a zone where we know there is something because of its movement. Each cloud's particle gets color samples that stores in a hue histogram. It permits to generate a dinamic color filter for the race defining a umbral over the histogram.

Each race has its own 3D position estimation for the tracked object based on the position of all race's particles. Therefore, a trajectory can be generated.