Visual tracking

Introduction

This work deals with the development of the use of a video camera in order to add useful information for tasks of a mobile robot roving in natural environments. When the robot executes a mission, the ground knowledge can be improved adding information other than geometrical, such as color and texture.

Our approach

We have proposed and implemented as a complementary representation, the nominative region model which indicates every region's nature in an image. Using this nominative model of the scene, it is possible to select a specific landmark and to track this object.

A color segmentation algorithm provides a synthetic description of the scene. Regions obtained from the segmentation stage are then characterized by their color and texture and afterward identified to obtain their nature (grass, rocks,...). Probabilistic methods are used to determine the nature of current elements in the environment. Then, one specific landmark is chosen according to its nature and shape and, this representation is tracked through an image sequence. The tracking is done using a comparison between the currently image and the landmark model.

Experiments:

Publications

Links to similar research