3D detection of people

Abstract

The aim of this work is to obtain the three-dimensional location of people in a scene using the information provided by several cameras; these cameras must have a range of view in common. The place chosen to record the videos for this work is the Nexus Lab in the Universität Stuttgärt. In this work, it is possible to distinguish two main tasks: ð€‚¾ The first task is to obtain a model of the camera which let us turn the twodimensional information from the images taken from the cameras, in three-dimensional information in a known reference-frame; this topic is studied in chapter 2. ð€‚¾ The second task is to obtain relevant information from the images of each camera, that is, the location of the people, which appear in an image scene, must be known to compute their three-dimensional location. This information will be obtained by reviewing different types of methods. This topic is described in chapter 3. Finally, the three-dimensional location of the people in the room has been accomplished by integrating the two-dimensional information obtained from the images of each camera with the model of the cameras and knowing their locations in the room. In order to quantify the measurement error of the system, several video tests have been recorded. The error analysis is accomplished in chapter 4.