Stereo Vision Based Object Detection for Mobile Robots (Stereo Visione per il Riconoscimento di Oggetti nella Robotica Mobile)

L'anteprima di questa tesi è scaricabile in PDF gratuitamente.
Per scaricare il file PDF è necessario essere iscritto a Tesionline.
L'iscrizione non comporta alcun costo. Mostra/Nascondi contenuto.

1.2. ACQUISITIONMETHODS 3 detected object can easily fail in different scenarios. A first reason is connected to the fact that monocular computer vision depends on the quality of the camera, so there is a strict relation with the camera construction deficiencies and the data-stream itself. One case is when the image is underexposed as in Figure 1.1b; it makes very hard for a 2D image processing application to recover the missing information. This problem is probably solved with the use of better camera sensors. A second reason is linked to the fact that the computer vision system uses 2D images which are just a projection of the 3D world on a plane. As shown in Figure 1.1c, a good match is dependent on the context. In the first case, while the system could apparently match the model template (a tree), after a zoom-out we observe that the tree was in fact another picture itself, lying in a different 3D geometric shape (an advertisement panel in this case). In the second case, it could be really hard for the system to recognize a single template among objects with similar characteristics. 1.2 Acquisition methods The ability to realize sensors that can provide three-dimensional data (3D) has always aroused considerable interest both in scientific and industrial fields because of the many applications where these sensors could be used. Interestingly the three-dimensional passive sensors based on traditional cameras are potentially cheap and less invasive in the environment in which they are used compared with sensors based on active technologies. So, though there are many methods to get distance information and convert them in a 3D representa- tion, in the context of mobile robots the most used approaches are: Time of Flight (TOF): system which estimates the distance from the sensor to a surface measuring the time that an emitted signal takes to hit and return from a target object. The usual technologies are: • Pulsed light source with digital time counters: It consists of a pulsed laser and an imaging integrated circuit for a fast computation of the distance of every pixel. • Range gated imagers: A shutter in front of the camera sensor is opened and closed at the same frequency with which the light pulses are emitted. According to the arrival time, part of the returned light will be blocked by the shutter closing making possible the distinction between far and near objects. • RF-modulated light sources with phase detectors: It works modulating the outgoing beam with a Radio Frequency carrier; the distance is inferred measuring the phase shift of that carrier in the receive side. Knowing the propagation speed of a ray and using precise systems to measure the time of flight, a simple approximation of the distance d is given by: d˘ c¢ t 2 where c is the speed of the ray (the light speed in case of laser sensors) and t is the flying time. On one side, the advantages of this system are the simplicity and the speed; on the other, some of the disadvantages are the multiple reflections, the cost and the disturbance caused by the sun or other sensors of the same type running at the same moment.

Anteprima della Tesi di Mattia Di Gaetano

Anteprima della tesi: Stereo Vision Based Object Detection for Mobile Robots (Stereo Visione per il Riconoscimento di Oggetti nella Robotica Mobile), Pagina 5

Tesi di Laurea Magistrale

Facoltà: Ingegneria

Autore: Mattia Di Gaetano Contatta »

Composta da 93 pagine.

 

Questa tesi ha raggiunto 101 click dal 13/12/2012.

Disponibile in PDF, la consultazione è esclusivamente in formato digitale.