This thesis is placed in the context of Advanced Driver Assistance Systems (ADAS), which are systems designed to improve everyday driving experience and safety. In particular, our work addresses the visual aid topic, which allows drivers to have a clear view of their surroundings thanks to cameras installed on the vehicle. The aim of the thesis is to study and propose solutions for the issues a icting current commercial systems. What these systems typically do is merging together images coming from di erent cameras into a single, bird-eye view image, which is then shown to the driver. By doing this, several artifacts will be visible in the overlapping areas between cameras. The main idea behind our study was to reduce and possibly remove image artifacts by exploiting 3D information acquired from the surrounding scene. We propose one rst approach, based on point cloud surface reconstruction, which we found out to not be adequate to solve some of the issues we wanted to solve. Therefore, we propose a second approach, based on a mesh representation of the scene, which instead yields appreciable results.
Questa tesi si inserisce nel contesto degli Advanced Driver Assistance Systems (ADAS), i quali sono sistemi sviluppati per migliorare l' esperienza di guida quotidiana e la sicurezza. In particolare, il nostro lavoro a ronta l' argomento dell' aiuto visivo (visual aid), il quale aiuta il guidatore ad avere una visione chiara dell' ambiente circostante. Lo scopo della tesi e di studiare e proporre soluzioni per i problemi che a iggono gli attuali sistemi commerciali. Ci o che questi sistemi tipicamente fanno e fondere insieme immagini provenienti da telecamere di erenti in un' unica immagine con vista a volo d' uccello, che viene quindi mostrata al guidatore. Facendo ci o, diversi difetti saranno visibili nelle zone di sovrapposizione fra le telecamere. L' idea principale dietro il nostro studio e quella di ridurre e possibilmente elimare i difetti visivi sfruttanto l' informazione 3D acquisita dalla scena circostante. Proponiamo un primo approccio, basato sulla ricostruzione super ciale a partire da nuvole di punti, che abbiamo scoperto non essere adeguato a risolvere alcune delle problematiche che volevamo risolvere. Di conseguenza, proponiamo un secondo algoritmo, basato sulla rappresentazione in forma di mesh della scena, che invece d a risultati apprezzabili.
Improving an ADAS surrounding visual aid
VELO, ALBI
2014/2015
Abstract
This thesis is placed in the context of Advanced Driver Assistance Systems (ADAS), which are systems designed to improve everyday driving experience and safety. In particular, our work addresses the visual aid topic, which allows drivers to have a clear view of their surroundings thanks to cameras installed on the vehicle. The aim of the thesis is to study and propose solutions for the issues a icting current commercial systems. What these systems typically do is merging together images coming from di erent cameras into a single, bird-eye view image, which is then shown to the driver. By doing this, several artifacts will be visible in the overlapping areas between cameras. The main idea behind our study was to reduce and possibly remove image artifacts by exploiting 3D information acquired from the surrounding scene. We propose one rst approach, based on point cloud surface reconstruction, which we found out to not be adequate to solve some of the issues we wanted to solve. Therefore, we propose a second approach, based on a mesh representation of the scene, which instead yields appreciable results.| File | Dimensione | Formato | |
|---|---|---|---|
|
tesi.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
45.68 MB
Formato
Adobe PDF
|
45.68 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/112122