Metodologie di apprendimento automatico applicate alla generazione di dati 3D

Il framework in oggetto, è un ambiente ideato con lo scopo di applicare tecniche di Machine Learning (in particolare le Random Forest) alle funzionalità dell'algoritmo di stereo matching SGM (Semi Global Matching), al fine di incrementarne l'accuratezza in versione standard. Scopo della presente tesi è quello di modificare alcune impostazioni di tale framework rendendolo un ambiente che meglio si adatti alla direzionalità delle scanline (introducendo finestre di supporto rettangolari e ortogonali e il training di foreste separate in base alla singola scanline) e ampliarne le funzionalità tramite l'aggiunta di alcune nuove feature, quali la distanza dal più vicino edge direzionale e la distintività calcolate sulle immagini Left della stereo pair e gli edge direzionali sulle mappe di disparità. Il fine ultimo sarà quello di eseguire svariati test sui dataset Middlebury 2014 e KITTI e raccogliere dati che descrivano l'andamento in positivo o negativo delle modifiche effettuate.

Sistema di visione stereo su architettura ZYNQ

Lo scopo della tesi è creare un’architettura in FPGA in grado di ricavare informazioni 3D da una coppia di sensori stereo. La pipeline è stata realizzata utilizzando il System-on-Chip Zynq, che permette una stretta interazione tra la parte hardware realizzata in FPGA e la CPU. Dopo uno studio preliminare degli strumenti hardware e software, è stata realizzata l’architettura base per la scrittura e la lettura di immagini nella memoria DDR dello Zynq. In seguito l’attenzione si è spostata sull’implementazione di algoritmi stereo (rettificazione e stereo matching) su FPGA e nella realizzazione di una pipeline in grado di ricavare accurate mappe di disparità in tempo reale acquisendo le immagini da una camera stereo.

Depth Estimation in Stereo Biomedical Images via Proxy-Supervised Deep Learning

In order to estimate depth through supervised deep learning-based stereo methods, it is necessary to have access to precise ground truth depth data. While the gathering of precise labels is commonly tackled by deploying depth sensors, this is not always a viable solution. For instance, in many applications in the biomedical domain, the choice of sensors capable of sensing depth at small distances with high precision on difficult surfaces (that present non-Lambertian properties) is very limited. It is therefore necessary to find alternative techniques to gather ground truth data without having to rely on external sensors. In this thesis, two different approaches have been tested to produce supervision data for biomedical images. The first aims to obtain input stereo image pairs and disparities through simulation in a virtual environment, while the second relies on a non-learned disparity estimation algorithm in order to produce noisy disparities, which are then filtered by means of hand-crafted confidence measures to create noisy labels for a subset of pixels. Among the two, the second approach, which is referred in literature as proxy-labeling, has shown the best results and has even outperformed the non-learned disparity estimation algorithm used for supervision.

Mitigating non-Lambertian surfaces issues in Stereo Matching with Neural Radiance Fields

Depth estimation from images has long been regarded as a preferable alternative compared to expensive and intrusive active sensors, such as LiDAR and ToF. The topic has attracted the attention of an increasingly wide audience thanks to the great amount of application domains, such as autonomous driving, robotic navigation and 3D reconstruction. Among the various techniques employed for depth estimation, stereo matching is one of the most widespread, owing to its robustness, speed and simplicity in setup. Recent developments has been aided by the abundance of annotated stereo images, which granted to deep learning the opportunity to thrive in a research area where deep networks can reach state-of-the-art sub-pixel precision in most cases. Despite the recent findings, stereo matching still begets many open challenges, two among them being finding pixel correspondences in presence of objects that exhibits a non-Lambertian behaviour and processing high-resolution images. Recently, a novel dataset named Booster, which contains high-resolution stereo pairs featuring a large collection of labeled non-Lambertian objects, has been released. The work shown that training state-of-the-art deep neural network on such data improves the generalization capabilities of these networks also in presence of non-Lambertian surfaces. Regardless being a further step to tackle the aforementioned challenge, Booster includes a rather small number of annotated images, and thus cannot satisfy the intensive training requirements of deep learning. This thesis work aims to investigate novel view synthesis techniques to augment the Booster dataset, with ultimate goal of improving stereo matching reliability in presence of high-resolution images that displays non-Lambertian surfaces.