Javascript must be enabled for the correct page display

MORE: Simultaneous Multi-View 3D Object Recognition and Pose Estimation

Parisotto, Tommaso (2021) MORE: Simultaneous Multi-View 3D Object Recognition and Pose Estimation. Master's Thesis / Essay, Artificial Intelligence.


Download (3MB) | Preview
[img] Text
Restricted to Registered users only

Download (493kB)


In computer vision the problem of 3D object recognition has been tackled by many successful algorithms. In particular, in recent times, convolutional neural networks based on multi-view approaches figure among the best performing classifiers on 3D object datasets such as Princeton ModelNet, constituting de facto state-of-the-art. Although the research on the architectures for such networks has been widely explored, there has been little investigation on the benefits of employing best-views estimation algorithms to select the views for prediction. Many objects have regular 3D shapes which carry symmetrical features and sets of projections of such objects carry redundancy of information which can be reduced selecting only very informative views. To demonstrate the employability of best-view selection algorithms, we propose an entropy estimation model to retrieve best-views for a multi-view convolutional neural network to perform simultaneous recognition and pose estimation on 3D objects. We demonstrated that our model learns features to generate entropy maps that approximate closely the entropy evaluation of depth-images projections of a 3D object. With such evaluation we select a small number of highly-informative camera poses to observe the object in space. We demonstrated that the views obtained from such positions are descriptive enough to achieve accuracy scores comparable to other state-of-the-art approaches on the ModelNet10 dataset.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Mohades Kasaei, S.H. and Wiering, M.A.
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 19 Mar 2021 12:35
Last Modified: 19 Mar 2021 12:35

Actions (login required)

View Item View Item