Javascript must be enabled for the correct page display

Open-Ended 3D Object Recognition: Investigating the Effect of Deep Features, Color Spaces, and Similarity Measures

Țoca, Andreea-Mihaela (2020) Open-Ended 3D Object Recognition: Investigating the Effect of Deep Features, Color Spaces, and Similarity Measures. Bachelor's Thesis, Artificial Intelligence.

[img]
Preview
Text
AI_BA_2020_ANDREEAMIHAELATOCA.pdf

Download (9MB) | Preview
[img] Text
Toestemming.pdf
Restricted to Registered users only

Download (97kB)

Abstract

Object recognition is a challenging task in unpredictable environments, where service robots need to react fast, and cooperate with human-users. Lack of training data and limited use of object information, such as only shape information, reduces the robustness of object recognition algorithms. This study investigates the influences of shape information, color spaces, and similarity measures in open-ended 3D object recognition. Towards this end, three experimental setups, color-only, shape-only, and color-shape, were evaluated in both offline and online setups. Following the OrtohraphicNet construction, the experiments were conducted using two deep learning networks, Mobilenet-v2, and VGG16 in combination with four parameters: orthographic image resolution, similarity distance functions, k-values (in a KNN algorithm), and three color spaces. In the online evaluation, extensive experiments showed that the k-nearest neighbor algorithm had a neglectable influence over the system's performance. It was also observed that the color information improves the performance over shape information alone, color-only RGB obtaining the best result. The online evaluation showed a decrease in accuracy compared with the offline results, predicting the lack of robustness of classical train-test experiments in open-ended domains. The results' hierarchy from offline followed in the online evaluation as well.

Item Type: Thesis (Bachelor's Thesis)
Supervisor name: Mohades Kasaei, S.H.
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 13 Aug 2020 08:26
Last Modified: 13 Aug 2020 08:26
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/23079

Actions (login required)

View Item View Item