Wu, Jing (2021) Lifelong 3D object recognition: A comparison of deep features and handcrafted descriptors. Bachelor's Thesis, Artificial Intelligence.
|
Text
Bachelor Thesis JJ Wu s2655012.pdf Download (1MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (123kB) |
Abstract
Over the past few decades we have been finding more and more uses for service robots. While the early robots that were employed by humans were often stationary and operating in static environments, nowadays the demand for service robots capable of more complex behaviour in dynamic environments has been increasing. Object recognition has come a long way the past decade with the improvements in software and hardware allowing widespread implementation using Convolutional Neural Networks (CNN). Unfortunately, the batch training that is required of CNNs is hard to employ in a system that has to be capable of lifelong learning while operating in a dynamic open-ended domain at the same time. Using OrthographicNet, a system designed specifically towards the purpose of achieving lifelong 3D object recognition in open-ended domains, we aim to compare the performances of deep features (MobileNetV2, VGG19_fc1) and handcrafted descriptors (GOOD, ESF). We performed offline experiments for our preliminary comparisons and found during the online testing phase that the deep feature descriptor MobileNetV2 is capable of learning the most object categories whilst providing decent accuracy and learning speed.
| 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: | 15 Jun 2021 14:57 |
| Last Modified: | 15 Jun 2021 14:57 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/24576 |
Actions (login required)
 |
View Item |