Please use this identifier to cite or link to this item:
https://scholarbank.nus.edu.sg/handle/10635/151409
Title: | PATTERN RECOGNITION APPROACHES FOR EVENT-BASED VISION | Authors: | ROHAN GHOSH | Keywords: | event-based vision, pattern recognition, action recognition, object recognition, feature estimation, feature tracking | Issue Date: | 20-Aug-2018 | Citation: | ROHAN GHOSH (2018-08-20). PATTERN RECOGNITION APPROACHES FOR EVENT-BASED VISION. ScholarBank@NUS Repository. | Abstract: | Abstract: This thesis is a compilation of approaches aimed at event-based pattern recognition applications,such as object recognition, feature tracking and gesture recognition. For object recognition, I document two different ways of realizing a classifier (CNN and slow-ELM) for event-based data streams. Focus is on achieving invariance to numerous factors of variation, some of which can only be associated with event-based vision. In both cases, good performance is demonstrated, in response to different variations. Next, a spatiotemporal feature learning algorithm is proposed. By inculcating spatiotemporal slowness, features are learnt directly from raw spike-event data, which are robust and invariant to visual transformations, such as translation, spatial scaling and rotation. The applicability of such a feature learning method to the problem of feature tracking is demonstrated. Finally, using the same principles, a two-stage deep learning system is proposed for spatiotemporal classification problems. The system outperforms the state-of-the-art in event-based gesture recognition. | URI: | http://scholarbank.nus.edu.sg/handle/10635/151409 |
Appears in Collections: | Ph.D Theses (Open) |
Show full item record
Files in This Item:
File | Description | Size | Format | Access Settings | Version | |
---|---|---|---|---|---|---|
GhoshR.pdf | 4.54 MB | Adobe PDF | OPEN | None | View/Download |
Google ScholarTM
Check
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.