Is that my hand? An egocentric dataset for hand disambiguation
Research output: Journal Publications and Reviews › RGC 21 - Publication in refereed journal › peer-review
Author(s)
Related Research Unit(s)
Detail(s)
Original language | English |
---|---|
Pages (from-to) | 131-143 |
Journal / Publication | Image and Vision Computing |
Volume | 89 |
Online published | 5 Jul 2019 |
Publication status | Published - Sept 2019 |
Link(s)
DOI | DOI |
---|---|
Attachment(s) | Documents
Publisher's Copyright Statement
|
Link to Scopus | https://www.scopus.com/record/display.uri?eid=2-s2.0-85069947714&origin=recordpage |
Permanent Link | https://scholars.cityu.edu.hk/en/publications/publication(f4131288-ae1c-47b6-a061-9fef27a2f81b).html |
Abstract
With the recent development of wearable cameras, the interest for research on the egocentric perspective is increasing. this opens the possibility to work on a specific object detection problem of hand detection and hand disambiguation. However, recent progress in egocentric hand disambiguation and even hand detection, especially using deep learning, has been limited by the lack of a large dataset, with suitable variations in subject, activity, and scene. In this paper, we propose a dataset that simulates daily activities, with variable illumination and people from different cultures and ethnicity to address daily life conditions. We increase the dataset size from previous works to allow robust solutions like deep neural networks that need a substantial amount of data for training. Our dataset consists of 50,000 annotated images with 10 different subjects doing 5 different daily activities (biking, eating, kitchen, office and running) in over 40 different scenes with variable illumination and changing backgrounds, and we compare with previous similar dataset.
Hands in an egocentric view are challenging to detect due to a number of factors, such as shape variations, inconsistent illumination, motion blur, and occlusion. To improve hand detection and disambiguation, context information can be included to aid in the detection. In particular, we propose three neural network architectures that jointly learn the hand and context information, and we provide baseline results with current object/hand detection approaches.
Hands in an egocentric view are challenging to detect due to a number of factors, such as shape variations, inconsistent illumination, motion blur, and occlusion. To improve hand detection and disambiguation, context information can be included to aid in the detection. In particular, we propose three neural network architectures that jointly learn the hand and context information, and we provide baseline results with current object/hand detection approaches.
Research Area(s)
- Egocentric perspective, Hand detection
Bibliographic Note
Research Unit(s) information for this publication is provided by the author(s) concerned.
Citation Format(s)
Is that my hand? An egocentric dataset for hand disambiguation. / Cruz, Sergio; Chan, Antoni.
In: Image and Vision Computing, Vol. 89, 09.2019, p. 131-143.
In: Image and Vision Computing, Vol. 89, 09.2019, p. 131-143.
Research output: Journal Publications and Reviews › RGC 21 - Publication in refereed journal › peer-review
Download Statistics
No data available