Hand gesture recognition with generalized hough transform and DC-CNN using Realsense

Bo Liao, Jing Li, Zhaojie Ju, Gaoxiang Ouyang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

526 Downloads (Pure)

Abstract

Hand gesture recognition plays an important role in human-computer interaction. With the development of depth cameras, color images combined with depth images can provide richer information for hand gesture recognition. In this paper, we propose a hand gesture recognition system based on the data captured by Intel RealSense Front-Facing Camera SR300. Considering that the pixels in depth images collected by RealSense are not one-to-one to those in color images, the recognition system maps depth images to color images based on generalized Hough transform in order to segment hand from a complex background in color images using the depth information. Then, it recognizes different hand gestures by a novel double-channel convolutional neural network containing two input channels which are color images and depth images. Moreover, we built a hand gesture database of 24 different kinds of hand gestures representing 24 letters in the English alphabet. It contains a total of 168,000 images which are 84,000 RGB images and 84,000 depth images. Experimental results on our newly collected hand gesture database demonstrate the effectiveness of the proposed approach, and the recognition accuracy is 99.4%.
Original languageEnglish
Title of host publication8th International Conference on Information Science and Technology
Subtitle of host publicationICIST 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages84-90
ISBN (Electronic)978-1-5386-3782-1
ISBN (Print)978-1-5386-3783-8
DOIs
Publication statusPublished - 9 Aug 2018
Event2018 Eighth International Conference on Information Science and Technology - Cordoba, Spain
Duration: 30 Jun 20186 Jul 2018

Publication series

NameIEEE ICIST Proceedings Series
ISSN (Electronic)2573-3311

Conference

Conference2018 Eighth International Conference on Information Science and Technology
Country/TerritorySpain
Period30/06/186/07/18

Fingerprint

Dive into the research topics of 'Hand gesture recognition with generalized hough transform and DC-CNN using Realsense'. Together they form a unique fingerprint.

Cite this