Cited 0 time in webofscience Cited 0 time in scopus

Driver Gaze Estimation on Specific Zones: Effective Training with Limited Training Samples

Title
Driver Gaze Estimation on Specific Zones: Effective Training with Limited Training Samples
Authors
Tayibnapis, Iman RahmansyahPark, JaehyeongLee, Jin HeeKwon, Soon
DGIST Authors
Park, Jaehyeong; Lee, Jin Hee; Kwon, Soon
Issue Date
2020-01-21
Citation
ICEIC(International Conference on Electronics, Information, and Communication) 2020
Type
Conference
Abstract
One of the important parts in the advanced driver assistant system (ADAS) is monitoring driver condition. In monitoring driver condition, driver gaze estimation is one of vital parts. In order to train the gaze zone estimator as a classification task, a high cost must be paid to gather a large annotated dataset. To reduce the effort, we used a transfer-learning method using pre-trained convolution neural network (CNN) model to project the gaze estimation task having large and reliable dataset into new classification task to overcome lack of annotated dataset for gaze zone estimation. We project the gaze estimation task on mobile devices using regression. Furthermore, we added a spatial weights CNN and showed the effect of the addition on the performance. We tested the proposed method on our building simulation test bed. The result showed an accuracy of 79.8~87% in case of an unseen driver for estimating 10 zones in-vehicle
URI
http://hdl.handle.net/20.500.11750/14079
Publisher
IEIE
Related Researcher
  • Author Kwon, Soon  
  • Research Interests computer vision; deep learning; autonomous driving; parallel processing; vision system on chip
Files:
There are no files associated with this item.
Collection:
Division of Automotive Technology2. Conference Papers


qrcode mendeley

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE