TY - GEN
T1 - An implementation of an affine BRISK for mobile heterogeneous parallel processors
AU - Lee, Chulhee
AU - Rhee, Chae Eun
AU - Lee, Hyuk Jae
PY - 2016/3/10
Y1 - 2016/3/10
N2 - As the computing power of a mobile processor increases, a number of computer vision applications are widely used by a mobile device. Feature extraction is one of the crucial tasks in a computer vision area and it is used for object recognition, panorama image generation and so on. The affine SIFT (ASIFT) feature is robust to scale, rotation and viewpoint changes but its high computational complexity makes it challenging for ASIFT to be used in applications by a mobile device. To reduce the complexity of ASIFT while maintaining affine invariance, this paper proposes an affine BRISK which replaces SIFT by BRISK of which complexity is much less than SIFT. By fully exploiting the parallelism of affine transformation for viewpoint invariance, this paper achieves the speed up by parallel execution of feature extraction and affine transformation running on the mobile CPU and GPU, respectively. The proposed affine BRISK feature is robust to a viewpoint change with a comparable accuracy. The processing time is 290.64ms on a Galaxy LTE-A mobile.
AB - As the computing power of a mobile processor increases, a number of computer vision applications are widely used by a mobile device. Feature extraction is one of the crucial tasks in a computer vision area and it is used for object recognition, panorama image generation and so on. The affine SIFT (ASIFT) feature is robust to scale, rotation and viewpoint changes but its high computational complexity makes it challenging for ASIFT to be used in applications by a mobile device. To reduce the complexity of ASIFT while maintaining affine invariance, this paper proposes an affine BRISK which replaces SIFT by BRISK of which complexity is much less than SIFT. By fully exploiting the parallelism of affine transformation for viewpoint invariance, this paper achieves the speed up by parallel execution of feature extraction and affine transformation running on the mobile CPU and GPU, respectively. The proposed affine BRISK feature is robust to a viewpoint change with a comparable accuracy. The processing time is 290.64ms on a Galaxy LTE-A mobile.
UR - http://www.scopus.com/inward/record.url?scp=84965122900&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84965122900&partnerID=8YFLogxK
U2 - 10.1109/ICCE.2016.7430680
DO - 10.1109/ICCE.2016.7430680
M3 - Conference contribution
AN - SCOPUS:84965122900
T3 - 2016 IEEE International Conference on Consumer Electronics, ICCE 2016
SP - 435
EP - 436
BT - 2016 IEEE International Conference on Consumer Electronics, ICCE 2016
A2 - Bellido, Francisco J.
A2 - Diaz-Sanchez, Daniel
A2 - Vun, Nicholas C. H.
A2 - Dolar, Carsten
A2 - Ling, Wing-Kuen
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - IEEE International Conference on Consumer Electronics, ICCE 2016
Y2 - 7 January 2016 through 11 January 2016
ER -