TY - GEN
T1 - Visual Localization Based on Deep Learning - Take Southern Branch of the National Palace Museum for Example
AU - Tu, Chia Hao
AU - Lu, Eric Hsueh Chan
N1 - Funding Information:
Acknowledgement. This research was supported by Ministry of Science and Technology, Taiwan, R.O.C. under grant no. MOST 109–2121–M–006–013–MY2 and MOST 109–2121–M–006–005.
Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - Visual localization uses images to regress camera position and orientation. It has many applications in computer vision such as autonomous driving, augmented reality (AR) and virtual reality (VR), and so on. The convolutional neural network simulates biological vision and has a good image feature extraction ability, so using it in visual localization can improve regression accuracy. Although our team has built an image indoor localization model for Southern Branch of the National Palace Museum, this paper tries to use new network and loss function to achieve better positioning accuracy. In this paper, we use ResNet-50 as backbone network, and change the output layer to 3-dimensional position and 4-dimensional orientation quaternion, and use learnable weights loss function. We compare different pretrained models and normalization methods, and the best result improves the positioning accuracy by about 60%.
AB - Visual localization uses images to regress camera position and orientation. It has many applications in computer vision such as autonomous driving, augmented reality (AR) and virtual reality (VR), and so on. The convolutional neural network simulates biological vision and has a good image feature extraction ability, so using it in visual localization can improve regression accuracy. Although our team has built an image indoor localization model for Southern Branch of the National Palace Museum, this paper tries to use new network and loss function to achieve better positioning accuracy. In this paper, we use ResNet-50 as backbone network, and change the output layer to 3-dimensional position and 4-dimensional orientation quaternion, and use learnable weights loss function. We compare different pretrained models and normalization methods, and the best result improves the positioning accuracy by about 60%.
UR - http://www.scopus.com/inward/record.url?scp=85145183472&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85145183472&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-21743-2_4
DO - 10.1007/978-3-031-21743-2_4
M3 - Conference contribution
AN - SCOPUS:85145183472
SN - 9783031217425
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 39
EP - 50
BT - Intelligent Information and Database Systems - 14th Asian Conference, ACIIDS 2022, Proceedings
A2 - Nguyen, Ngoc Thanh
A2 - Trawiński, Bogdan
A2 - Nguyen, Ngoc Thanh
A2 - Tran, Tien Khoa
A2 - Tukayev, Ualsher
A2 - Hong, Tzung-Pei
A2 - Szczerbicki, Edward
PB - Springer Science and Business Media Deutschland GmbH
T2 - 14th Asian Conference on Intelligent Information and Database Systems , ACIIDS 2022
Y2 - 28 November 2022 through 30 November 2022
ER -