TY - GEN
T1 - Adaptive Voxelization Strategy for 3D Object Detection
AU - He, Jyun Hong
AU - Chen, Xiu Zhi
AU - Lin, You Shiuan
AU - Yang, Chen Yu
AU - Chen, Yen Lin
AU - Chiang, Hsin Han
N1 - Funding Information:
This project is financially sponsored by the Ministry of Science and Technology. MOST-109-2628-E-027 -004 -MY3, MOST-110-2218-E-027 -001. This study was also supported by the Ministry of Education of Taiwan under Official Document No. 1100156712, and also jointly developed with Wistron.
Funding Information:
This project is financially sponsored by the Ministry of Science and Technology. MOST- 109-2628-E-027 -004 -MY3, MOST- 110-2218-E-027 -001. This study was also supported by the Ministry of Education of Taiwan under Official Document No. 1100156712, and also jointly developed with Wistron.
Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - 3D object detection techniques rely on features obtained from point cloud data structures to identify and label the frame range of objects. Past techniques of converting point cloud data into voxel grids or image sets would make the data unnecessarily large, and it would be impractical to slice all space into a voxel grid of the same scale for objects of different types and depths. In this paper, the RGB-D depth camera is used to obtain the original point cloud information, and the mature 2D target detection technology and advanced 3D deep learning are used to locate the target. In addition, the voxel grid structure is improved, and the proportion and size of the voxel grid are appropriately adjusted by adopting the method of image category adaptation and spatial depth clustering to obtain more accurate point cloud features and achieve fast and accurate 3D object detection.
AB - 3D object detection techniques rely on features obtained from point cloud data structures to identify and label the frame range of objects. Past techniques of converting point cloud data into voxel grids or image sets would make the data unnecessarily large, and it would be impractical to slice all space into a voxel grid of the same scale for objects of different types and depths. In this paper, the RGB-D depth camera is used to obtain the original point cloud information, and the mature 2D target detection technology and advanced 3D deep learning are used to locate the target. In addition, the voxel grid structure is improved, and the proportion and size of the voxel grid are appropriately adjusted by adopting the method of image category adaptation and spatial depth clustering to obtain more accurate point cloud features and achieve fast and accurate 3D object detection.
UR - http://www.scopus.com/inward/record.url?scp=85138727487&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85138727487&partnerID=8YFLogxK
U2 - 10.1109/ICCE-Taiwan55306.2022.9869047
DO - 10.1109/ICCE-Taiwan55306.2022.9869047
M3 - Conference contribution
AN - SCOPUS:85138727487
T3 - Proceedings - 2022 IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2022
SP - 423
EP - 424
BT - Proceedings - 2022 IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2022
Y2 - 6 July 2022 through 8 July 2022
ER -
