ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper Improvement of Pose Estimation using Integrated NMS with Rotated Images
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Dohun Kim, Wonjong Kim
Issue Date
2023-06
Citation
International Technical Conference on Circuits/Systems, Computers and Communications (ITC-CSCC) 2023, pp.718-721
Publisher
IEEE
Language
English
Type
Conference Paper
Abstract
Pose estimation is a challenging problem in computer vision, which involves estimating the position and orientation of objects in an image. Non-Maximum Suppression (NMS) is a widely used technique to suppress multiple detections of the same object in an image. However, in some cases, NMS can lead to a loss of accurate pose estimation. In this paper, we propose an integrated NMS approach with rotated images to improve pose estimation accuracy. Our approach involves rotating the image and applying NMS to the rotated image to suppress multiple detections. We then rotate the suppressed detections back to the original image and estimate the pose of the object. We evaluate our proposed approach on MPII benchmark datasets, and the results show that our approach improves 8% in mAP compared to YOLO V7
KSP Keywords
Benchmark datasets, Computer Vision(CV), Estimation accuracy, Non-maximum suppression, Orientation of objects, Pose estimation, Position and orientation