합성곱 신경망 기반 맨하탄 좌표계 추정
Estimation of Manhattan Coordinate System using Convolutional Neural Network
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 논문에서는 도심 영상에 대해 맨하탄 좌표계를 추정하는 합성곱 신경망(Convolutional Neural Network) 기반의 시스템을제안한다.도심영상에서맨하탄좌표계를추정하는것은영상조정, 3차원장면복원등컴퓨터그래픽스및비전문제해결의기본이 된다. 제안하는 합성곱 신경망은 GoogLeNet[1]을 기반으로 구성한다. 합성곱 신경망을 훈련하기 위해 구글 스트리트뷰 API로 영상을 수집하고 기존 캘리브레이션 방법으로 맨하탄 좌표계를 계산하여 데이터셋을 생성한다. 장면마다 새롭게합성곱신경망을학습해야하는PoseNet[2]과달리,본논문에서제안하는시스템은장면의구조를학습하여맨하탄좌표계를추정하기 때문에 학습되지 않은 새로운 장면에 대해서도 맨하탄 좌표계를 추정한다. 제안하는 방법은 학습에 참여하지 않은구글스트리트뷰영상을검증데이터로테스트하였을때3.157의중간오차로맨하탄좌표계를추정하였다.또한,동일검증데이터에 대해 제안하는 방법이 기존 맨하탄 좌표계 추정 알고리즘[3]보다 더 낮은 중간 오차를 보이는 것을 확인하였다.
Abstract
In this paper, we propose a system which estimates Manhattan coordinate systems for urban scene images using a convolutional neural network (CNN). Estimating the Manhattan coordinate system from an image under the Manhattan world assumption is the basis for solving computer graphics and vision problems such as image adjustment and 3D scene reconstruction. We construct a CNN that estimates Manhattan coordinate systems based on GoogLeNet [1]. To train the CNN, we collect about 155,000 images under the Manhattan world assumption by using the Google Street View APIs and calculate Manhattan coordinate systems using existing calibration methods to generate dataset. In contrast to PoseNet [2] that trains per-scene CNNs, our method learns from images under the Manhattan world assumption and thus estimates Manhattan coordinate systems for new images that have not been learned. Experimental results show that our method estimates Manhattan coordinate systems with the median error of 3.157 for the Google Street View images of non-trained scenes, as test set. In addition, compared to an existing calibration method [3], the proposed method shows lower intermediate errors for the test set.
참고문헌 (14)
- C. Szegedy, "Going deeper with convolutions", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.1~9 (2015)
- A. Kendall, "PoseNet : A convolutional network for real-time 6-dof camera relocalization", Proceedings of the International Conference on Computer Vision, pp.2938~2946 (2015)
- H. Lee, "Automatic upright adjustment of photographs with robust camera calibration", IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(5), pp.833~844 (2014)
- J. M. Coughlan, "Manhattan World: Compass direction from a single image by bayesian inference", Proceedings of the International Conference on Computer Vision, pp.941~947 (1999)
- K. H. Jang, "Practical modeling technique for large-scale 3d building models from ground images", Pattern Recognition Letters, 30(10), pp.861~869 (2009)
- P. Denis, "Efficient edge-based methods for estimating manhattan frames in urban imagery", Proceedings of European Conference on Computer Vision, pp.197~210 (2008)
- B. Li, "Simultaneous vanishing point detection and camera calibration from single images", Proceedings og the International Symposium on Visual Computing, pp.151~160 (2010)
- Tutorial on Gabor filters, "J. R. Movellan", Univ. of California, ~ (2005)
- M. Zhai, "Detecting vanishing points using global image context in a non-Manhattan world", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.5657~5665 (2016)
- C. Wu, "Towards linear-time incremental structure from motion", Proceedings of the International Conference on 3D Vision, pp.127~134 (2013)
- "Google Street View Image API"[URL]
- "TensorFlow: Large-scale machine learning on heterogeneous systems"[URL]
- D. P. Kingma, "Adam : A method for stochastic optimization", Proceedings ot The International Conference on Learning Representations, pp.2938~2946 (2015)
- E. Tretyak, "Geometric image parsing in man-made environments", International Journal of Computer Vision, 97(3), pp.305~321 (2011)