Guided Filter를 이용한 GA-Net Stereo Matching 개선
Improvement of GA-Net Stereo Matching using Guided Filter
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
두 대의 카메라의 양안 시차를 이용해 깊이를 판별하는 stereo matching 분야에서는 최근에 개발되어 가장 효율적인 알고리즘인 GA-Net이 존재한다. 본 논문에서는 Guided filter라는 경계 처리 알고리즘을 이용해 GA-Net에서 얻어낸 결과를 개선하는 방법을 제안한다. GA-Net에서 얻어낸 결과와 원본 스테레오 이미지를 Guided filter로 합성해 세밀한 경계를 이전보다 정확하게 처리할 수 있는 방법을 제시한다. 실험 결과 GA-Net에서 얻어진 결과는 Guided filter를 사용하여 평균 오차가 16.3% 감소하였다. 또한, Guided filter에서 사용자가 설정해야 하는 파라미터를 변화하면서 결과를 측정해 파라미터의 변화에 대한 결과 이미지의 경향성을 분석하였다.
Abstract
In the field of stereo matching, the depth is determined by the disparities between two cameras and GA-Net shows the most accurate performance. In this paper, we improve the depth results from GA-Net using guided filtering as an edge enhancement technique. Depth results from GA-Net and original stereo images are merged to produce more detailed depth images. Experimental results showed that the average error of the depth acquired by guided filter was decreased by 16.3%. We analyzed the depth results by controlling user-defined parameters in guided filtering.In the field of stereo matching, the depth is determined by the disparities between two cameras and GA-Net shows the most accurate performance. In this paper, we improve the depth results from GA-Net using guided filtering as an edge enhancement technique. Depth results from GA-Net and original stereo images are merged to produce more detailed depth images. Experimental results showed that the average error of the depth acquired by guided filter was decreased by 16.3%. We analyzed the depth results by controlling user-defined parameters in guided filtering.
참고문헌 (18)
- 황경조, "2D 및 3D 정보 융합 기술을 이용한 표지판 검출", 한국차세대컴퓨팅학회 논문지, 7(6), pp.61~70 (2011), 한국차세대컴퓨팅학회
- 나민영, "스마트 디바이스 제어를 위한 비전 기반 실시간 손 포즈 및 제스처 인식 방법", 한국차세대컴퓨팅학회 논문지, 8(4), pp.27~34 (2012), 한국차세대컴퓨팅학회
- 임지연, "Global R&D Trend and Challenges in Artificial Intelligence SW", 한국차세대컴퓨팅학회 논문지, 12(4), pp.33~46 (2016), 한국차세대컴퓨팅학회
- H. Hirschmuller, "Stereo processing by semiglobal matching and mutual information", IEEE Transactionson Pattern Analysis and Machine Intelligence, 30(2), pp.328~341 (2008)
- A. Seki, "Sgm-nets: Semiglobal matching with neural networks", IEEE Conference on Computer Vision and Pattern Recognition, pp.6640~6649 (2017)
- Schonberger, Johannes L., "Learning to fuse proposals from multiple scanline optimizations in semiglobal matching", EuropeanConference on Computer Vision (ECCV), pp.739~755 (2018)
- Q. Yang, "A non-local cost aggregation method for stereo matching", IEEE Conference on Computer Vision and Pattern Recognition, pp.1402~1409 (2012)
- N. Mayer, "A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation", IEEE Conferenceon Computer Vision and Pattern Recognition, pp.4040~4048 (2016)
- J. Pang, "Cascade residual learning: A two-stage convolutional neural network for stereo matching", IEEE International Conference. on Computer Vision Workshops, pp.887~895 (2017)
- A. Kendall, "End-to-end learning of geometry and context for deep stereo regression", IEEE International Conference on Computer Vision, pp.66~75 (2017)
- J.-R. Chang, "Pyramid stereo matching network", Conference on Computer Vision and Pattern Recognition, pp.5410~5418 (2018)
- F. Zhang, "Ga-net: Guided aggregation net for end-to-end stereo matching", IEEE Conference on Computer Vision and Pattern Recognition, pp.185~194 (2019)
- K. He, "Guided Image Filtering", IEEE Transactions on Pattern Analysis and Machine Intelligence, 6, pp.1397~1409 (2013)
- N. Mayer, "A large dataset to train convolutional networks for disparity, optical flow and scene flow estimation", IEEE Conference on Computer Vision and Pattern Recognition, pp.483~499 (2016)
- A. Geiger, "Are we ready for autonomous driving? the KITTI vision benchmark suite", IEEE Conference on Computer Vision and Pattern Recognition, pp.3354~3361 (2012)
- "Vision Benchmark Suite"[URL]
- "Middlebury Dataset"[URL]