모바일 스테레오 비전 시스템을 위한 다양한 스테레오 정합기법의 오차율 비교 = Comparison of error rates of various stereo matching methods for mobile stereo vision systems

다국어 초록 (Multilingual Abstract)

In this paper, the matching error rates of modified area-based, energy-based algorithms, and learning-basedstructures were compared for stereo image matching. Census transform (CT) based on region and life propagation(BP) algorithm based on energy were selected, respectively.Existing algorithms have been improved and implementedin an embedded processor environment so that they can be used for stereo image matching in mobile systems.
Even in the case of the learning base to be compared, a neural network structure that utilizes small-scale parameterswas adopted. To compare the error rates of the three matching methods, Middlebury’s Tsukuba was selected as atest image and subdivided into non-occlusion, discontinuous, and disparity error rates for accurate comparison. Asa result of the experiment, the error rate of modified CT matching improved by about 11% when compared withthe existing algorithm. BP matching was about 87% better than conventional CT in the error rate. Compared to thelearning base using neural networks, BP matching was about 31% superior.

국문 초록 (Abstract)

본 논문에서는 스테레오 영상정합을 위하여 개선된 영역기반, 에너지 기반 알고리즘, 학습기반 구조의 정합 오류율을 비교하였다.
영역기반으로 census transform(CT), 에너지 기반으로 belief propagation(BP) 알고리즘을 선정하였다. 기존 알고리즘을 개선하고 모바일 시스템에서 스테레오 영상정합에 활용가능 하도록 임베디드 프로세서 환경에서 구현하였다. 비교 대상이 되는 학습기반의 경우에 도 적은 규모의 파라메터를 활용하는 신경망 구조를 채택하였다. 세 가지 정합방법의 오류율 비교를 위해 테스트 이미지로 Middlebury 데이터 세트 가운데 Tsukuba를 선정하고 정합 성능의 정확한 비교를 위해 비폐색, 불연속, 시차 오류율 등으로 세분화하였다. 실험 결과 CT 매칭의 오차율은 기존 알고리즘과 수정된 알고리즘으로 비교하였을 때 약 11% 성능 개선되었다. BP 매칭은 오류율에서 기존 CT 에 비하여 약 87% 우수하였다. 신경망을 이용한 학습기반과 비교 하였을 때 BP 매칭이 약 31% 우수함을보였다.

참고문헌 (Reference)

1 Ekaterina Mezenceva, "The study of the Semi-Global Block Matching Algorithm Implementing Parallel Calculation with GPU" 2021

2 M. B. Hisham, "Template Matching using Sum of Squared Difference and Normalized Cross Correlation" 2015

3 J. Zbontar, "Stereo matching by training a convolutional neural network to compare image patches" 17 (17): 2-, 2016

4 Lucas de Paula Veronese, "Stereo Matching with VG-RAM Weightless Neural Networks" 27-29, 2012

5 Jure Zbontar, "Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches" 17 (17): 2287-2318, 2016

6 Martull Sarah, "Realistic CG Stereo Image Dataset With Ground Truth Disparity Maps" 2012

7 I. Aleksander, "RAM-Based Neural Networks" World Scientific 18-30, 1998

8 Jong il Park, "Parallel Implementation of Improved Census Transform Stereo Matching algorithm with Open-MP" 35 (35): 2012

9 Zhao wei, "Panoramic image matching based on graph cuts algorithm" 2012

10 Ramin Zabih, "Non-parmetric local transforms for computing visual correspondence" 801 : 151-158, 2005

11 Qian. Zhang, "New Stereo Matching Method Based on Improved BP Algorithm" Walter de Gruyter GmbH 8 (8): 464-479, 2015

12 "Middlebury Stereo Datasets"

13 Haesol Park, "Look Wider to Match Image Patches With Convolutional Neural Networks" Institute of Electrical and Electronics Engineers (IEEE) 24 (24): 1788-1792, 2017

14 "KITTI Stereo Data Set"

15 J. N. Sarvaiya, "Image Registration by Template Matching Using Normalized Cross-Correlation" 2009

16 Aditya Rachman Putra, "FPGA implementation of template matching using binary Sum of Absolute Difference" 2016

17 Xiaoqing Ye, "Efficient Stereo Matching Leveraging Deep Local and Context Information" Institute of Electrical and Electronics Engineers (IEEE) 5 : 18745-18755, 2017

18 Chunsheng Guo, "Belief propagation algorithm for background estimation based on local maximum weight matching" 2012

19 Hamid Laga, "A Survey on Deep Learning Techniques for Stereo-Based Depth Estimation" Institute of Electrical and Electronics Engineers (IEEE) 44 (44): 1738-1764, 2022