A Fast Epipolar Line Matching Method Based on 3D Spherical Panorama

Shuai Liu,Min Sun,Jun Chen,Xiang Wei,Lingli Zhao

doi:10.1109/access.2019.2961184

Abstract

A fast epipolar line matching method based on 3D spherical panorama is proposed in this paper in order to overcome the low automation degree of feature matching in applying the 3D panoramic model measurement. First, the search scope of panoramic image matching was restricted from 2D to 1D strip-shaped buffer zone through the epipolar constraints. The search scope was further refined using color invariant correlation coefficient. Finally, a matching calculation model was constructed based on features to complete the panoramic feature matching. The experiment verified the effectiveness and feasibility of the proposed method in reducing the mismatching of spherical panoramic features. The proposed method also effectively solved the automatic search for corresponding points and resolved geometric information. The findings help lay a solid foundation for the resampling and measurement of the panoramic measurement model and the generation of depth maps.

Highlights

Image matching is a core technology that realizes automatic search for corresponding points and the automatic resolution of geometric information in digital photogrammetry and computer vision
The current panoramic image matching work was completed by establishing constraint conditions based on initial matching, including vehicle-mounted system Global Positioning System/Inertial Measurement Unit (GPS/IMU) data [10]–[12], epipolar line, optical flow, scale, sky point and other multi-combinational constraint conditions [4], or kinematic relation satisfying some conditions and features [11], or matching through accurate camera calibration and extraction of connection points [12], [19]–[24]
The current study focuses on completing the matching work by establishing the epipolar constraints using existing 3D image parameters after the panoramic stereo model is constructed

Summary

A Fast Epipolar Line Matching Method Based on 3D Spherical Panorama

This work was supported in part by the National Natural Science Foundation of China under Grant 41761079, in part by the Universities Joint Special Foundation of Yunnan Provincial Science and Technology Department in China under Grant 2018FH001-046, Grant 2018FH001-056, and Grant 2017FH001-06, in part by the key Projects of the Universities Joint Special Foundation of Yunnan Provincial Science and Technology Department in China ‘‘Study on the estimation method of pomegranate yield based on machine learning geometric feature panoramic image pattern recognition’’, in part by the Young and Middle-Aged Academic and Technical Leader Reserve Project of Yunnan Province in China under Grant 201905C160014, and in part by the Top Young Talent Project of Yunnan Province in China.

INTRODUCTION

GEOMETRIC MAPPING RELATIONS OF EPIPOLAR LINES UNDER SPHERICAL PROJECTION

EXPERIMENT AND DISCUSSION

Findings

CONCLUSION