研究目的
Investigating the use of dense optical flow for distortion-robust spherical camera motion estimation to overcome the limitations of conventional sparse feature point tracking methods.
研究成果
The proposed dense optical flow-based approach for spherical camera motion estimation is more stable and robust to strong distortions induced by pitch rotations compared to sparse feature point tracking methods. It provides a reliable consensus of camera motion by utilizing all available information in an image, leading to lower errors and increased robustness to distortion.
研究不足
The method may not work across large translational displacements, which can make it difficult to compute dense optical flow.
1：Experimental Design and Method Selection:
The approach involves decomposing dense optical flow into epipolar geometry and a dense disparity map, followed by reprojection to estimate 6DoF camera motion.
2：Sample Selection and Data Sources:
Images captured using a Ricoh Theta S spherical camera.
3：List of Experimental Equipment and Materials:
Ricoh Theta S spherical camera.
4：Experimental Procedures and Operational Workflow:
Two spherical images are rectified to an equirectangular stereo pair through multiple image rotations via an iterative energy minimization based on equirectangular dense optical flow. The dense disparity map is then utilized to reproject one of the images to a third image to give the full 6DoF estimate.
5：Data Analysis Methods:
The approach is evaluated against a sparse feature descriptor (A-KAZE) in terms of robustness to distortion induced by pitch rotations.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
