研究目的
To enhance the ToF sensor to estimate depth of specular surfaces by combining absolute depth (ToF) and relative shape with polarization cues (SfP).
研究成果
The proposed depth reconstruction framework for specular objects that combines few absolute depth cues and relative shapes with SfP can precisely reconstruct depth values of specular objects in the controlled environment and demonstrates adequate reconstruction to a dense depth image for real car in the outdoor environment.
研究不足
The method assumes that one object has the same refractive index for the whole object and that depth for parts of the specular object can be obtained. The evaluation dataset needs to be extended to adapt various objects which have different material property as well as moving camera.
1：Experimental Design and Method Selection:
The framework combines ToF depth cues and SfP for depth reconstruction of specular objects. Superpixel segmentation with planarity constraints is used to overcome the ill-posedness of SfP with a single view.
2：Sample Selection and Data Sources:
Datasets were recorded in different environments – a controlled indoor environment and a real outdoor environment.
3：List of Experimental Equipment and Materials:
A single ToF sensor (TEDTB-7Z-TCDK-GC2) with a linear polarizer (Thorlabs LPNIRE100B) was used.
4：Experimental Procedures and Operational Workflow:
The angles of the linear polarizer were set to 0°, 45°, 90°, 135°. The method involves correcting π-ambiguity for azimuth angles, estimating refractive index with ToF measurements, and reconstructing absolute depth from SfP normals.
5：5°. The method involves correcting π-ambiguity for azimuth angles, estimating refractive index with ToF measurements, and reconstructing absolute depth from SfP normals. Data Analysis Methods:
5. Data Analysis Methods: The reconstruction error was compared against the method of Levin et al. using root mean squared error (RMSE) in meter as metric.
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
