3D Vision for Regular Scenes

3D Vision for Regular Scenes

There is a long and rich history of 3D reconstruction in computer vision . For natural scenes captured by regular cameras, which contains lots of shape features such as corners, edges, it is easy to locate the features and find correspondences between adjacent views. If the camera is fixed, then we can mosaic images together to obtain a panorama. Otherwise if the camera is moving, we can apply stereo or structure from motion to reconstruct the 3D information of the scene. A brief comparison between the different methods in listed below:

1. Multi-view Stereo: Rich texture features are required for establishing correspon- dences. Given both correspondences and calibrated camera motions, depth can be recovered. This method assumes that the lighting is distant and fixed. In other words, the texture features do not depend on the lighting conditions.
   
2. Shape-from-motion: Rich texture features are required for establishing correspon- dences. Given only correspondences, motion and shape structure can be computed by factorization. Same as the above approach, this method assumes that the texture features do not depend on the lighting conditions.
   
3. Shape-from-shading: Constant surface albedos are assumed in this method. Dif- ferent methods have been developed to deal with distant or near-field lighting, and orthogonal or perspective projection camera models. Initial values on occluding boundaries are required as a starting point for this approach.
   8
   

4. Shape-from-silhouette: Given camera motions and image silhouettes, surfaces of 3D object can be reconstructed. This method requires multiple cameras working simultaneously to achieve high resolution reconstruction.
   
5. Photometric stereo: This method reconstructs the surface normal by using several images of the same surface taken from the same viewpoint but under illumination from different directions. In each case there is a well-defined light source direc- tion from which to measure the surface orientation. Therefore, the change of the intensities in the images depends on both local surface orientations and illumination directions. Varying lightings are the key to this solution.
   
6. Shape-from-structured light: By projecting structured patterns onto the surface, the correspondences can be easily established between different views. This approach requires special devices to create structured patterns such as laser projectors.
   

Most of the conventional methods assume distant lighting and require rich features and correspondences, or expensive experimental setups (e.g. structured light). So it is still challenging to obtain an affordable solution for 3D reconstruction from endoscopic images with the existing methods. 

---

版权声明
本博客所有的原创文章，作者皆保留版权。转载必须包含本声明，保持本文完整，并以超链接形式注明作者编程随想和本文原始地址：
https://idreamboats.blogspot.com/2013/03/3d-vision-for-regular-scenes.html

---