[11] semi-dense depth maps by applying the direct image

11 presents the method of depth map based to register the
input images to depth map. This method is to synthesize the new viewpoint
images. Besides, this method can further accelerated by quad tree decomposition
and view-independent visibility priority.

Applying Markov
Random Field (MRF) into infer plane parameters can helps to identify the
position and orientation of triangular facets of image 12. Finally, this approach construct a 3D wire-frame
scene model.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

13 proposed s system named Free-viewpoint Television
(FTV) to utilize images took by camera arrays. The images of different
viewpoints can be generated based on FTV system for single objects.

14 introduced a direct monocular Simultaneous
Localization and Mapping (SLAM) algorithm. SLAM builds the consistent maps of
the large-scale environment. Pose-graph of key frames is reconstructed with
associated semi-dense depth maps by applying the direct image alignment.

An automatic
system is proposed by 15 in order to build a visual model from images which is
one of the pioneers fully automatic structure from motion of urban environment.
This paper considered the uncalibrated image sequences acquired with a
hand-held camera and is according to the features that matched with the
multiple views.  

            16 applied the approach of image-based reconstruction to
provide the combined sparse-dense method in the work of city reconstruction form
the unstructured image collections. Figure 2.3 demonstrates the example of
dense reconstruction after depth map fusion. A knowledge database is then
transferred to the target of the incrementally building a virtual
representation of a local surrounding.

From the images
taken along the street, an automatic approach is proposed to compose the
street-side photo-realistic 3D models 17. Multi view segmentation algorithm is proposed to
recognize segments each image at pixel level into semantically classes as
depicted in Figure 2.4. The classes include ground, sky, building and so on.
The proposed system splits the buildings into independent blocks and analyze
the structure of each block by using the prior of building regularity.