Kévin Boulanger

Automatic Tour Into the Picture

Automatic Tour Into the Picture (ATIP) is an extension of the Tour Into the Picture method by Youichi Horry et al. that allows an approximative but visually convincing 3D walk-through inside a single image by rendering a box textured using the input image data. The original algorithm requires a long and tedious user interaction to determine the box dimensions and the perspective parameters, and imposes several constraints on the input image orientation. Our method provides automatic and fast camera calibration for any view orientation without using a calibration target nor a tripod. Our method reduces the user interaction, hence only a couple of seconds are required between the input image loading and the final walk-through.

Here are some results (click on the thumbnails to see larger images). The original image is at the top left corner, the three other images are rendered from another point of view using OpenGL.

Input image for the ATIP algorithm Result of the ATIP algorithm
Result of the ATIP algorithm Result of the ATIP algorithm

Input image for the ATIP algorithm Result of the ATIP algorithm
Result of the ATIP algorithm Result of the ATIP algorithm

Input image for the ATIP algorithm Result of the ATIP algorithm
Result of the ATIP algorithm Result of the ATIP algorithm

The software is made of two parts, ATIP maker (demo video here) and ATIP navigator. The navigator has a single purpose: it reads the content of an .atip file created by ATIP maker (containing geometric information and textures) and renders a textured box. The navigator works on PDAs (Personal Digital Assistants) due to the simplicity of the rendering step, as shown in the following image:

ATIP Navigator on a Pocket PC

ATIP Maker uses detection of vanishing points in the input image to calibrate a camera and so is able to define the perspective parameters of the image. However, with this kind of algorithm, lines have to be present in the image to make this detection possible. In our method, the detection is more robust and allows the camera calibration using noisy lines or very imprecise vanishing points (due to hand drawing for example). Here are some example of difficult pictures to analyze (left column) and a result during navigation (right column) after a successfull analysis.

Input image that is difficult to process   Result of the ATIP algorithm   Difficult scene due to noisy lines in the image. There are no straight lines as in architectural scenes.
Input image that is difficult to process   Result of the ATIP algorithm   Difficult scene due to the strong presence of curved lines on the white part of the building. They make the vanishing point coordinates estimation difficult.
Input image that is difficult to process   Result of the ATIP algorithm   Difficult scene due to the imprecise drawing (no ruler was used). The lines are not perfectly straight and they do not intersect in precise points (so the vanishing points estimation is more difficult).

Publications


Downloads


ATIP Maker demo video (5 MB)

If you have any comment, you can contact me at info@kevinboulanger.net