Pinhole Camera:
Image I(x) = T(x)*S(x), T(x): [0,1],aperture function: light through pinhole; S(x): image formed by a pinhole camera
Shape image = Small aperture T(x)
Night: various illumination sources, I(x): superimposed copies of the window shape
Daylight: diffuse illumination, I(x): convolution of outdoor scene and window shape (very blurry)
Pinspeck (Inverse Pinhole) Camera
I_window(x)-I_occludewindow(x) = T_hole(x)*S(x)
Limitations:
1. To extract background, it requires at least two images or a video.
2. To increase signal to noise ratio:
Pinhole camera requires a more sensitive light sensor to capture as more as possible lights.
Pinspeck camera requires increased illumination of background with respect to the lights blocked by occludes.
Calibration:
Distortion: orientation between camera and project surface.
Solution: find homography between wall(project surface, usually wall) and camera.
Applications:
1. Seeing what is outside the room
2. Seeing light sources
3. Seeing the shape of the window