Application Example: Edutainment Game - Introduction to Video and Image Processing - page 167

Fig. 12.6 The effect of applying different threshold values

does is to find a mapping between two different coordinate systems—here between
(x, y) and (x ,y ) . This mapping is known as a homography, see Sect. 10.3. Once
the calibration is done and the mapping function found, this function is then applied
to each found foot before the Game-block assesses whether the player steps on the
right answer or not. Assuming neither the camera nor the projector are moved, the
calibration only has to be done ones. Good practice is, however, to do a calibration
each time the system is started.
The solution to the calibration problem is given in Sect. 10.3. It requires knowl-
edge of the position of the same four points in the two camera systems. SB and
Mick found these four points by placing an infrared light source at each of the four
positions where the four corners of the graphics are projected onto the floor, see
Fig. 12.5 . The positions of these four corners in the (x ,y ) coordinate system are
equal to the (known) size of the projected graphics: ( 0 , 0 ) , ( 0 ,Y MAX ) , (X MAX , 0 ) ,
and (X MAX ,Y MAX ) . The corresponding (x, y) positions of these four corners are
found manually.

12.3 Segmentation

Mick and SB decided to apply a thresholding approach as the first step in segmenting
the feet from the rest of the image. Unfortunately, when they set up the system at the
bartender's house it turned out that the infrared images where not always as nice as
those produced in Mick's living room, see Fig. 12.4 . In fact, the images were heavily
contaminated by noise, see Fig. 12.2 , and choosing a suitable threshold value turned
out to be a delicate matter. On one hand, a low threshold value would segment the
feet but also produce a lot of noise. On the other hand, a high threshold value would
eliminate noise, but also parts of the feet. In the end, a moderate threshold value was
chosen, see Fig. 12.6 . They tried to eliminate the remaining noise by morphology
and/or a median filter, while at the same time thinking about the framerate. None
could do a perfect job and the conclusion was a 7

7 median filter and then remove
the final groups of noise pixels in the Representation sub-block, see Fig. 12.2 .In
Fig. 12.7 the effect of different median filters is illustrated.

×