Designing Vision Systems that See Better - Advances in Embedded Computer Vision - page 264

Fig. 13.4 CASSI-based spectral classification in the experiment: a photo of CASSI system, b
photo of objects, c CASSI measurement, d Channel 12, e Channel 18, and f ACE scores of 1, 2,
14, 7, and 425 for the “G”, “P”, “?”, “O”, and “Y” letters by referencing the “Y” letter's spectrum.
Note: the color range is shown as a score of spectral similarity
In the reconstruction, the datacube was decoded by using the measurement model
of the computational design and the TV-based TwIST [ 13 ] was used to search the
solution with parameters of 20 spectral channels and 200 iterations. Note 20 frames
were used to reconstruct the datacube from CASSI measurements. In Fig. 13.4 f, the
ACE scores are visually shown to quantify the performance of CASSI-based spectral
classification and the ACE scores are obtained by using ACE algorithm on the CASSI
reconstruction. In the detection, some partial pixels of the “Y” letter were referenced
to detect the rest of the “Y” letter. The true positive score was returned as 425 and
the false positive score is 24, resulting in the ratio of 18. In the result, the pixels with
the target spectrum are highly scored keeping the other pixels relatively low. Thus,
the CASSI-based spectral classification effectively detects and classifies the desired
pixels with a single frame in the scene.
13.3 Motion Compensation Coded Exposure Camera
In a traditional single-exposure camera system, the camera shutter opens and closes
for a fixed amount of time. This means that the corresponding exposure time is
static, and fast moving objects or camera motion can cause motion blur. This is
because the exposure time defines a temporal period where light from a moving