morphing for 12view animal scan
proposed by Prof.Satoshi Nishimura / Jichi Med Univ(JAPAN)

---

hardware set up 1

security cameras
960x480pix x 4cameras x2set (total 8camera)


painted black


3d mounted, camera,

---

result by set up 1


captured images


result movie

8videos are not enough for photometry, roundview must be better

---

hardware set up 2

12cameras

2K recorder x 3


image processing

1. acquired images
x3
(camera position adjusted)

2. prepare background images

pickup images without object (automatically)
calculte min, ave, max for each pixels

x3

3. object contours and gravity center

blurring, binarization, and pickup contours with evalutaion

(calculated for 12camera positions.)

to

4. 3D remodelling (failed)

compare a pair of cameras using AKAZE
from camera real positions, object 3D positions are calculated

currently, camera number and resolutions are not enough for photomery..

5. morphing

morphing, using similar approach


raw image (pair)


calculate contour and gravitiy center

images positons adjusted by gravity center
calculate AKAZE and determine exterior contour
for each pixels, 10 closest AKAZE positions are picked up
calculate XY movement from 10 AKAZE position difference of two images
by changing, weight of two averages, images gradually morhing
0% from imageL, 100% from imageR
10% from imageL, 90% from imageR
20% from imageL, 80% from imageR
30% from imageL, 70% from imageR
40% from imageL, 60% from imageR
50% from imageL, 50% from imageR
60% from imageL, 40% from imageR
70% from imageL, 30% from imageR
80% from imageL, 20% from imageR
100% from imageL, 0% from imageR
captured images

---

result by set up 2


raw & contour movie, MP4

with morphing


morphing and raw, MP4x2



morphing and raw, MP4



how systems are running, MP4


morphing and raw, MP4

---