Minutes of OMNIVIEWS Meeting
Lisbon June 29, 2001
IST: Jose’ Santos-Victor, Claudia Decco’, Nial Winters, Jose’ Gaspar, Pedro Soares, Alex Bernardino
CMP: Tomas Pajdla, Branislav Micusik
DIST: Giulio Sandini
Presented ideas for new mirrirs profiles (mixed mirrors) to have on the same mirror both a contant vertical resolution and a bird’s eye view on different parts of the image.
Mixed Mirrors could be designed in two ways:
a) fovea/retina (fovea looks at ground but seems too small: Tg=0.23cm)
b) retina/retina (ground and cylinder is mapped to periphery)
The two profiles may be smoothly joined.
Profiles of the mirror have been recomputed and differences with respect to theory are slightly different
A Software package (Matlab) to design profiles of mirrors (also mixed) havs been presented.
IST gave the code to CMP
A new mixed mirror will be realized and tested.
Presented different designs of mixed mirrors.
1) Combined with blind zone
2) Combined with no blind zone
3) Bird’s eye view (based on standard sensor)
4) Single design (logp sensor – current mirror)
Influence of the following two parameters is not known:
1) Astigmatism due to non planar mirrors.
2) Field Curvature
Include an optical specialist in the second phase
Distributed a CD containing the latest versions of the drivers for Windows-NT/2000 and Linux.
Also the code for acquiring log-polar images from Matlab. This has been a joint effort of IST and Observit.
It was noted that the images in MATLAB are displayed with rings mapped into vertical lines while for omnidirectional views it would be better to map rings to horizontal lines
Modify the code accordingly
Tomas also pointed out that before processing it would be better to interpolate between successive pixels of the rings to account for the “half-pixel” shift of the rings of the sensor’s structure.
Add function to do that
Should we think about removing the half-pixel shift for next phase?
Presented application for navigation using Nomadic robot + Camera. In particular:
c) may be target pose through interactive reconstruction)
a) Path following
b) Topological maps
Pointed out thet for bird’s eye view it would be better to start from a Cartesian array.
Rectangular instead of polar fovea for second phase?
Also no good results so far because of poor quality of the images. To be checked later today.
1) Topological Navigation
a) Corridor guidelines
Presented the outline of a demo for tracking image regions during motion of objects. The method, grossly speaking is based on a correlation method applied to warped templates of the area of the image to be tracked. It could be used for surveillance.
DIST is working on the transmission demo. Live images from Genova to Lisbon being transmitted on a GSM, Phone and internet.
Besides, a new viewer for log-polar images is being developed,
Presented a live demo for surveillance. “Interesting” regions are extracted and tracked on the basis of difference between a continuously updated background and the current image. Demo in MATLAB runs at 2 fps.
Also continued the evaluation of the quality of mirror:
1) some errors on the “ceiling” part
2) quasi-central projection
New technology for mirrors is being tested (different company) based on pressed glass coated with a reflective surface (much cheaper).
Also the possibility of designing a new mirror with all pixels having an aspect ratio of 1 was discussed (currently it is close to 1.3).
Investigate this for the second phase
A discussion about the strong points of OMNIVIEWS pointed out:
1) For the same number of pixels in the resulting image a standard solution will not be better (and our approach is “simpler”). We need to provide some numbers to show this. For example the computational cost of remapping images from standard cameras.
2) Processing is zero in our case
3) Less power consumption (due to less pixels)
4) Polar sampling is best for panoramic image (is there an alternative to “log”?)
Date of final review was fixed to September 27-28 (with a second choice 20-21).
Date for submitting the second-phase proposal is mid-January 2002 (to be checked at a later date)
For the final review we need to prepare:
1) A draft final report
Giulio will circulate the structure of the final report for discussion
It would be important if in the final report we could include simulations of pictures obtained with higher resolution “log-polar “ sensor. In order to do that we need high resolution images taken from the current mirrors.
Demos (to be better specified at a later stage)
1) Surveillance live demo by CMP
2) Live Navigation demos from IST
3) Possible surveillance demo from IST (Alex algorithm)
4) Transmission demo live from a panoramic camera in Genova
About the second phase it was discussed the addition of other partners both from academic/industry for the “research” work and from industries for real applications. Important application areas are:
3) Transmission (also related to surveillance)
At the beginning of September provide an updated outline of demos
In the afternoon we worked in the lab where we were shown a few demos from IST and the three cameras from IST, CMP and DIST were compared. The “low image quality” problem encountered by IST was solved. The images now look of better quality. Images were acquired in the corridor at IST with two cameras and compared.
The meeting ended at 8:30 pm