IU/IRRT/MTG/9-24

IU/IRRT Task Group Meetings

24 September 2004

Attendees: Antolovic, Arnab Cutzu, Himebaugh, Holloway, Ira, Leykin, Johnson.

Camera Recommendation (Leykin). Alex's report generally concurs with J.P. Mellor's report to the IRRT group. However, the $3,100 price of the Bumblebee is well below the cost quoted by J.P. for a low-end stereo camera.
- The one-inch resolution quoted for range resolution ("30'" feet or meters?) is assumed to be with a stablized camera. One inch would be adequate at 30 feet and might be adequate at 30 meters, but this leaves open the question of resolution under shaking. Opinions varied about the effective exposure speed of the camera at 30 frames-per-second.
- The group agreed that some form of peripheral vision is required, so that more cameras are needed, particularly in the horizontal range. Monoscopic peripheral vision will suffice. (SDJ note: isn't side-ways range finding even more important on steep banks, and wouldn't a $3,000 stereo camera be relatively cheap for that purpose?)
- No one but SDJ though that very cheap cameras would be viable for speed and attitude estimation.
- action (Cutzu, Leykin): Florin and Alex will finalize a camera recommendation during the week of 9/27. Early in the week would be desirable.
The image group believes that high luminosity headlights are needed to reduce daylight shadow problems in image processing.
Controller Status (Himebaugh). The vehicle-side requirements are still evolving, but the objective is a board layout for Doug's visit to Bloomington on 9/27. In the absence of more definite requirements, the B649 seminar should proceed with a higher-level command interface prototype that conforms to the published message format on the network side and appropriate hard-real-time RCOM tasks in RT-Linux.
Action (Himebaugh): Bryce will publish a common repository name in \scratch on the CS network.
Image Samples (Bowers). Cory was not present. Default image format is still to be determined. Some argued TIFF. For now it doesn't matter.
Action (Bowers, Dhua, all): A collection of still images will be place in the repositor the week of 9/27 to enable more substative discussion of image feature extraction next week.
Vision System Design. The preliminary discussion was quite vague, centering on multitasking feature extraction processes and various ways CBR might be deployed in the scheme. A main discussion item involved the need for mode and context information and the idea of configuring the vision system depending on the context. But no concrete details emerged.