[ Discussions | TOC | Search | Post | Reply ]
Category: Design Note
From: Forwarded from Steve Tell
Time: 12:02:19 PM
Remote Name: 22.214.171.124
---------- Forwarded message ---------- Date: Sun, 21 Jan 2001 22:11:15 -0500 (EST) From: Steve Tell <firstname.lastname@example.org> To: Ashes Ganguly <email@example.com> Subject: progress Re: Flood
Hi Ashes, Good to meet you and the team yesterday.
Before leaving town yesterday, I had a chance to prime the 2x6 in white, and today I returned early enough to be able to spraypaint black stripes and install it in the camera's field of view, although not where it will be subject water except in serious floods. Still, it should provide much better test images than anything I had set up previously. I was running out of daylight, so I didn't paint any borders down the side of the pole - both black and white stripes run all the way to the edge.
see http://www.telltronics.org/creekcam/0121/0121-1705.jpg, http://www.telltronics.org/creekcam/0121/0121-1710.jpg
I also temporarily installed the prototype 6-LED array that I showed everyone yesterday. Its tied to the tree on the left. It appears that a 2-inch LED spacing like this array as is just barely usable at this distance. While the LED current is higher than in the single-LED experiment, they appear dimmer because they are not pointed directly at the camera. They're quite directional. In this case, it was a choice of having the segment of pipe vertical, or pointing the LEDs more directly at the camera. I chose the former. It may be possible to arrange for the LEDs to not be pointing perpendicular to the pipe, but rather to tilt upwards. Also notice the difference in aparent intensity of the LEDs in the image. It wasn't aparent before installation. I'm not sure what factors contribute to this.
I'll need you and the team to let me know if this exact LED-array design produces sufficiently-useful images, or if I need to experiment more before building a full-sized array and setting up permanent power for it.
see http://www.telltronics.org/creekcam/0121/0121-1750.jpg, http://www.telltronics.org/creekcam/0121/0121-2100.jpg
I didn't have time to carry a sprit level out back, so both the striped pole and LED array were only set to vertical by eye, but they seemed a lot closer to vertical on-site than they look in the image. The camera might be somewhat rotated, and its lens might have significant geometric distortion, all of which the software should be able to ignore once we verify that the measurement markers are indeed veritical.
I thought of a feature that would probably be to the team and to me: have the image-processing software generate a modified image that shows what it is "looking at." Once you do the various pre-processing steps such as thresholding or edge-detection, and are ready to do the measurement by counting stripes or LEDs, somehow composite the processed image with the original. This will help a lot in determining if the algorithms you come up with are identifying the intended features.
P.S. I thought I recalled seeing an annoucement about Intel open-sourcing some part of their image processing library. A quick search found it; I just downloaded the linux version from:
So that looks like an option you guys can consider.
-- Steve Tell firstname.lastname@example.org