barrel detection is unreliable

[touretzkyds]

Barrel detection is very sensitive to lighting. The images below were tall taken without moving either the robot or the barrels. The only change was that I moved my hand to change the amount of light falling on the barrels. As you can see, in full fluorescent lighting the barrels are not detected. Shading the barrels caused one or the other to be detected, but it was difficult to detect both at the same time. This robot was running firmware version 1.0.3 and my vex-aim-tools software. But if I used the on-board Dashboard vision viewer, I had the same problem.

[jpearman]

They seem to be rather dark, noisy images. I wouldn't consider that a bug, there's a limit to the object detection.

[touretzkyds]

I don't see how darkness is the problem here. Shielding the barrels, making the images darker, improved the recognition. It's the brighter image where recognition fails. And this wasn't an unusual environment; it's just the desk in my office, with normal fluorescent lighting.

These images were taken using one of the production robots I received 12 months ago. But if you look at the last two images in #15, they don't look that different from this one, except for the color variations.

Not sure what to say about the "noisy" part. I'm not processing these images in any way, just taking the raw camera image from the robot, drawing a bounding box when a barrel is detected, and displaying it in a window. I can try again with another robot.

[jpearman]

All I meant was that it looks like a very low light situation causing the camera gain to be high and creating a lot of digital (or analog, idk) noise in the image. A more typical setting that AIM is used in gives better results.
an example.

Anyway, I'm not sure this is the place to discuss as it really in not related to the websocket code.