Only intermittently finding the 2 trailer colors

[ellisk]

Hi all,
I've been trying to get the camera to track the two colors using the code at http://first.wpi.edu/FRC/frcupdates.html, and it sort of works. It seems to almost never recognize the colors, except every once in a while. Even when it stays in the exact same place as where it recognized the image, I still cannot get more times where it recognizes the two colors.

The camera is correctly focused (we viewed it, and it can clearly see the colors on the trailer). The settings are correct in terms of brightness etc.

We then tried to increase the range of acceptable colors, and it still didn't work. Increasing other tolerances in Target.h also didn't help, and strangely didn't give us any false positives (just the same huge number of false negatives). Does anyone else have the same problems or know what causes this?

Thanks,
-Kevin

[RedHeadRobotics]

We had to play with the colors a lot. We found that different lighting also made a difference. Try an incandenscent maybe instead of a flourescent. Just trying different things can help. We did a lot of trial and error.

[ellisk]

Originally Posted by RedHeadRobotics

We had to play with the colors a lot. We found that different lighting also made a difference. Try an incandenscent maybe instead of a flourescent. Just trying different things can help. We did a lot of trial and error.

We've done a lot of experimentation with different tolerances and camera settings, and it still is very touch-and-go. Do you recommend any particular settings to play with?

[EdenA]

The api definitely needs to be more documented. Trial and error is not fun for anyone.

[Peter Randall]

We found two problems - the vertical control for the servos controlling the camera were reversed from our installation. You need to check the motion of the camera after a find. Second, check the color of Friend. We spent an hour chasing our own tail. Once we reset these, the demo code worked fine.

PGR

[Peter Randall]

We found two problems - the vertical control for the servos controlling the camera were reversed from our installation. You need to check the motion of the camera after a find. Second, check the color of Friend. We spent an hour chasing our own tail. Once we reset these, the demo code worked fine.

PGR

[Greg McKaskle]

It sounds like some, perhaps most, of the issues are with the gimbal construction or setup. If there still seem to be camera issues, I'd be glad to help out, and I'd love to hear what sorts of lighting works/doesn't work with the default values that were shipped.

A brief explanation. The hue colors are quite wide to allow different types of lights to still work. The green fabric in particular will tint to a different color based on sunlight, indoor fluorescent, indoor halogen, etc. The downside to having the colors set this wide is that they may identify too many other materials and colors as being the target color. Making it too narrow may miss the target when the light color changes. The current settings were selected to work in classrooms, shops, event lighting, and hopefully outdoors. If you find that you need to widen them further we'd really like to know the circumstances. If you decide to narrow them, that is fine, but hopefully not necessary.

The rest of the camera settings are hopefully fine provided the camera adapts the exposure to the scene lighting in a good way. The camera is set to use a light metering to adapt to the amount of light. This seems to work quite well with a lag of 0.5 sec or less. Certain lighting conditions can confuse it, and for this, the brightness setting is handy. This will shorten or lengthen the exposure computed using the light meter. With the green active targets, setting this low will keep the sensor from saturating, making the green light look green instead of white. Similarly, a spotlight on the target will cause glare on the target. If the background color is dark, the camera will average things out and will pick an exposure that is too long and will make the glare worse. Again, shortening the exposure with a small brightness number fixes this. Finally, if there are bright windows or lights shining into the camera, it may pick an exposure too short, making the targets appear dark. Raising the brightness in this case will help.

Again, we hope the defaults work in most cases and want to help and to hear when they don't.

Greg McKaskle

[yoyodyne]

Has anyone had a problem with the FindColor() function finding a color exactly every Nth call? We haven't had much time to look into this yet. We think we are waiting enough time between calls

To do this test we edited the now old Iterative robot example and added the following to the constructor:

if (StartCameraTask(10, 0, k160x120, ROT_0) == -1)
97 {
98 dprintf( LOG_ERROR,"Failed to spawn camera task; Error code %s",
99 GetVisionErrorText(GetLastVisionError()) );
100 }
101
102 // Set up camera tracking parameters.
103 tdata = GetTrackingData(GREEN, PASSIVE_LIGHT);


The periodic autonomous function as shown below prints the "found value..." to the console exactly every 5th time with 4 ShowActivity("\n") between.

void Team116_Demo::AutonomousPeriodic(void) {
158 // feed the user watchdog at every period when in autonomous
159 GetWatchdog().Feed();
160 ParticleAnalysisReport par; // particle analysis reports
161
162 m_autoPeriodicLoops++;
163
164 // generate KITT-style LED display on the solenoids
165 SolenoidLEDsKITT( m_autoPeriodicLoops );
166
167 if ((m_autoPeriodicLoops % 60) == 0) {
168
169 if (FindColor(IMAQ_HSL, &tdata.hue, &tdata.saturation, &tdata.luminance, &par)
170 && par.particleToImagePercent < MAX_PARTICLE_TO_IMAGE_PERCENT
171 && par.particleToImagePercent > MIN_PARTICLE_TO_IMAGE_PERCENT) {
172 // drive toward color
173 m_robotDrive->Drive((float)1.0, (float)par.center_mass_x_normalized);
174 ShowActivity("%s found: follow x value: %f", tdata.name, par.center_mass_x_normalized);
175 }
176 else
177 {
178 m_robotDrive->Drive(0,0);
179 ShowActivity("\n");
180 }
181 }


We have tried calling FindColor() at different rates with the same result. Any help would be appreciated.

Thanks,

Greg