ColorLightSensor battery voltange dependency

[s.frings]

Hello all,

I noticed that the ColorLightSensor seems to depend much on the battery voltage. With fresh alcaline batteries, it always detect all four colors of the balls (from the Mindstorms 2.0 set) properly.

With rechargeable nickel batteries (7.8 Volt), the green color is often detected as black. After some time (7.4 Volt), green cannot be detected anymore at all and yellow balls are often detected as red. If tried different angles and istanced and I eliminted ambient light without any success.

This is really strange to me, because

a) Independently of the battery type, the device should work down to 1 Volt per Cell (6 Volts in total) because both alcaline and nickel batteries are empty below that level.

b) The hardware developer kit says that the four input connectors provide 4.3 Volt, so why does it depend on the battery voltage?

Any Ideas how to fix this?

Debugging Info, collected with 7.4 Volt battery Voltage using four balls with a distance of about 8mm:

Blue ball:
raw r/g/b/n: 133,150,273,97
cal r/g/b/n: 23,44,127,32
Color: 2, BLUE

Green ball:
raw r/g/b/n: 119,162,134,89
cal r/g/b/n: 19,61,32,22
Color: 1, BLACK

Yellow ball:
raw r/g/b/n: 376,252,176,114
cal r/g/b/n: 175,113,43,54
Color: 4, YELLOW

Red ball:
raw r/g/b/n: 347,135,140,114
cal r/g/b/n: 154,14,16,55
Color: 5, RED

[s.frings]

Ooops, I played around a little more and now I cannot detect the green ball anymore even with fresh alcaline batteries.

So it might not be related to the batteries. But anyway - my main problem is still there: I cannot detect the green ball!

More debugging data, taken with fresh alcaline batteries at 9.1 Volts:

Blue ball:
raw r/g/b/n: 122,132,244,88
cal r/g/b/n: 24,36,112,21
Color: 2, BLUE

Green ball:
raw r/g/b/n: 118,150,126,88
cal r/g/b/n: 20,51,27,21
Color: 1, BLACK

Yellow ball:
raw r/g/b/n: 337,229,167,115
cal r/g/b/n: 149,94,38,61
Color: 4, YELLOW

Red ball:
raw r/g/b/n: 299,125,125,104
cal r/g/b/n: 131,18,15,43
Color: 5, RED

[s.frings]

I did a simple visual test.

If I switch the ColorLightSensor to expose green light, and let it shine on the green ball, I see visually only few reflection. The ball appears nearly black to my eyes. So it is no wonder that the "cal green" value is only 50.

If I do the same with the red or blue ball, I see a great difference. The blue ball reflects the red light very good, and also the blue ball reflects blue light very good. In both cases, the "cal red" and "cal blue" values are above 100.

It seems that the color of the green ball does not match the color of the green Led. To check this further, I used a piece of green paper, but noticeable lighter color than the ball has.

This paper can be detected cleary as green, and also the measured value is much better ("cal green" is greater than 100).

Is there anbody who could tell post his values for a green ball (Using the ColorSensor Test" sample program and using the corrected classes.jar).

[gloomyandy]

Hi,
Before going too far down this road you may be better to wait for the next release of leJOS (coming soon now I hope). This has an updated color recognition algorithm (as used by the latest Lego firmware). If you understand how to use a developer snapshot you could try it out now. Otherwise if you want to take a look at the new code it is here...
http://lejos.svn.sourceforge.net/viewvc ... iew=markup
Take a look at the readValue method...

Andy

[s.frings]

I found a workaround, by replacing the ColorLightSensor.readColor() by the following method:

public static Color getFarbe() {
int[] raw=new int[4];
farbSensor.readRawValues(raw);
//LCD.drawInt(raw[0]-raw[3],3,0,7);
//LCD.drawInt(raw[1]-raw[3],3,4,7);
//LCD.drawInt(raw[2]-raw[3],3,8,7);
boolean rot=(raw[0]-raw[3]) > 100;
boolean gruen=(raw[1]-raw[3]) > 50;
boolean blau=(raw[2]-raw[3]) > 100;
Color farbe;
if (rot && gruen)
farbe=Color.YELLOW;
else if (rot)
farbe=Color.RED;
else if (blau)
farbe=Color.BLUE;
else if (gruen)
farbe=Color.GREEN;
else
farbe=Color.BLACK;
//LCD.drawString(farbe.name()+" ", 12, 7);
return farbe;
}

It detects only the four colors that I am interested in. The trick is, to define lower threshold for green, than for the other colors.

Anyway, I think the green LED should have a darker color or the green ball should have a brighter color, so that both match. Then the ColorLightSensor.readColor() would work as it should.

If the color detection code part of the hardware, or can we change it in Lejos?

[gloomyandy]

Did you see my previous reply?

Andy

[gloomyandy]

Oh and using the raw values is probably not a good idea for any general code. The values returned from one sensor to the next will vary. That is why Lego supply each sensor with individual calibration data...

Andy

[s.frings]

Oh, yes. I can see that my problem will be solved in the new class source. I never noticed the problem with the Lego Firmware because I started with version 1.29.

I found snapshots in the SVN repository, which I will use now. Thanks for your help.

[s.frings]

It works fine with the new classes.jar snapshot from today.

The ColorLightSensor class has been renamed to ColorSensor. The Color class does not support the name() function anymore, so I need to rewrite my programs a litte - but that's not a problem.

I also updated the VM and the menu application using the snapshot files, just to ensure compatibility.

[gloomyandy]

Make sure you also use the linker (jtools.jar) from the snapshot... In general you are best to use all of the files in lib along with the new firmware and menu code...

[s.frings]

Sure, I replaced all jar files and binaries that are in the snapshot directory.

[Coppola]

This paper can be detected cleary as green, and also the measured value is much better ("cal green" is greater than 100). Is there anbody who could tell post his values for a green ball (Using the ColorSensor Test" sample program and using the corrected classes.jar).