Auto Calibration Line follower

[ahmedragia21]

we used manual calibration (using a pot) to sense the white color using Cds LDRs, it didnt work cuz of room light , we shield it , didnt work too ..., i want to use autocalibration with 89c52 and cuz it doesnt have analog port ,gonna use ADC ..first i will get the value of the sensor when its on the white line and store it , and check if the sensor's reading is equal to the precomputed value while the bot is moving , now the bot moved on the white line

is this better than manual calibration and would work ?

 

[philba]

Uh, are you using IR based sensors? You should have a daylight filter if that's the case. I have several line followers that use the fairchild sensors with daylight filters and have never had a problem with ambient.

Baring that, I would figure out how to measure both and then use the average of the two to determine which your sensor is over.

note that checking against a measured value will have problems. your system will have noise and thus your measurements will bounce around a bit. as long as the measurable difference between white+ambient and background_color+ambient is above 2X the noise level, you will be fine.

note also that each sensor will have a somewhat different response.

 

[ahmedragia21]

thanks for ur reply , so the method wont work ??
the ground is white lines on green flower ....
we shielded it but the LDRs still ambient to light ...
so ur opinion is to use Infrared ?

but we dont want to use manual calibration ( using the pot ) for every floor :S

 

[philba]

what sensors are you using?

also, is the background a solid color? If it is some sort of pattern, you will probably have trouble.

 

[ahmedragia21]

im using two cds LDRs with two Red LEDs and a compartor .

yea , its a green color , with white lines on it .

 

[philba]

CDS? hmmm, they are very slow. not the best choice. do you know the spectral response? I'd bet it peaks about where sunlight peaks. as such, I don't see how you could make this work well.

I would replace those with matched IR LEDs and phototransistors. by matched, I mean same wavelength.

 

[bloody-orc]

i'm using this combo:
normal 13000mcd red LED connected to 5V through 270ohm resistor.
4k7 Light dependant resistor connected to microcontrollers ADC input.
all haused in a wooden cube. a piece of cloth around it to shield ambitient light out. and works on all surfaces and colors just perfectly (2 of those sensors and line running betwean them).

 

[ahmedragia21]

i still have ambient light with LDRs and errors even i did shield it from all light

 

[bloody-orc]

i think that errors come from your schematic not light. do you have all the caps needed? are you doing any data communication with a PC? if yes, then this might cause some problems. if your uC is fast, you could try reading 5 values and then calculate the midpoint {(value1+value2...value5)/5}.

 

[philba]

there may well be noise problems but I'm pretty sure that ambient light is an issue that needs to be dealt with. I really think that you need to use IR with "daylight filters". I think you can make a filter with exposed camera film (the dark part). the LDRs will pick up stray light and make your problem much harder.

 

[odense]

I do not think you need Calibration to find exact point. Try to use Lagrange Intepolating Polynominal. Using the theory of 3 points, you can find exact point from centre.
But you are using LDRs so it is too much affected by ambient light so you should think how to modulate the signal to a range out of ambient light.
Best,