Calculate error across multiple measurements

[ThiroSmath]

Basically, I'm working on a program that calculates distance to an object from an image. I wanted to see if there's a way of calculating the error produced with some samples, but it doesn't look like it is a clear linear progression.

For instance, when measuring 50, 40, 30 and 20 cm the program returns 52.6649, 41.9110, 31.5879 and 20.9529 cm respectively. How can I calculate an error function, if it is possible at all?

My apologies if this doesn't belong in calculus, I figured that the error function would require derivatives or differentials.

 

[Deleted member 4993]

Basically, I'm working on a program that calculates distance to an object from an image. I wanted to see if there's a way of calculating the error produced with some samples, but it doesn't look like it is a clear linear progression.

For instance, when measuring 50, 40, 30 and 20 cm the program returns 52.6649, 41.9110, 31.5879 and 20.9529 cm respectively. How can I calculate an error function, if it is possible at all?

My apologies if this doesn't belong in calculus, I figured that the error function would require derivatives or differentials.

In measurement there can be two types of error - systematic error & random error. It looks like your measurement has consistently "over-estimated" the length.

Thus it most probably has "systematic error" which could be due to:

faulty measurement device, or

incorrect modelling, or

incorrect calculation, or ........

First eliminate the sources of systematic errors.

 

[ThiroSmath]

In measurement there can be two types of error - systematic error & random error. It looks like your measurement has consistently "over-estimated" the length.

Thus it most probably has "systematic error" which could be due to:

faulty measurement device, or

incorrect modelling, or

incorrect calculation, or ........

First eliminate the sources of systematic errors.

Actually, you're right. An image resizing operation was the cause for the overestimation. After fixing that, I'm getting values much closer to the real distance (the relative error is now consistently less than 2% or 0.02, either over or under the real distance, as opposed to always being around 5% over).

I still would like to know if there is any better way to estimate the error than going on a case-by-case, though.