Thursday, 7 April 2005

Precision and Accuracy

I was pleasantly surprised at the accuracy of the tracks I was getting out of my GPS unit until the other day.

Then I noticed something odd; the tracks in places follow a very smooth curve, apparently matching the road I was following. At certain points there are glitches, where the track jumps a few metres then carries on in a smooth curve.

Diverging GPS tracks

The Above image shows the same section of path walked on two separate occasions. Note how the track for the higher resolution data (yellow) deviates from the former track. Also note the sudden re-alignment and the consequent parallel nature of the paths. (Scale of 1m/pixel)

This highlights to me, the difference between precision and accuracy. The precision of GPS units looks impressive, but precision is not the same as accuracy.

Precision is how well defined a value is, for example 3.141593, is precisely defined to the 6th decimal point. This does not mean, however that the value is correct - i.e. it is not necessarily accurate.

Accuracy is a measure of how correct a value is, for example 3.14 ± 0.02. This tells us that there’s no point in quoting a precision of six decimal places (as above) because the remaining four decimal places will almost certainly be wrong.

Relating this back to my GPS example we can see that the unit precisely tracked a curve, but not necessarily the correct curve. The curve looks correct because it precisely follows a neat curve, but it is not accurate because it is the wrong curve.

I’m assuming that the jump occurs when the unit locks to a new satellite, or a building reflects the signal.

No comments: