0
$\begingroup$

This image show a histogram (200 bins) of accumulated distances from a radar distance meter (very noisy). The peak around 7 meters is an object. At thought this looked kind of like a normal distribution, at least if you ignore values <4m (which for this application is reasonable).

What I am trying to do is to filter out true distances based on the probability distribution.

alt text

  • 0
    Thanks for cross-posting on Cross Validated: here's the link to additional answers: http://stats.stackexchange.com/questions/5430/how-would-you-filter-this2011-05-18

4 Answers 4

1

If it wasn't for the spike at 0 and the mode at around seven, it would look roughly lognormal to me

0

The density function of distance-from-mean of a bivariate normal distribution takes the form $Ax e^{-Bx^2}$, which look pretty much like your noisy graph.

  • 0
    Ok, I see your point. Maybe I should have pointed out that it is a 1D radar.2010-12-13
0

I would fit a straight line to the bins near but outside the peak and subtract this from the data, claiming that this level is the background. Then you can do a 2D fit of the Gaussian parameters (mean and sigma) and see how it turns out.

0

This is a (tri-modal) mixture distribution.