I have data points in 3D space, and I'd like to quantify how well they conform to a straight line.

Specifically, this is a body tracking problem. I have multiple options available for the 3D positions of body parts. For example, there are points for the knee, calf, and foot. I need to select one correct point for each body part. Since the knee to foot is a rigid link, the correct points should be fairly aligned. I want to reject combinations of knee/calf/foot positions which are not in a straight enough line.

The line of best fit for 3D points can be found using the method described in this thread. Is there a metric for how well this line fits the data?

  • $\begingroup$ Please explain how a line can be a "best fit" to points in three dimensions. How did you find it? In what sense is it intended to fit the points? $\endgroup$
    – whuber
    Commented Mar 6, 2016 at 16:43
  • $\begingroup$ I used the information on this thread - the line of best fit passes through the centroid of the points, and is in the direction of a certain eigenvector. My specific problem is for body tracking. For example, I have options for the position of the knee, calf, and foot. I need to select correct points from these options. The knee, calf and foot data points should be in a roughly straight line. I want to reject combinations of points that are not "straight enough". $\endgroup$
    – Vermillion
    Commented Mar 6, 2016 at 20:04
  • 2
    $\begingroup$ I think it's a great question (+1). However, I recommend (a) editing it to include the details in your comment and (b) downplaying $R^2$, because it's neither applicable nor relevant to your problem. $\endgroup$
    – whuber
    Commented Mar 6, 2016 at 21:13


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.