Let's assume we have series of frames with a known object. I want to track the object using Kalman filter using constant velocity / acceleration models. My questions are:

  1. Which model to use? CV or CA?

  2. If I use CV, how to calculate velocity of object? and what is the unit of velocity? pixels/sec or m/s?

  3. If I use CA, how do we evaluate velocity and acceleration of the object?

  4. If object moves closer to camera, it drastically shifts in both and y pixels. How do we accommodate this change? CV and CA models fail.

  • $\begingroup$ You probably need to combine it with computer vision. If applying computer vision to the entire frame is too expensive you might be able to use the prediction step of the Kalman filter as the first guess for the computer vision. $\endgroup$
    – fibonatic
    Jul 10, 2019 at 9:46
  • $\begingroup$ What's your motion model ? Do you use 3D world coordinates or 2D projection coordinates ? $\endgroup$
    – Fat32
    Jul 11, 2019 at 14:59

1 Answer 1

  1. Depends on the properties of the tracked object.
  2. pixels / frame. You may convert to time units by the FPS.
  3. The model will estimate those given the detection per frames.
  4. You may use angle based models. They may require additional measurements.

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.