Kalman Tracking - Dispersion Measurements

Question

I do some things to track moving objects with the ceiling camera pointing down. I reached the point where I can determine the position of the desired object in each frame.

I am learning to use the Kalman filter to track the position of an object and speed through the scene, and I have reached a stumbling block. I created my system and have all the necessary parts of the Kalman filter, except for the variance of the measurements.

I want to be able to assign a significant deviation for each measurement to allow the correction phase to use the new information in a reasonable way. I have several measures assigned to my detected objects, which theoretically can be useful in determining how accurate the position is, and it is logical to try to combine them to obtain a suitable variance.

I am approaching this in the right way, and if so, can someone point me in the right direction to continue?

Any help is greatly appreciated.

Answer 1

I think you're right. According to this post:
Merging sensors with a Kalman filter
determination of variance is 100% experimental. It seems to me that you have everything you need to get good variance estimates.

Answer 2

sorry for the late reply. I personally encountered the same problem in my previous project. I found the advice given by Gustav Hendeby in his lecture slides “Fusion Sensor” ( Slide Page ) is extremely valuable.

Summarizing:

(1) The SNRs of your measurement noise and process noise determine the behavior of your filter. A high noise level / noise measurement makes your filter slower (low-pass filter), which usually provides smoother tracking, on the contrary, if you set a low measurement noise, you essentially have a high-pass filter that tends to have more jitter .

(2) There are numerous articles in the literature on how to properly establish this noise model. However, usually a lot of “setup” is required, depending on your application. Typically, measurement noise is what we can measure / characterize based on equipment specifications. Therefore, it is recommended to correct the “R” (measurement noise covariance) and adjust Q (process model noise covariance).