I have a sequence of images, and I need to find and track the creation of the objects, then their movement and then their disappearance. There can be up to $3$ objects overall, and sometimes there are fewer, or none. Between consecutive images there is a maximum distance that an object can move.

The practice I use is as follows.

  1. Using a neural network, I estimate the locations for the objects in each image separately, up to $3$ locations in each image

  2. Filtering out clear mistakes, e.g., random locations with no continuation over time.

After a little research, I found that with some effort I can translate this problem into a hidden Markov model, and this one can be solved with the Viterbi algorithm. The problem is that for each image there are more than $100$ possible object locations, and with $3$ objects we get over $10^5$ different states.

Is there a designated algorithm for such object-tracking over time? Or otherwise, if there's a good and efficient way I can fit Viterbi algorithm for this problem?

  • $\begingroup$ What is the motivation for such tracking? If the motivation is military, I doubt many are willing to help you build a targeting system for free $\endgroup$ Nov 19, 2022 at 14:45


You must log in to answer this question.