what are the algorithms exist for feature tracking and which algorithm is best for tracking in image sequence?

Hi everyone, i have 5 images of a person taken at different angles, i want to find the features and track them. i want to know what the feature tracking algorithms existed and which one is better.

Answer 1

[Walter Roberson]

Every possible function of up to numel(YourMatrix) inputs and up to the same number of outputs is a "feature". Some functions are more useful than others and so some features are more useful than others. You need to read research papers to decide what features you will implement, and that is not something that we will do for you.

[Spandan Samiran]

lucky for you there is no downvote option in Matlab forums

[John Szymanski]

lol tru

[Walter Roberson]

Suppose we start creating a list of feature tracking algorithms, like Kanade–Lucas–Tomasi, Sethi and Jain 87, Hwang 89, Salari and Sethi 90, Krishnan and DanielRaviv 95, MHT filters, PHD filters, particle filters, Arroyo, Yebes, Bergasa, Daza, Almazán 15 ...

So, these feature tracking algorithms: how many of them are there? What is the answer to the question "what are the algorithms that exist for feature tracking" ?

So we try, https://scholar.google.com/scholar?q=feature+tracking

"About 2,950,000 results (0.16 sec)"

That's more than three, right?

Does Google Scholar find all of the algorithms? Probably not: it doesn't index everything (not even all published papers), and people keep writing new algorithms.

If we were to go through Google Scholar and create an exhaustive list of every feature tracking algorithm there, how many might we be looking at? How many different feature tracking algorithms can exist? Because we need to know all of the feature tracking algorithms that can exist in order to be able to answer the question of which one is "best" (for some meaning of "best".)

This leads us to a question of what feature tracking algorithms are -- as opposed, for example, to chequebook balancing algorithms. Given an image, what defines a "feature" ?

The answer to that turns out to be what I stated earlier: that every possible distinct function with at most (input size) inputs and a fixed number of outputs is a distinct feature calculation usable for feature tracking purposes. (Variable number of outputs cannot be used without padding to a fixed size.)

Answer 2

[Spandan Samiran]

If you have Computer Vision System Toolbox and Image Processing Toolbox, there are examples of how to implement feature trackers in the documentation (e.g. here).

[Walter Roberson]

True, but that does not tell you what feature tracking algorithms exist and which one is better (and why).