Find the position of objects in two dimensions XY

Dear sir/Madam,

I am specislized in the area of machine learning and data mining. I am looking to extract the position of objects in two dimensions (XY) from Video file make analysis if these trajectories using machine learning algorithms thereafter.

I would like you to advise me which toolbox in matlab I can use?

Thank you so much

Answer 1

[Nicolas Schmit]

The relevant toolboxes are the Image Processing Toolbox and the Computer Vision Toolbox.

[Walter Roberson]

You asked which toolboxes were the relevant ones; the above was the answer.

As you are talking about machine learning, I might also suggest Neural Network toolbox, depending how you were thinking of doing the machine learning.

[Walter Roberson]

Image Processing Toolbox

Computer Vision Toolbox

https://www.mathworks.com/products/neural-network.html

It would have taken you less than a minute to search for those, which would have been faster than waiting for me to search out the links and write the posting.

[Walter Roberson]

Code to track moving objects:

https://www.mathworks.com/help/vision/examples/motion-based-multiple-object-tracking.html

[Mohamed Elwakdy]

Hi Walter,

All the trajectories of the objects are available in the array of tracks in the code. This means after I test this code on any video file, I can get the trajectories of different objects into this array. Is this right?

Thank you so much

[Mohamed Elwakdy]

This means after I use this code https://au.mathworks.com/help/vision/examples/motion-based-multiple-object-tracking.html for a video file and run it. I will get all positions of the objects if I write "tracks" on the command window of Matlab .. right?

[Walter Roberson]

"I will get all positions of the objects if I write "tracks" on the command window of Matlab"

Not quite. tracks is used as a shared variable. It is created near the top of function MotionBasedMultiObjectTrackingExample and all of the other functions are nested inside that one, so they are accessing the shared copy. Once the function MotionBasedMultiObjectTrackingExample has finished executing, it would return and its workspace would disappear, removing the tracks variable. So you would only be able to examine the tracks variable if you put in a breakpoint, or if you change that first line from

function MotionBasedMultiObjectTrackingExample()

to

function tracks = MotionBasedMultiObjectTrackingExample()

"This means after I test this code on any video file, I can get the trajectories of different objects into this array. Is this right?"

The tracks variable only keeps a record of the objects in the current view, the ones that have not yet gone through the deletion process. Once the code has decided that an object is gone, it removes the record of it; it does not keep it for any kind of archive purposes.

Also, the code does not keep a record of all of the historical positions of an object: it only keeps enough information on an object to be able to do motion estimation.

It would not be difficult to make the changes to track all of the history of an object and to keep those around -- though I would suggest that at the time that an object is deleted, that you flush the information about it to a file rather than keeping it around in memory.

[Mohamed Elwakdy]

Thank you so much for your great help. Thank you so much really for everything.

I think that to overcome that the code has decided that an object is gone and remove the record of it that I should select a video file where the object is walking/running in a straight line. In this case, the object will not be lost and can keep all the trajectories of these objects... right?

[Walter Roberson]

Sure. But it will still only know the current position and one or two previous positions.

You could alter the code slightly to keep a "unique identifier" for each object being tracked. At each time point in which you detect a new object, you allocate it the next available identifier. At each time point in which an object is detected, you write out to a file its unique identifier and the timestamp and the current position of the object. Do not bother to keep long internal histories, and allow any object to be deleted the same way it is now.

At the end, close the file. You can then post-process the file to group by identifier number to track an individual object, or you could group by timestamps to see what was on camera at any given time.