Still experimenting with my people counting..
1. Once a blob is identified and displayed after filtering, how do You keep if from being lost when other blobs pass nearby?
2. Is it possible to erode a blob based on it´s initial COG to prevent it from being merged with other blobs? I.e reduce the blob to a small round circle.
3. Is it possible to get the COG displayed for all found blobs?
4. Is there any other way to track a person moving around over a fixed background. Perhaps locking into a returning shape found in people (head+shoulders for example). Any ideas?
I have a Dynapel auto tracking system at my office. It uses a fixed camera to find moving objects, give each found object an identity and then send coordinates to direct a speed dome camera towards the object and zoom in with a 27x motor zoom lens to the boundary lines created around the object. It can track about 20 objects in realtime and alternate between them in a prdefined cycle. When setting this system up You declare size of object of interest first in a near position, and the in a far position. Two object sizes can be set and tracked. Normally it would be people and cars. There are filters to avoid the camera getting stuck on moving leafs etc.
How is this most probably done? If two objects identified as "people" cross each others paths the system still keeps them apart. In setup mode You can see the ectangle around the object as it moves, and also what i believe is the COG of the rectangle/object with a trailing line.
For me it all comes back to HOW to track people moving around without losing them when they are near other people or pass behind a tree. I suspect this is beyond what RR can do, but i am still interested. There are several cheap products on the market today that can track faces, identify loitering people or people falling etc, but what are the mechanics behind it? Are they working with a reference image bank?
Something else that struck me; can RR based on a sample image cancel out an entire background in live video? Much like the movement does, but a step further by killing the entire backgrund. Or is this what Movement does?
I´ll keep trying to learn :-)