HI,
I would appreciate some help....
I am trying to write a program in MATLAB that will segment moving objects, for
example cars, from a series of images (extracting each frame from a video) and
that superimposes on each object its bounding box.
Has any one got an idea?
The images are RGB. and the bounding boxes must be seen on the RGB image not
binary or grayscale.
Image segmentation and tracking .... HELP!!! anyone got a clue?
Started by ●March 6, 2010
Reply by ●March 7, 20102010-03-07
Skms-
> I would appreciate some help....
> I am trying to write a program in MATLAB that will segment moving
> objects, for example cars, from a series of images
> (extracting each frame from a video) and that superimposes on each
> object its bounding box.
>
> Has any one got an idea?
>
> The images are RGB. and the bounding boxes must be seen on the RGB image not binary or grayscale.
Is the camera stationary? If so then suggest something like this to start:
1) "Posterize" or otherwise reduce pixel variation to short range of tones or gray shade values, like 0-255.
2) Subtract (or some people say XOR) successive images.
3) Put a box around regions where you have differences. This result would include both where the moving object used
to be and where it moved to.
Obviously 3) is more difficult, because you may get slight differences in non-moving parts of the image due to
lighting (reflections and shadows) caused both by the moving object and/or other objects outside the scene. Step 1)
helps with reducing those but still your 'box algorithm' should be wise about excluding outliers.
-Jeff
> I would appreciate some help....
> I am trying to write a program in MATLAB that will segment moving
> objects, for example cars, from a series of images
> (extracting each frame from a video) and that superimposes on each
> object its bounding box.
>
> Has any one got an idea?
>
> The images are RGB. and the bounding boxes must be seen on the RGB image not binary or grayscale.
Is the camera stationary? If so then suggest something like this to start:
1) "Posterize" or otherwise reduce pixel variation to short range of tones or gray shade values, like 0-255.
2) Subtract (or some people say XOR) successive images.
3) Put a box around regions where you have differences. This result would include both where the moving object used
to be and where it moved to.
Obviously 3) is more difficult, because you may get slight differences in non-moving parts of the image due to
lighting (reflections and shadows) caused both by the moving object and/or other objects outside the scene. Step 1)
helps with reducing those but still your 'box algorithm' should be wise about excluding outliers.
-Jeff
