I have some python code which uses cv2 to perform motion detection and tracking. However it often detects one object as two or more separate objects. I think this might be due to kernel size. Increasing iterations only makes the white bits larger, instead of filling in the object properly. The line in question is:
thresh = cv2.dilate(thresh, None, iterations=2)
Where None is, I can input the kernel size. I've attached two pictures, one is the result of a kernel size of 3x3:
the other 5x5:
As you can see, neither are successful, and the result is 3 or 4 small boxes drawn over the animal rather than one large one with the correct outline eg:
How can I determine correct kernel size? Does image resolution matter? I have been doing reading on morphology and just don't really understand it.
Any help appreciated.