How to Improve Blobs Detection Accuracy?

Question

Consider the following picture: it is a frame from a video of few people talking to each other. I have tried to do blob detection on that video, but when two persons are near each other, it is considered as one blob. Is there a way to improve the blob detection, so that I can detect people more efficiently?![Blobs][1]

Answer 1

This problem is called Occlusion.

This is a typical problem where you need help of Kalman Filter.

Kalman filter takes previous values and predicts the future values. It is highly useful in noisy situations or inaccurate situations.

A lot of discussions are there in SO, which you can find by simple searching. But i found two of them really useful

1) Question on tracking a moving ball using Kalman filter

2) Use Kalman filter to track the position of an object, but need to know the position of that object as an input of Kalman filter. What is going on?

And Kalman filter is already implemented in OpenCV and inbuilt functions are available.

Also check this SO on multiple blob tracking : Multiple Blob Tracking

This is not only method. You can find plenty of papers regarding this on googling.

Paper 1 : This paper explains another method. Might be useful.

A lots of papers exclusively on this subject can be found here.