Balaje Kalyanaraman

PhD Student at the University of Newcastle, Australia

16 May 2014

Tracking Objects without color

Object tracking is one of the most important aspects of Computer Vision. Nowadays human security guards are being replaced by automated tracking robots for two main reasons; ONE: robots can detect even the subtlest of motions and record them efficiently; TWO: robots don’t get bored staring into the same space, even for days.

Object tracking can be done in numerous ways. A simple way is to detect the color of the object and track it. But this has many problems. First, the object must be of a uniform color. No object in nature has a color so uniform, that it covers the entire object, like a tree which has green as well as yellow leaves. Second, the background. Suppose we want to track a red object in a red background. This can’t be done using color filtering.

So what is the solution? The answer is quite simple. Detect changes in the environment directly from sequence of images i.e. subtract two different frames to find out the changes. Let $$F(t_1)$$ be the input frame at time $$t_1$$ and similarly $$F(t_2)$$ be the input frame at time $$t_2$$. Now the difference between $$F(t_1)$$ and $$F(t_2)$$ gives us only the pixels which has changed in the time-interval $$t_2-t_1$$.

$I = | F(t_2) - F(t_1) |$

The resulting image I has the information about the moving object only. Contours can be extracted from this image to track the object. A sample video (made by me!) demonstrates this method.

This video attempts to track the ball as it jumps across the frame, and the result is quite good. This is done using OpenCV’s Image Processing Library. OpenCV offers this function to compute the absolute difference between the two Image matrices.

cv::absdiff(firstimage, secondImage, outputImage);


Stay Tuned.

tags: opencv - objecttracking - imageprocessing