User:S1150038

’’Mean-Shift Tracking Algorithm’’ is an iterative method that tracks non-rigid object by locating the maximum likelihood using the object's colour-histogram. It is rather robust in dealing with rotation and deformation of the object. Application include computer vision  and  Human–computer interaction.

History
The core procedure "mean-shift" was first presented by Fukunaga and Hostetler in 1975. And later became popular after Cheng, Yizong published the paper.

Steps
The first step of the tracking procedure is to generate the colour histogram of the target in the first frame. Then we apply a technique called "back-projection" to the next frame. This is done by substituting the intensity of each pixel by its probability in the target image histogram. With the probability distribution image, the next step is to find the mass center of the window in the second frame which is located at the position where the object is selected in the first frame. The equation is as shown below:
 * Back-projection
 * Mean-Shift

$$x=\frac{M_{10}} {M_{00}}$$, $$y=\frac{M_{01}} {M_{00}}$$

Where $$M_{00},M_{01},M_{10}$$ are moments of the window. The result will be the new location of the window and when the difference between locations are converged, the object is found.

Development
In 1998, Bradski from intel improved this method to CAMShift  (Continuously Adaptive Mean-Shift) which updates the histogram of the target continuously to improve the robustness.