CamShift

CamShift, which stands for Continuously Adaptive Mean-Shift, is an iterative algorithm used in computer vision for object tracking. It builds upon the Mean Shift algorithm by introducing adaptive window sizing. Mean Shift itself is a non-parametric clustering algorithm that iteratively shifts a window towards the mode (peak) of a probability distribution, often derived from image color histograms.

CamShift enhances this by adjusting the size and shape of the tracking window in each iteration. It uses the spatial distribution of the color probability density within the current window to determine how to move the window to better match the target. Crucially, it also analyzes the compactness of this distribution. If the distribution is compact, it suggests the target is well-defined and potentially moving away, so the window size is reduced. Conversely, if the distribution is spread out, it indicates the target might be closer or larger, leading to an increased window size.

This adaptive resizing makes CamShift more robust to changes in target scale and distance from the camera. The algorithm typically works by first identifying a target region, calculating a color probability model for that region, and then applying the Mean Shift procedure to find the new location of the target's probability mode in subsequent frames. The window size is updated based on the output of the Mean Shift search. CamShift is particularly effective for tracking objects with distinctive color properties, but it can be sensitive to significant illumination changes or occlusion.