OpenCV 4 with Python Blueprints - Second Edition

The sliding-window approach

One of the earliest approaches to create an architecture that can detect multiple objects in the scene was the sliding-window approach. With this approach, you first build a classifier for objects of interest. Then, you pick a rectangle (a window) of a size that is several or many times smaller than the image where you want to detect an object. After that, you slide it across all possible locations in the image and classify whether there is an object of the chosen type in each position of the rectangle.

During sliding, a sliding size of between a fraction of the box size and the complete box size is used. The procedure is repeated with different sizes of the sliding window. Finally, you pick the window positions ...