R-CNN designed by Ross Girshick has opened a new door for using deep learning algorithms to detect object.

There are mainly four steps to catch the target.

First, input an image.

The main algorithm that R-CNN uses to select proposals is the selective search algorithm. Separating the whole picture into small blocks and assessing the similarities between two adjacent blocks are required as the first step. Then computers can start merging the blocks with similar color histogram and similar histogram of oriented gradient (HOG).

Compute all the proposals with CNN features .

Support Vector Machine (SVM) is used to classify the region and Bounding Box Regression is selected to adjust the real place.