Real-time template matching based on gradient direction vector product

In order to increase the speed of image matching, we propose a novel real-time matching algorithm which contains two stages: online stage and offline stage. We realize the information extraction of template image in offline stage and the calculation of similarity measure value in online stage. Gradient direction vector product is used to obtain the value of similarity measure, which makes the algorithm against occlusion, nonlinear illumination, and clutter. For increasing the speed of image matching, optimized measures such as coarse-to-fine pyramid search strategy, contour extraction, contour approximate and stop conditions, are used. The experimental results show that the proposed algorithm has strong real-time and it is robust to rotation, scale and heavy noise.