Object Detection Based on SIFT and Clustering Algorithm

Abstract

Detecting objects from complex scenes is a classic problem in computer vision. It is the basis of object recognition and image understanding. In this paper, we propose an object detection algorithm based on SIFT feature and clustering algorithm. We first extract feature points from the template and the target image. Then, after three times of clustering and filtering, the rotation, scaling and translation of the object from the template image to the target image are estimated. Finally, using the transformation matrix, we can obtain the position of the object in the target image. Our experiment result proved that the proposed algorithm outperforms RANSAC in effectiveness and stability. Our method also avoids manually setting of thresholds that is required in the RANSAC method.