Research on target localization with large viewing angle

Abundant and effective affine-invariant features directly influence the localization accuracy when the target is moving in a large field of view. The SIFT algorithm is widely used in feature extraction and stereo matching, but is not efficient in this case. In this paper, we propose a robust target localization method using stereo vision system when the 3D target is in large viewing angle, the method is based on the new algorithm that integrates MSER detector with SIFT descriptor and can greatly increase the localization accuracy. The whole method is divided into three parts: initial image matching is made through extracting abundant affine-invariant features from pairs of images captured from the stereo vision system. RANSAC is adopted to detect and reject the erroneous correspondences. The final image matching is finished with the large field of view. Several localization experimental results of the moving target in large viewing angle indicate that the proposed method has good performance of precision and robustness in localization.