An approach based on stereovision is proposed for ball-shaped robot localization. Shi-Tomasi detector, scale
invariant feature transform(SIFT) descriptor and KLT tracker are used to extract and track the features from image sequence
taken by a stereo camera. Euclidean distance between SIFT descriptors of features in both images is computed for stereo
matching. An efficient closed-form method is adopted to estimate the frame-to-frame incremental motion in real time.
Moreover, additional techniques, including bucketing of features, RANSAC based outlier rejection and Kalman filtering, are
applied to improve accuracy and robustness of the estimated motion. The experimental results demonstrate the feasibility of
the proposed stereovision-based localization.