Abstract: This paper proposed a algorithm for detecting moving object based on camera
rotation motion. First, the rotational parameter models are built for global motion images.
Secondly, SIFT point pairs are established based on motion prediction between successive frames
and RANSAC algorithm is used to eliminate the outliers. With the least square method precisely
computing the motion compensation, an updating strategy based on residual image, which updates
the feature point set sequentially, can adjust the change of the background. Finally, motion object
is detected by Frame difference method. This novel algorithm remains the advantages of SIFT and
raises the detection rate significantly. The experiment results demonstrate that this algorithm can
effectively detect objects while achieving real-time performance.

Key words: moving object detection, motion compensation, camera rotational model, SIFT