Corner feature is a usually used local invariant. It is widely applied for national defense and industry technology. Using different methods: the splitting-merging algorithm

the improved Hough transform method and the extended corner invariant algorithm

different corner features is extracted to recognize occluded objects in this paper. Finally

Dempster-Shafer evidential reasoning method is adopted for data fusion at report level. This method can use information from different sources and decrease uncertainty of the recognition system. The classification experiment shows that this method effectively improves the matching precision of the recognition system.