Frame Decimation for Structure and Motion

A frame decimation scheme is proposed that makes automatic extraction of Structure and Motion (SaM) from handheld sequences more practical. Decimation of the number of frames used for the actual SaM calculations keeps the size of the problem manageable, regardless of the input frame rate. The proposed preprocessor is based upon global motion estimation between frames and a sharpness measure. With these tools, shot boundary detection is first performed followed by the removal of redundant frames. The frame decimation makes it feasible to feed the system with a high frame rate, which in turn avoids loss of connectivity due to matching difficulties. A high input frame rate also enables robust automatic detection of shot boundaries. The development of the preprocessor was prompted by experience with a number of test sequences, acquired directly from a handheld camera. The preprocessor was tested on this material together with a SaM algorithm. The scheme is conceptually simple and still has clear benefits.