Selection method for low complexity feature detection algorithms

Interactive smart TV is a growing field which allows users to click on objects as they appear on the screen and get a context dependent reaction. Automated systems allowing preparation of regular video clips for interactive-TV must possess object tracking functionality, which can be provided by the well known feature detection algorithms SIFT or SURF. SURF has a significantly lower computational complexity than SIFT, and under certain conditions has still managed to equal SIFT's feature detection fidelity. However, these conditions are not always valid in an interactive TV preparation scenario, and consequentially SURF can fail where SIFT would not have. In this work we propose a low complexity, automatic SIFT and SURF algorithm selection method. The method applies selection criteria to the particular object being tracked and reverts to SIFT in cases where SURF might fail. The resulting algorithm has a reduced overall complexity close to SURF's while preserving the highest possible tracking performance available with SIFT. Our method has been successfully tested and implemented as part of a larger interactive TV preparation system.