Real-Time Speed Variation Detection of Moving Human Using Polar Projection Feature

This paper proposes a method to detect speed variation of a moving human target, based on polar projection feature. Through even-grid-polar mapping, two polar projection feature images, namely r-projection and q-projection, are obtained respectively. It is found that the frequency of double-peak in q-projection feature image series in a motion cycle is sensitive to speed variation and can be utilized to distinguish running from walking. Experimental results show that the proposed method is robust to rotation, scaling, translation and small non-rigid deformation in spite of multi-angle cases.