An Action Classification Algorithm Based on MEI and LPP

A foreground extraction algorithm based on back-ground subtraction and edge detection is proposed to obtain the foreground with a little change. An action classification method based on Motion Energy Image (MEI) and Locality Preserving Projections (LPP) is proposed. The high dimensional feature space is non-linearly reduced to lower dimensional space, which outperforms PCA and 2DPCA. The nearest-neighbor classifier is adopted to distinguish different actions. This algorithm needn´t extract the period of the video, which is indispensable in some other methods. Experimental results show that the algorithm is simple, and achieves higher classification accuracy with less running time.