Motion History Histograms for Human Action Recognition

Novel methods for feature extraction based on motion history images and evaluation with regard to altering viewing angles

“…In [2] several important variants of the MHI are presented, like the motion energy image (MEI), which represents the binary version of the MHI . [3] refines the idea of the MHI with respect to the temporal aspect which results in the motion history histogram (MHH) that provides information about temporal occurrences of movements in an image.…”

“…Nevertheless, it shows good classification rates, especially for differing movements and in combination with other feature extraction methods (see [3]). The main disadvantage of the MHI is the limited discrimination rates regarding similar movements such as running or walking.…”

“…In [3] a support vector machine (SVM) is used. Compact representations of the motion features, so-called feature vectors, are generated from the training data.…”

Human motion recognition using support vector machines

“…In particular, Ref. [14] proposed a new feature extracting algorithm called motion history histogram, which is an improvement over the motion history image [12] in terms of encoding the time span of movement, and provided an FPGA implementation. Given a set of features that is believed to be able to characterize the motion of interest, most recognition algorithms are based on either template matching [12,13] or state-space matching that usually uses Hidden Markov Models (HMM) [11].…”

“…The majority of relevant work in motion recognition focuses on motion feature selection, including extracting features from 2D tracking data [1][2][3][4][5][6][7] or 3D tracking information [8,9], or extracting motion information directly from images [10][11][12][13][14]. In particular, Ref.…”