Detection & classification of imperceptible motion using video decomposition

As human vision system (HVS) is highly sensitive to motion, motion saliency is an important field of research in video signal processing. But, HVS is particularly insensitive to subtle motions with low amplitude. Though, in many practical fields, e.g., biomedical science, earth science, plasma science etc., these low amplitude motions are significant for predicting certain crucial events, most of the signal processing methods fail to analyze them as they are difficult to detect in natural scenes. Thus, a specialized manual intervention is generally required to analyze these data. The situation worsens in presence of noise, inherent to any imaging system, as it is difficult to distinguish imperceptible motions in noisy environment. In this paper we propose a robust method to detect and classify imperceptible motion in a video sequence. The proposed algorithm exploits a total variation (TV) based video decomposition to detect the motion in a scene and detected motion is classified by training a support vector machines (SVM) after the detection. This classification of subtle motion can be used in several areas for diagnosing abnormalities.