Application of the Unusual Motion Detection Using CHLAC to the Video Surveillance

Abstract

Cubic Higher-Order Local Auto-Correlation (CHLAC) is feature vector that simultaneously represent motion and shape. The system learns a sample set of “usual motion” to create a “usual subspace” with PCA. Feature vectors are then similarly extracted from unknown input data, and accurate detection of “unusual motion” is achieved by measuring the deviation from the usual subspace. Therefore, by defining unusual motion as “motion that is outside the usual motion,” this method can detect unusual motion without an actual model of unusual motion, which differs depending on the situation, and furthermore, is difficult to define. This paper reports on the fast CHLAC that we have developed, so that these capabilities of CHLAC can be put to practical use as an unusual motion detection system that operates in real time. This paper also demonstrated the effectiveness of this method through example tests, conducted using real images both indoors and outdoors.