Action and simultaneous multiple-person identification using cubic higher-order local auto-correlation

We propose a new method - cubic higher-order local auto-correlation (CHLAC) - to address three-way data analysis. This method is a natural extension of higher-order local auto-correlation (HLAC) (N. Otsu and T. Kurita, 1988), which deals only with two-way data. Both methods use "correlation" to summarize relative positions or motions within a local data region, and these can be calculated simply with a low computational load. Moreover, our new method (CHLAC) offers several preferable properties as well as HLAC: shift-invariance to data (rendering the method segmentation-free), additivity for data, and robustness to noise in data. In this study, we applied this method to action and simultaneous multiple-person identification from a motion-image sequence through the property of data additivity. Experimental results showed that this method performed well.