Discovery of Correlation from Multi-stream of Human Motion

Abstract

Human motion data is practically used in some domains such as movies, CGs and so on. Creators use motion data to produce exciting and dangerous scenes without real actors. Human motion data has following features: 1. correlation between body parts We control all body parts in cooperation and motion data consists of the information of cooperation. For example, we swing both arms in turn to keep walking straight. There exists correlation between arms and legs. 2. correlation in the ow of contents There is tendency that certain movement likely to occurs after another movement. For example, once we raise our hands, we probably put our hands down in certain interval of time. This is correlation in time flow of motion.

Because human motion data has these kinds of features, motion data should be treated as multi-stream [4]. Multi-stream includes unexpectedly frequent or infrequent co-occurrences among different streams. This means that an event on one stream is related to another events which locate on other streams and seem to have nothing to do with the former event. Time series pattern of stock price is a good example. Rise and fall of price on some stocks obviously cause price of one stock to rise and fall. If we analyze the multi-stream of time series for some stock price and can discover correlation between all streams, the correlation help us to decide better time to buy stocks.