A real-time human-motion recognition method is proposed that uses fuzzy associative inference. It transforms space-time patterns into state-transition patterns, which are then recognized by means of fuzzy associative inference using associative memories. The tracking data is given as time-series data, from which the characteristic states are extracted. Each human motion has a specific state-transition pattern that consists of characteristic states. To recognize these motions, the specific state-transition patterns of the motions are defined as fuzzy rules and these fuzzy rules are implemented in a fuzzy associative memory system. This method is independent of the person being measured and the speed of the motion. In real-time experiments, this method was able to recognize three basic tennis motions (forehand stroke, backhand stroke, and smash) for six unspecified people. The recognition ratio of the fuzzy associative memory system is better than that of conventional fuzzy inference and a multi-layer perceptron.
|Number of pages||6|
|Publication status||Published - 1 Dec 1994|
|Event||Proceedings of the 3rd IEEE Conference on Fuzzy Systems. Part 3 (of 3) - Orlando, FL, USA|
Duration: 26 Jun 1994 → 29 Jun 1994
|Conference||Proceedings of the 3rd IEEE Conference on Fuzzy Systems. Part 3 (of 3)|
|City||Orlando, FL, USA|
|Period||26/06/94 → 29/06/94|