Efficient fall detection based on event pattern matching in image streams

Abstract

This research proposes sliding window fall detection match (SW-FDM), a rule-based fall detection method based on event pattern matching from the human body posture image streams. Fall and post-fall (long lie) rules are expressed as a pattern, and complex event processing (CEP) systems is required to process them in a temporal ordering relationship as well as in a large scale of streams. Those patterns can be detected with event selection strategies such as Skip Till Next Match, and Skip Till Any Match. However, existing strategies generate either duplicate or missing alarms. In addition, processing cost is a severe problem when the size of event streams is large. SW-FDM applies a concept of sliding window, so it is able to detect correct matches constantly, and it reduces processing cost without a duplicate computation. The experiment proved that SW-FDM results in a more accurate and efficient performance. Also, it was shown that the improvement of efficiency becomes greater as an increasingly larger size of data sources are sent to the implemented CEP systems.