Rozpoznávání lidských činností pomocí detekce anomálii

Abstract

Methods for detecting anomalies are well described for both static and dynamic data of various formats, but this is not the case for human activities. Each person performs different activities and this is the problem of anomaly detection when the same activity is performed differently by different people. For example, in waving, a hand may be raised less or more, or it may wave at a faster or slower rate.

This work deals with anomaly detection in 4 datasets of real activities using neural network. For the used datasets a basic statistical analysis was carried out to determine the predominant lengths of activities according to the number of frames, their averages, medians and the amount of activities divided to groups according to the number of frames.