Automatic Recognition of Abnormal Human Actions with Semi-supervised Training: A Literature Review
Diana K. Guevara-Flores, Josefina Guerrero-García, David Pinto-Avendaño · Research in Computing Science · 2019
In this paper is presented the literature review and the first tests performed for the development of a method consisting of differentiating typical actions in a given environment from those that can be categorized as abnormal or atypical.The objective of this method will be to determine which are the typical situations taking into account temporal and spatial information of a given environment to generate an alarm when a potentially undesirable situation arises.The main difference between this system and those existing in the literature is that it does not seek recognition of pre-established actions, such as running or sitting, and that the system can adapt to different environments.For the development of this research the use of Deep Learning is proposed and due to the complexity of the attributes required by the classifier, the use of a semi-supervised method is proposed.