oalogo2  

AUTHOR(S): 

Nenad Katanic, Krešimir Fertalj

 

TITLE

Towards Physical Intrusion Detection Method Based on Machine Learning and Context-Aware Activity Recognition in Real-Time

pdf PDF

KEYWORDS

activity recognition, machine learning, context-aware, real-time, dense-sensing, accelerometer, physical intrusion

ABSTRACT

Sensor-based human activity recognition is getting increasingly popular in various applications. Most of the related work within dense-sensing based approaches assume that large number of different multimodal sensors are placed on the objects in the environment (which is rarely the case in today’s real life home environments), that sensor data is not processed in real-time and that activity to be classified is always performed within the same context, thus perform poorly when tested in real life scenarios. In this paper we report on the current status and future steps towards a generic context-aware method for human activity recognition, based on a real-time raw sensor data stream coming from a minimum number of sensors placed in the environment. We propose a hybrid method based on state-of-the-art data-driven and knowledge-driven approaches. Proposed method is being developed and will be validated on the example of the application for robust physical intrusion detection on home doors in real life environment.

Cite this paper

Nenad Katanic, Krešimir Fertalj. (2016) Towards Physical Intrusion Detection Method Based on Machine Learning and Context-Aware Activity Recognition in Real-Time. International Journal of Signal Processing, 1, 196-202