Tampering Classification Using Accelerometer Data

Abstract: Network Video Door Stations are IP-based door stations for two-way communication, identification, and remote entry control. They have a number of different sensors e.g video, audio and accelerometer that can measure external data. These door stations serve multiple purposes including that of acting as a security feature and they are often exposed to malicious intent. The goal of this master thesis is to utilize machine learning techniques to classify tampering events on the Axis A8105-VE using a three dimensional low cost MEMS accelerometer. A reliable system using support vector machine was developed and tested on the Axis A8105-VE. The classification scheme developed achieved an average accuracy of 99.8% with a response time of 1.624 seconds. The data used in this thesis contains 2119 observations introducing 13 different environments. The feature vector used in the binary classification consists of 41 features focusing on probabilistic-, periodic-, frequency- and generic measurements of the time-series signal based on the accelerometer data.