Basic Principles of Sensor Output Analysis
The first area of interest is the Database.
To allow the data to be analysed we must first store it somewhere. To do this Dundee is using what is known as a database. The Sensor Network supplies the database with data from the sensors. Each database record contains three pieces of information:
Timestamp: Every time a signal is sent to the database, the time at which the sensor was activated is stored.
Device ID: This is a unique number to identify the sensor.
Value: This notes the new sensor reading. A sensor reading will initially have no value.
Only a change in the sensor value is recorded in the database. The reason behind this is obvious if you consider a movement detector activating every second. Each time the sensor activates, it would send a signal to the database giving it a value depending on whether it has detected movement or not. Now for the majority of the time, if the sensor is not detecting anything it is sending a null value to the database and this is being recorded each time. Over a period of time the storage requirements for such a database would become huge as it is filled with useless null values.
To solve this problem, only changes in a sensor value are recorded. This change is then recorded as the information of most relevance.
The second area of interest is Data Analysis.
First of all, software is used to create mathematical models that give the probability of what the user may be doing or where they may be going within the home.
Data mining is another area of Data Analysis that is required. Data mining involves going through past results and test data. The aim is to classify data which is consistent throughout each test of the Sensor Network. Consistent data is important as it is used to train the mathematical models, increasing their accuracy in the future.
