PIR-Based Motion Patterns Classification for AmI Systems
2013
The analysis of human motion has become a major application area in computer vision. The vast majority of applications require using video cameras as main sensor for acquisition, analysis and detection of human motion. Some applications are very sensitive to the acquisition of images of people that may feel violated his right to privacy. Especially in ambient intelligence applications, non intrusive nature, whose scope is the user’s home. The scope of this work was constrained to the analysis of human motion obtained from signals acquired for PIR sensors. We propose a method to classify PIR signals regarding two parameters: speed and distance from sensor to the subject. Our method allows to get compact devices that perform local computation and are able to transmit middle abstraction level context information in an efficient way. Signals from a PIR sensor array generated by several subjects have been registered to get patterns. Subjects moved at a certain speed at a given distance from sensors, so patterns are indexed by these two parameters. The patterns are then used to classify other tests signals regarding the mentioned parameters. The hit rate, for the method which combines information from all the sensors in the array, results in 79% for distance classification, 96% for speed classification and 77% for classification regarding both parameters simultaneously. We consider that this is a satisfying result for a non intrusive method.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
11
References
2
Citations
NaN
KQI