Back to search

National Technical University of Athens (NTUA)

Τεχνικές χαμηλές κατανάλωσης ενέργειας για ανίχνευση προσώπων και κατηγοριοποίηση οχημάτων στα ασύρματα δίκτυα αισθητήρων

Abstract

dc:description

The subject of this Doctoral Thesis is seismic personnel detection and seismic and acoustic vehicle recognition and classification in Wireless Sensor Networks, with emphasis on energy efficiency and low power consumption. These applications form part of a complete physical security system used to protect large areas of interest such as large facilities or borders, where permanent power supply is not always available. Wireless Sensor Networks consist of small, battery powered devices, also known as Sensor Nodes, which are able to sense the surrounding environment with electronic sensors, process and/or store data and communicate wirelessly. The span of life of the nodes depends on their hardware, the processing capabilities of the processing unit and the algorithms used, but mostly on the way hardware and communications are used by the node's software. The unique characteristics of WSNs make them ideal for physical security applications. However, personnel and vehicle detection is quite demanding in signal processing and Sensor Nodes often require a faster and more capable microprocessor, which consume extra power reducing their operational life. In this work we investigate both hardware and software low- power techniques, which will enable successful personnel detection and vehicle recognition on a simple Sensor Node, equipped with acoustic and seismic sensors and a simple 8-bit microcontroller. In the field of personnel detection the special characteristics of the seismic vibrations, produced by a walking person, are examined. Ground vibrations are recorded using a piezoelectric accelerometer and processed accordingly by the wireless sensor node, which decides if they are generated from a walking person or some other seismic event. The proposed algorithms are tested in simulation level with real signals. They are finally implemented in hardware on the Mica2Dot sensor node. Using power saving techniques such as scheduled disabling of unused subsystems of the node, average power consumption can be reduced by 82%, reaching 9,89 mW. Maximum estimated battery life of the Personnel Detection Node can reach 37,53 days, when powered by 2 simple 2700 mAh, AA type batteries. In the field of acoustic and seismic vehicle recognition approaches other than traditional frequency-domain signal processing or wavelets are investigated. Such methods are very demanding in processing resources and energy. A time-domain signal processing and encoding method, used in the speaker identification task, known as TESPAR, is introduced instead. The method is adapted to the new environment and further extended to accurately encode the acoustic and seismic signatures of vehicles almost in real time. The special characteristics of the encoded signal, that enable good discrimination between different vehicle types, are grouped in a one-dimensional matrix, the feature vector. Classification of the feature vectors is performed by an Artificial Neural Network and a minimum distance classifier. Simulation tests with real acoustic and seismic signals from two vehicles, a heavy wheeled truck and a tracked vehicle, showed that performance can reach as high as 88%. The performance is comparable with existent, more complex methods using the same dataset. The method is implemented in hardware on the Mica2 sensor node and the resulting Vehicle Recognition Node is able to recognize three classes of vehicles (car, bus, and motorcycle) based on their acoustic signature with an accuracy of 62-90%. Average power consumption can reach 3,35 mW when classification is done by the sensor node and maximum battery life is estimated to 110 days.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2007

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Mazarakis, Georgios
  • Μαζαράκης, Γεώργιος

Subjects

dc:subject × 18

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/16935

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Mazarakis, Georgios; Μαζαράκης, Γεώργιος. Τεχνικές χαμηλές κατανάλωσης ενέργειας για ανίχνευση προσώπων και κατηγοριοποίηση οχημάτων στα ασύρματα δίκτυα αισθητήρων. National Technical University of Athens (NTUA), 2007. http://hdl.handle.net/10442/hedi/16935