Article Preview
TopIntroduction
Indoor object localization and tracking is a growing field with direct applications in both the private and public sectors. The ability to track people and objects indoors, in real-time, is a valuable tool both for business and government. In the public sector localization can provide lifesaving information to military, law enforcement, and emergency services (Harris, 2013), as well as inform traffic and city management. In the construction industry object tracking is a growing field with important applications to asset and building site management (Akanmu et al., 2013).
GPS provides highly-accurate and functional localization outdoors but performs poorly indoors where direct line-of-sight to GPS satellites is not possible. As there is no direct means to allow GPS to function indoors, novel indoor localization systems must utilize a completely different methodology. Alternatives to GPS primarily utilize existing wireless systems and expand their functionality to include localization. Prominent methods include IEEE 802.11 wireless LAN, ultrasonic and infrared methods, laser range finding, and Radio Frequency Identification (RFID) (Sanpechuda and Kovavisaruch, 2008).
All of these methods however, suffer from noise and multipath issues. To combat this many turn to the Kalman filter (KF). The KF is a predictive filtering model that can be applied to noisy systems to predict the actual system state based on recursive sampling. Used widely in GPS tracking, navigational guidance systems, and image detection, it is a powerful tool for noise reduction and system control. When applied to localization problems KF can greatly reduce the effects of multipath and time of arrival (TOA) errors and increase the feasibility of these systems for real-world applications (Bekkali et al., 2007; Lee and Park, 2006).
Non-linear Kalman filters have been applied to ultrasonic localization networks, similar to the MIT Cricket (Ko et al., 2006), laser rangefinding robots (Martinelli, 2008), and wireless LAN patient tracking (Paul and Wan, 2008).
RFID localization involves the use of multiple RFID readers set up in a grid pattern. Using the readers a system can track the location of a tag within the grid using TOA. Non-linear KF have been applied to such systems and found to reduce noise and increase localization accuracy (Nick et al., 2011a,b).
Existing work has used only simulated data and statistical models and has lacked real-world experimentation. This paper presents the application of linear and non-linear KF to a commercial RFID localization system operating in indoor environments. The RFID system is set up in an indoor environment containing noise and NLOS situations. Real-time data is collected, and KF techniques are designed based on the output and statistics of the system. The performance of the KF is analyzed and linear and non-linear implementations of the KF are also compared.