Opportunistic Seamless Localization

Phd Thesis Maarten Weyn



Abstract

Asset tracking, location based direct marketing, context aware data mining, people tracking, geofencing and navigation are only a few location-based services and applications which are used today. Besides outdoor localization, mostly based on GPS, a lot of indoor systems based on technologies such as WiFi, Ultra-Wideband, Radio-Frequency Identification (RFID), ultrasound and Wireless Sensor Networks are developed and used in different areas. These areas include healthcare, airline industry, transport and logistics, search and rescue, military, tracking and safety, and social networking applications such as friend finders.

Every technique and technology used for localization has its own specific properties and advantages, but also its specific disadvantages. One of the common disadvantages of many existing localization systems is the need for dedicated devices and proprietary infrastructure. Multi-modal systems which use the data coming from different systems and sensors will be the only possibility to allow affordable localization in different situations.

The future of localization systems most likely will evolve towards systems that can adapt and cope with any available information provided by mobile clients without the need to install any additional dedicated infrastructure. This type of localization is called opportunistic localization. It is defined as: "An opportunistic localization system is a system, which seizes the opportunity and takes advantage of any readily available location related information in an environment, network and mobile device for the estimation of the mobile device absolute or relative position without relying on the installation of any dedicated localization hardware infrastructure."

A lot of research has been done focussing on obtaining a very high accuracy with special dedicated hardware. In a lot of cases, this leads to research which obtains superfluous accuracy in an unrealistic lab environment using expensive, dedicated hardware. However, more and more people use portable devices which contain an increasing number of internal sensors. This thesis focuses on developing and demonstrating an adaptive localization algorithm, which uses commonly used portable devices.

Focusing on laptops, PDAs and smartphones, a selection of possible localization technologies is identified. WiFi, GPS, GSM and Bluetooth sensors are combined with internal accelerometers and a digital compass. A recursive Bayesian filter is used to seamlessly switch between and fuse the heterogeneous sensor data. This Bayesian filter, a particle filter, combines the possible sensor data coming from GPS, the accelerometers and compass by an adaptive motion model based on pedestrian dead reckoning. The sensors which can be used to infer a location (WiFi, GPS, GSM and Bluetooth) are combined using an adaptive measurement model. The WiFi, GSM and GPS measurement model use a Kernel method based algorithm. Bluetooth is mainly used for object binding.

Besides explaining the different algorithms, three experiments are discussed. The first one tests the adaptive character of the fusion and benchmarks it to the singular technologies. This experiment shows the possibility to obtain a mean accuracy of 3 m without the need to install any additional dedicated hardware for localization. The two other experiments focus on the opportunistic WiFi measurement model which is proposed in order to tackle hardware dependencies when using a pattern matching approach. The experiments show an improvement of more than 14% compared to another particle filter, a Kalman filter and two k-Nearest Neighbour implementations

Finally, a dynamic recalibration algorithm is proposed which needs no additional hardware, sensors, access point locations or user input to maintain an up-to-data signal-space to physical-space mapping. A possible mapping can be a WiFi fingerprint database.

All work presented in the thesis focuses on the use of multi-modal, opportunistic sensor data to create a ubiquitous localization system.

View Full Text
BibTex


More info: maarten dot weyn at uantwerpen dot be

Comments