Recently, new wireless communication techniques, such as Bluetooth and ZigBee, have become widely used. Their transmission ranges are relatively narrower than those of traditional wireless communication techniques, such as mobile phone and wireless LAN. These techniques are available for everyone unlike mobile phone networks which belong to specific communication carriers. Moreover, we can introduce them with relatively low price. System administrators in various kind of facilities, universities, or companies can construct a new wireless infrastructure by appropriately using them. Such infrastructures accomplish ubiquitous environments in which every user can connect to the network anytime and anywhere. On the other hand, with the development of sensor techniques, sensors become smaller and more reasonable. We expect that we can realize an object tracking system using the wireless infrastructure in which lots of low-priced infrared sensors are embedded. In this thesis, we propose an object tracking method in a narrow-range area, namely a microcell, in which the deployed sensors can communicate each other using Bluetooth or ZigBee. From the viewpoint of implementation and operation costs, we first observe properties of object mobility in the micro-cell and make modeling them based on the observation results. In case of the micro cells, there are several factors that affect object mobility in them, such as object arrivals/departures, characteristics of their locations. The conventional mobility models do not consider these factors. From the observation result, we found that an object moves approximately along with a straight line and the object velocities follow a normal distribution. These characteristics indicate that we can reproduce object movements in the area by observing object arrivals and departures and estimating the correlation between them. The object arrivals and departures can be obtained relatively easily by setting sensors only to multiple boundary points of the micro-cell. To estimate the correlation between object arrivals and departures, we propose a Bayesian estimation based method which is extended to deal with time-series information. Additionally, we propose the reduction method of false estimations to improve the accuracy of the estimation. To evaluate our object tracking method, we use actual object flows and artificial object flows: the actual object flows are obtained from the observation; the artificial object flows are yielded by a traffic generator. We obtained actual object flows by observing a location where the traffic is heavy at busy time. The proposed method could completely estimate all of the correlations between object arrivals and departures that were obtained from the observation. Through several simulation experiments using the artificial object flows, we further clarified that the proposed method was more suitable for a situation where the object flows were biased. Moreover, we could improve the estimation accuracy by the reduction method of false estimations. More precisely, we accomplished the estimation accuracy of about 90 % when the object flows were biased and the object cell-arrival rate was not more than 3 [objects/s]. This accuracy was 58 % higher than that of the situation where the object flows were not biased.

Satoshi Aihara, Observation-Based Modeling of Object Mobility in a Micro-cell and Its Application to Object Tracking, Master's Thesis, March 2008.