A Multi-modality Framework for Energy Efficient Tracking in Large Scale Wireless Sensor Networks Lizhi Yang, Chuan Feng, Jerzy. W. Rozenblit, Jianfeng Peng Abstract- This paper considers the problem of tracking real- world objects in a large scale area using distributed wireless sensor networks. Due to the limited power supply of wireless sensors, prediction based tracking mechanisms have been com- monly used to conserve the energy consumption of the tracking algorithm. On the other hand, in order to preserve the quality of tracking (QoS), appropriate recovery approaches have to be incorporated into the tracking algorithm since the prediction may fail due to network topology changes, blind areas, the uncertainty and unpredictability of real-world objects' motion, etc. In this paper, a multi-modality tracking framework is pro- posed and an n-step prediction tracking algorithm is evaluated in the framework. The proposed framework is suitable for the tracking system in which sensors are randomly deployed. This paper exhibits how the network of multi-modality wireless sensors can reduce the power consumption of the tracking and preserve the quality of tracking as well. I. INTRODUCTION In the paper, we consider the problem of tracking real- world objects in a large scale area using distributed wireless sensor networks (DWSNs) in which sensors are randomly deployed. The sensors have abilities to sense the environment in various modalities, process the information, and forward it to a certain node for further processing[1 1]. Compare to a single-modality sensor network that can only provide partial information of the environment, a multi-modality sensing system can obtain more complete descriptions of the mon- itored environment through combining the fused data from various sensors with different capabilities and strengths[10]. Thus, a multi-modality wireless sensor network architecture can offer more flexibility and more resources for various tracking applications. However, when designing a tracking algorithm for specific tracking applications such as border control, battle field surveillance or traffic flow measuring, there are several constraints that are needed to be considered. Some of these constraints are inherent from the nature of wireless sensors, e.g., the sensors may have a limited power supply, a limited communication bandwidth or a limited computational power. Therefore, the algorithm must be designed to expend as little energy as is possible in order to maximize network's lifetime. Moreover, since surveillance and tracking systems are likely to be deployed in a critical or hostile environment where functional failures are vital sometime, the design priority should be given to the quality and the reliability of tracking[1]. Thus, in order to improve the energy efficiency Lizhi Yang, Chuan Feng, Jerzy W. Rozenblit and Jianfeng Peng are with the Department of Electrical and Computer Engineering, University of Arizona, Tucson, AZ 85721, USA (phone: 520-626-2117; fax: 520-626- 2117; email: {lyang, fengc, jr, jpeng} 0ece.arizona.edu). of the tracking algorithm as well as preserve the quality of tracking, it is necessary to develop an integrated framework that takes into account some specific important issues of wireless sensor system design. A. Decentralized wireless sensor network architectures A fully decentralized sensor network is defined in [12] as a system in which the data is processed by local sensors and global results are available locally. While a centralized architecture is theoretically optimal and also conceptually simple[8], it is not suitable in a large scale area because of the limited communication bandwidth of wireless sensors. Moreover, the failure of the fixed superior node may imply the failure of the entire system. On the other hand, given a decentralized architecture, it is able to utilize dynamic head selection techniques to enhance the robustness of tracking since there is no local point of failure leading to the global failure. While each sensor node has a limited communication bandwidth, it is capable of coordinating with other nodes to have the global results. In addition, each wireless sensor has its own processors to fuse the data from diverse sensors with a lighter processing load. Consequently, a decentralized architecture offers more scalabilities than a centralized ar- chitecture, i.e., it is more adaptive to large scale tracking applications. B. Sensor deployment strategies A deployment strategy decides how to deploy sensors including where and how many to deploy in a specific area and may vary with the application considered. It is critical for tracking applications since the positioning of sensor nodes affects sensing coverage, communication and computing cost[15]. A strategy can be predetermined or undetermined respectively when the environment is known or unknown[3]. In this paper, we focus our attention to the application of tracking real-world objects, e.g., tracking moving targets using huge numbers of sensors with a small detection range. Typically, sensors are distributed randomly, in a large scale region to be monitored. The possible scenario can be short range micro sonar sensors or acoustic sensors deployed in a country border to detect and track illegal intrusions. Ideally, sensors will be dropped from aircrafts or vehicles without any further adjustment. In many such contexts it will be far easier to deploy larger numbers of nodes initially than to deploy additional nodes or additional energy reserves at a later date [2]. Nevertheless, a random deployment strategy may lead to severe coverage problems subject to sensors' communication and detection constraints. 1-4244-0065-1/06/$20.00 C 2006 IEEE 916