A Line in the Sand:A Wireless Sensor Network for TargetDetection, Classification, and TrackingA. Arora, P. Dutta, S. Bapat, V. Kulathumani, H. Zhang, V. Naik, V. Mittal, H. Cao, M. Demirbas1M. Gouda, Y. Choi2T. Herman3S. Kulkarni, U. Arumugam4M. Nesterenko, A. Vora, and M. Miyashita51Department of Computer Science and Engineering, The Ohio State University{anish,duttap,bapat,vinodkri,zhangho,naik,mittalv,caohu,demirbas}@cse.ohio-state.edu2Department of Computer Sciences, The University of Texas at Austin{gouda,yrchoi}@cs.utexas.edu3Department of Computer Science, University of [email protected] of Computer Science and Engineering, Mi chigan State University{sandeep,arumugam}@cse.msu.edu5Department of Computer Science, Kent State University{mikhail,avora,mmiyashi}@cs.kent.eduAbstract. Intrusion detection is a surveillance problem of practical import that is well suited towireless sensor networks. In this paper, we study the application of sensor networks to the intrusiondetection problem and the related problems of classifying and tracking targets. Our approach is basedon a dense, distributed, wireless network of multi-modal resource-poor sensors combined into lo oselycoherent sensor arrays that perform in situ detection, esti m ation, compression, and exfiltration. Weground our study in the context of a security scenario called “A Line in the Sand” and accordinglydefine the target, system, environment, and fault models. Based on the performance requirements ofthe scenario and the sensing, communication, energy, and computation ability of the sensor network,we explore the design space of sensors, signal processing algorithms, communications, networking, andmiddleware services. We introduce the influence field, which can be estimated from a network of binarysensors, as the basis for a novel classifier. A contribution of our work is that we do not assume a reliablenetwork; on the contrary, we quantitatively analyze the effects of network unreliability on applicationp erform ance. Our work includes multiple experimental deployments of over 90 sensors nodes at MacDillAir Force Base in Tampa, Florida, as well as other field experiments of comparable scale. Based onthese experiences, we identify a set of key lessons and arti culate a few of the challenges facing extremescaling to tens or hundreds of thousands of sensor nodes.1 IntroductionDeeply embedded and densely distributed networked systems that can sense and control the environment,perform local computations, and communicate the results will allow us to interact with the physical worldon space and time scales previously imagined only in science fiction. This enabling nature of sensor actuatornetworks has contributed to a groundswell of research on both the system issues encountered when buildingsuch networks and on the fielding of new classes of applications [1,2]. Perhaps equally important is that theenabling nature of sensor networks provides novel approaches to existing problems, as we illustrate in thispaper in the context of a well-known surveillance problem.Background. The instrumentation of a militarized zone with distributed sensors is a decades-old idea,with implementations dating at least as far back as the Vietnam-era Igloo White program [3]. Unattendedground sensors (UGS) exist today that can detect, classify, and determine the direction of movement of in-truding personnel and vehicles. The Remotely Monitored Battlefield Sensor System (REMBASS) exemplifiesUGS systems in use today [3]. REMBASS exploits remotely monitored sensors, hand-emplaced along likelyenemy avenues of approach. These sensors respond to seismic-acoustic energy, infrared energy, and magneticfield changes to detect enemy activities. REMBASS processes the sensor data locally and outputs detectionand classification information wirelessly, either directly or through radio repeaters, to the sensor monitoringset (SMS). Messages are demodulated, decoded, displayed, and recorded to provide a time-phased record ofintruder activity at the SMS.Like Igloo White and REMBASS, most of the existing radio-based unattended ground sensor systemshave limited networking ability and communicate their sensor readings or intrusion detections over relativelylong and frequently uni-directional radio links to a central monitoring station, perhaps via one or more simplerepeater stations. Since these systems employ long communication links, they expend precious energy duringtransmission, which in turn reduces their lifetime. For example, a REMBASS sensor node, once emplaced,can be unattended for only 30 days.Recent research has demonstrated the feasibility of ad hoc aerial deployments of 1-dimensional sensornetworks that can detect and track vehicles. In March 2001, researchers from the University of Californiaat Berkeley demonstrated the deployment of a sensor network onto a road from an unmanned aerial vehicle(UAV) at Twentynine Palms, California, at the Marine Corps Air/Ground Combat Center. The networkestablished a time-synchronized multi-hop communication network among the nodes on the ground whose jobwas to detect and track vehicles passing through the area over a dirt road. The vehicle tracking informationwas collected from the sensors using the UAV in a flyover maneuver and then relayed to an observer at thebase camp.Overview of the paper. In this work, we define, investigate, design, build, and field a dense, dis-tributed, and 2-dimensional sensor network-based surveillance system using inexpensive sensor nodes. Suchan approach relaxes the 1-dimensional constrained motion model and instead offers fine-grained detectionand tracking within an area but along any arbitrary 2-dimensional path. In this model, intrusion data areprocessed locally at each node, shared with neighboring nodes if an anomaly is detected, and communicatedto an exfiltration gateway with wide area networking capability. The motivation for this approach comes fromthe spatial- and temporal-locality of environmental perturbations during intrusions, suggesting a distributedapproach that allows individual sensor nodes, or clusters of nodes, to perform localized processing, filtering,and triggering functions. Collaborative signal processing enables the system to simultaneously achieve bettersensitivity and noise rejection, by averaging across time and space, than is possible with an individual nodewhich averages only across time.Our approach thus demonstrates how dense,