R E S E A R C H F E A T U R E
• It provides online real-time monitoring. For exam-ple, when the system is capturing events, the static sensors can immediately inform users where the events are occurring, and the mobile sensors can later provide detailed images of these events.
• It’s event-driven, in the sense that only when an event occurs is a mobile sensor dispatched to capture images of that event. Thus, iMouse can avoid recording unnecessary images when nothing happens. • The more expensive mobile sensors are dispatched to
the event locations. They don’t need to cover the whole sensing field, so only a small number of them are required.
• It’s both modular and scalable. Adding more sophis-ticated devices to the mobile sensors can strengthen their sensing capability without substituting exist-ing static sensors.
Because mobile sensors run on batteries, extending their lifetime is an important issue. We thus propose a dispatch problem that addresses how to schedule mobile sensors to visit emergency sites to conserve their energy as much as possible. We show that if the number of emergency sites is no larger than the num-ber of mobile sensors, we can transform the problem to a maximum matching problem in a bipartite graph; otherwise, we group emergency sites into clusters
Incorporating the environment-sensing capability of wireless sensor networks into
video-based surveillance systems can provide advanced services at a lower cost than traditional
surveillance systems.The integrated mobile surveillance and wireless sensor system
(iMouse) uses static and mobile wireless sensors to detect and then analyze unusual events
in the environment.
Yu-Chee Tseng, You-Chiun Wang, Kai-Yang Cheng, and Yao-Yu Hsieh
National Chiao Tung UniversityT
he remarkable advances of microsensing micro-electromechanical systems (MEMS) and wire-less communication technologies have pro-moted the development of wireless sensor networks. A WSN consists of many sensor nodes densely deployed in a field, each able to collect environmental information and together able to sup-port multihop ad hoc routing. WSNs provide an inex-pensive and convenient way to monitor physical envi-ronments. With their environment-sensing capability, WSNs can enrich human life in applications such as healthcare, building monitoring, and home security.Traditional surveillance systems typically collect a large volume of videos from wallboard cameras, which require huge computation or manpower to analyze. Integrating WSNs’ sensing capability into these systems can reduce such overhead while providing more advanced, context-rich services. For example, in a secu-rity application, when the system detects an intruder, it can conduct in-depth analyses to identify the possible source. The “Related Work in Wireless Surveillance” sidebar provides additional information about other work in this area.
Our integrated mobile surveillance and wireless
sen-sor system (iMouse) consists of numerous static
wire-less sensors and several more powerful mobile sensors. The benefits of iMouse include the following:
iMouse: An Integrated
Mobile Surveillance and
Wireless Sensor System
so that one mobile sensor can efficiently visit each cluster.
SYSTEM DESIGN
Figure 1 shows the iMouse architecture, which con-sists of static and mobile sensors and an external server. The static sensors form a WSN to monitor the environ-ment and notify the server of unusual events. Each sta-tic sensor comprises a sensing board and a mote for communication. In our current prototype, the sensing board can collect three types of data: light, sound, and temperature. We assume that the sensors are in known locations, which users can establish through manual set-ting, GPS, or any localization schemes.1
An event occurs when the sensory input is higher or lower than a predefined threshold. Sensors can com-bine inputs to define a new event. For example, a sen-sor can interpret a combination of light and temperature readings as a potential fire emergency. To detect an explosion, a sensor can use a combination of temperature and sound readings. Or, for home security, it can use an unusual sound or light reading.
To conserve static sensors’ energy, event reporting is reactive.
Mobile sensors can move to event locations, exchange messages with other sensors, take snapshots of event scenes, and transmit images to the server. As Figure 2 shows, each mobile sensor is equipped with a Stargate pro-cessing board (www.xbow.com/Products/productsdetails. aspx?sid=229), which is connected to the following:
• a Lego car (http://mindstorms.lego.com/eng/ default.asp), to support mobility;
• a mote, to communicate with the static sensors; • a webcam, to take snapshots; and
• an IEEE 802.11 WLAN card, to support high-speed, long-distance communications, such as transmitting images.
The Stargate controls the movement of the Lego car and the webcam.
The external server provides an interface through which users can obtain the system status and issue com-mands. It also maintains the network and interprets the
(4) Move to emergency sites and take advanced actions (for example, take snapshots)
(1) Issue com mands to the network
(5) Report analyzed data through WLAN (3) Dispatch (2) Report events (2) Report events Report statuses Issue commands Home/office environment Internet Mobile sensors Event User Wireless link External server Static sensor A F G E D C B I H
Figure 1.The iMouse system architecture.Three main components make up the iMouse architecture: static sensors, mobile sensors, and an external server.The user issues commands to the network through the server (1). Static sensors monitor the environment and report events (2). When notified of an unusual event, the server notifies the user and dispatches mobile sensors (3), which move to the emergency sites, collect data, (4) and report back to the server (5).
tures, which the server suspects to indicate a fire emer-gency in the sensors’ neighborhoods.
The server notifies the users and dispatches mobile sensors to visit the sites. On visiting A and C, the mobile sensors take snapshots and perform in-depth analyses. For example, the reported images might indicate the fire’s source or identify inflammable material in the vicin-ity and locate people left in the building.
Each static sensor runs the algorithm in Figure 3. The server periodically floods a tree-maintenance message to maintain the WSN. It also records each static sensor’s location and state, which is initially set to normal. Tree-maintenance messages help the static sensors track their parent nodes. To distinguish new from old messages, tree-maintenance messages are associated with unique sequence numbers. The goal is to form a spanning tree in the WSN.
When a sensor receives an input above a threshold, indi-cating an event, the sensor reports that event to the server. To avoid sending duplicate messages, each sensor keeps a variable event flag to indicate whether it has already reported that event. When a sensor detects an event and the event flag is false, the sensor reports that event and sets the flag to true. The server collects multiple events and assigns them to mobile sensors in batches. When a mobile sensor visits an event site, it asks the local sensor to clear its event flag.
Mobile sensor dispatch and traversal problems
Because mobile sensors are battery powered, we assign them to emergency sites to conserve their energy as much as possible. Specifically, we consider a set L of m emergency sites to be vis-ited by a set S of n mobile sensors, where each site must be visited by one mobile sensor. We allow an arbitrary relationship between m and n. The goal is to maximize the mobile sensors’ total remaining energy after sites are visited.
Our dispatch solution depends on the relation-ship of m and n. When
m <= n, we can convert
the problem to one of finding a maximum matching in a weighted bi-partite graph G = (S
L, S L), where the
ver-tex set is S L and the edge set is the product S meanings of events from sensors. On detecting a
poten-tial emergency, the server dispatches mobile sensors to visit emergency sites to obtain high-resolution images of the scene. The dispatch algorithm also runs on the server.
System operations and control flows
To illustrate how iMouse works, we use a fire emer-gency scenario, as Figure 1 shows.
On receiving the server’s command, the static sensors form a treelike network to collect sensing data. Suppose static sensors A and C report unusually high
tempera-Figure 3.The algorithm executed by static sensors.Three types of messages activate a static sensor: tree-maintenance message, sensory input, and event message.
Sensory input
Tree-maintenance message Event message
No No No Yes Yes Yes Start
Check incoming messages/events
Send a status-change message to the server; set eventflag=TRUE
Above threshold?
eventflag=TRUE New sequence
number?
Set the sender as its parent node
Rebroadcast the tree-maintenance message
Relay the message to its parent node Lego infrared tower WLAN card Mote Stargate Lego car Webcam Infrared ray receiver USB hub Light detector 24 cm 24 cm
Figure 2.The mobile sensor. Attached to the Stargate processing board are a mote, a webcam, and an IEEE 802.11 WLAN card. A Lego car provides mobility.
L = {(si, lj)|si S, lj L}. We set the
weight of (si, lj) to ei– emv d(si, lj),
where eiis the current energy of si; emv
is the energy cost for a mobile sensor to move by one unit; and d(si, lj) is the
distance from si’s current location to lj.
The solution is the maximum match-ing P of G, which we can find through traditional maximum-weight match-ing solutions.2Alternatively, we can
set our objective to minimizing mobile sensors’ total moving distances. We can also use maximum-matching to achieve this by setting the weight of (si, lj) to –emove d(si, lj).
When m > n, some mobile sensors must visit multiple sites. To solve this problem, we divide emergency sites into n clusters (for example, by the classical K-means method) and assign each group to one mobile sensor. In this case, each mobile sensor’s cost will include moving to the closest site in each group and then traversing the rest of the sites one by one. Given a set of locations to be visited, we can use a heuristic to the traveling sales-man problem2to determine the
tra-versal order.
IMPLEMENTATION AND EXPERIMENTAL RESULTS
Our static sensors are MICAz motes (www.xbow.com/Products/ productdetails.aspx?sid=164). A MICAz is a 2.4-GHz, IEEE 802.15.4-compliant module allowing low-power operations and offering a 250-Kbps data rate with a direct-sequence spread-spectrum (DSSS) radio.
The Stargate processing platform consists of a 32-bit, 400-MHz Intel PXA-255 XScale reduced-instruc-tion-set computer (RISC) with a 64-Mbyte main memory and 32-64-Mbyte extended flash memory. It also has a daughterboard with an RS-232 serial port, a PCMCIA slot, a USB port, and a 51-pin extension connector, which can be attached to a mote. It drives the webcam through a USB port and the IEEE 802.11 WLAN card through its PCMCIA slot. The Stargate controls the Lego car via a USB port connected to a Lego
Related Work in Wireless Surveillance
Traditional visual surveillance systems continuously videotape scenes to capture transient or suspicious objects. Such systems typically need to automatically interpret the scenes and understand or predict actions of observed objects from the acquired videos. For example, Wu-chi Feng and his colleagues proposed a video-based surveillance network in which an 802.11 WLAN card transmits the information that each video camera captures.1
Researchers in robotics have also discussed the surveillance issue.2 Robots or cameras installed on walls identify obstacles or humans in the environment. These systems guide robots around these obstacles. Such systems normally must extract meaningful information from massive visual data, which requires significant computation or manpower.
Some researchers use static WSNs for object tracking.3,4These systems assume that objects can emit signals that sensors can track. However, results reported from a WSN are typically brief and lack in-depth informa-tion. Edoardo Ardizzone and his colleagues propose a video-based surveillance system for capturing intrusions by merging WSNs and video-processing techniques.5The system complements data from WSNs with videos to capture the possible scenes with intruders. However, cameras in this system lack mobility, so they can only monitor some locations.
Researchers have also proposed mobilizers to move sensors to en-hance coverage of the sensing field6and to strengthen the network connectivity.7Other work addresses the pursuer-evader game, in which a pursuer must intercept an evader in the field with the assistance of WSNs.8To our knowledge, no one has adequately addressed the integra-tion of WSNs with surveillance systems, which motivates us to propose the iMouse system.
References
1. W.C. Feng et al., “Panoptes: Scalable Low-Power Video Sensor Networking Technologies,” ACM Trans. Multimedia Computing, Comm., and Applications, vol. 1, no. 2, 2005, pp. 151-167.
2. J.H. Lee and H. Hashimoto, “Controlling Mobile Robots in Distributed Intelli-gent Sensor Network,” IEEE Trans. Industrial Electronics, vol. 50, no. 5, 2003, pp. 890-902.
3. X. Ji et al., “Dynamic Cluster Structure for Object Detection and Tracking in Wireless Ad-Hoc Sensor Networks,” Proc. IEEE Int’l Conf. Comm., IEEE Press, 2004, pp. 3807-3811.
4. W. Zhang and G. Cao, “DCTC: Dynamic Convoy Tree-Based Collaboration for Target Tracking in Sensor Networks,” IEEE Trans. Wireless Comm., vol. 3, no. 5, 2004, pp. 1689-1701.
5. E. Ardizzone et al., “An Integrated Architecture for Surveillance and Monitoring in an Archeological Site,” Proc. ACM Int’l Workshop Video Surveillance and Sensor Networks, ACM Press, 2005, pp. 79-85.
6. N. Heo and P.K. Varshney, “Energy-Efficient Deployment of Intelligent Mobile Sensor Networks,” IEEE Trans. Systems, Man and Cybernetics—Part A: Systems and Humans, vol. 35, no. 1, 2005, pp. 78-92.
7. P. Basu and J. Redi, “Movement Control Algorithms for Realization of Fault-Tolerant Ad Hoc Robot Networks,” IEEE Network, vol. 18, no. 4, 2004, pp. 36-44. 8. C. Sharp et al., “Design and Implementation of a Sensor Network System for
Vehicle Tracking and Autonomous Interception,” Proc. 2nd European Workshop Wireless Sensor Networks, IEEE Press, 2005, pp. 93-107.
infrared tower, as Figure 2 shows. An infrared ray receiver on the front of the Lego car receives commands from the tower, and two motors on the bottom drive the wheels.
Navigating a mobile sensor or robot is difficult without some auxiliary devices. David Johnson and colleagues used wallboard cameras to capture mobile sensors’ locations,3while
Jang-Ping Sheu and his colleagues suggested using signal strength to do so.4
Our current prototype uses the light sensors on the Lego car to navigate mobile sensors. We stick different col-ors of tape on the ground, which lets us easily navigate the Lego car on a board. In our prototyping, we implemented an experimental 6 6 grid-like sensing field, as Figure 4 shows. Black tape rep-resents roads, and golden tape repre-sents intersections. We constructed the system by placing two mobile sensors and 17 static sensors on the sensing field. For static sensors, a light reading below 800 watts simulates an event, so we cover a static sensor with a box to model a potential emergency.
We use a grid-like sensing field and a grid-like static sensor deployment only for ease of implementation. In general, the static WSN’s topology can be irregular.
Three factors affect the mobile sensors’ dispatch time:
• the time that a mobile sensor takes to cross one grid-unit (about 26 centimeters),
• the time that a mobile sensor takes to make a 90-degree turn, and
• the time that a mobile sensor takes to make snap-shots and report the results.
In our current prototype, the times are 2.5, 2.2, and 4.0 seconds, respectively.
Figure 5 shows experimental results with one mobile sensor initially placed at (0, 0) and two mobile sensors placed at (0, 0) and (5, 5). We generate some random events and evaluate the dispatch time (from when the server is notified of these events to when all event sites are visited) and the average time of each site (from when an event is detected to when a mobile sensor visits the site). Clearly, using two mobile sensors significantly reduces dispatch and waiting times.
At the external server, users monitor the system’s sta-tus and control mobile sensors through a user interface, as Figure 6 shows. The user interface includes six major components:
• The configure area lets users input system configu-ration information, such as mobile sensors’ IP addresses, ports, and sensors’ positions.
• The system-command area provides an interface to let users control the overall system, such as issuing a Figure 4. A 6 6 grid-like sensing field used in our experiment.Colored tape placed
on the floor (black for roads, golden for intersections) is used to navigate the Lego car.
Figure 5. Experimental performance of (a) dispatch time and (b) average waiting time. As the graphs show, using two mobile sensors reduces dispatch and waiting times.
Intersection
(golden tape) (black tape)Road
Emergency site Static sensor Mobile sensor 26 cm 26 cm (0,0) (5,5) 0 20 40 60 80 100 120 0 10 20 30 40 50 60 (a) (b) 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Number of event sites
Number of event sites
Total dispatch time (seconds)
Average waiting time (seconds)
Two mobile sensors One mobile sensor
Two mobile sensors One mobile sensor
tree-maintenance message, adjusting the WSN’s topol-ogy, and connecting and disconnecting a specified mobile sensor.
• The sensor-status area shows the current status of a static sensor being queried.
• The action-control area lets users control the mobile sensors’ actions, including movement and taking snapshots.
• The monitor area shows the WSN’s network topology and the mobile sensors’ patrolling paths. When a sen-sor detects an event, a fire icon appears in the corre-sponding site.
• The log area displays some of the system’s status mes-sages.
SIMULATION RESULTS
Our experiments considered only grid networks. To help us understand the sensor dispatch problem in gen-eral irregular WSNs, we developed a simulator. We eval-uated the average waiting time for an event location being visited and mobile sensors’ total energy con-sumption. We compared a greedy algorithm against the
K-means algorithm.
Given a set S of mobile sensors and a set L of emer-gency sites, the greedy algorithm contains a sequence of iterations. In each iteration, we assigned the mobile sen-sor with the smallest distance to the nearest location. We repeated this process until all mobile sensors were assigned to locations. If there were unvisited locations, the first mobile sensor reaching its destination picked its next location in the same greedy manner. This continued until all locations were visited.
In our simulation, the sensing field was 15 15 meters. We assumed that moving one meter takes one Joule and 10 seconds. A mobile sensor moves in a straight line from one location to another. Each experiment had 100 rounds, and in each round a certain number of events were generated at random locations. After each round, mobile sensors stayed at their final destinations and waited for the next schedules. We marked each simula-tion result with a 90 percent confidence interval.
Our comparison results under different numbers of event sites and mobile sensors showed that the greedy algorithm performs better than the K-means algorithm because of its time-critical nature. The K-means algorithm gives mobile sensors unbalanced job assignments, caus-Figure 6. User interface at the external server.The interface consists of six areas through which the user can monitor the system’s status and control the mobile sensors.
Action-control area
Monitor area Log area Sensor-status area
System-command area Patrolling path Direction of the mobile sensor Configure area Event WSN Link
ing some event sites to wait longer even when some mobile sensors are idle. On the other hand, the K-means algorithm is more energy-efficient because of its cluster-ing approach, which exploits event locality.
T
he iMouse system integrates WSN technologies into surveillance technologies to support intelligent mobile surveillance services. We can enhance or extend iMouse in several ways. First, we can improve mobile sensor navigation by, for example, integrating localization schemes to guide mobile sensors instead of using color tapes. Second, we can exploit coordination among mobile sensors, especially when they’re on the road. Finally, we need to further investigate how we can use mobile sensors to improve the network topology. ■Acknowledgments
Yu-Chee Tseng’s research is cosponsored by Taiwan’s Ministry of Education ATU Program; National Science Council of Taiwan grants no. 93-2752-E-007-001-PAE, 96-2623-7-009-002-ET, 2221-E-009-058-MY3, 95-2221-E-009-060-MY3, 95-2219-E-009-007, 95-2218-E-009-209, and 94-2219-E-007-009; Taiwan’s Ministry of Economic Affairs under grant no. 94-EC-17-A-04-S1-044; the Industrial Technology Research Institute of Taiwan; Microsoft; and Intel.
References
1. T. He et al., “Range-Free Localization Schemes for Large-Scale Sensor Networks,” Proc. 9th ACM Int’l Conf. Mobile Com-puting and Networking (MobiCom 03), ACM Press, 2003, pp. 81-95.
2. J.R. Evans and E. Minieka, Optimization Algorithms for Net-works and Graphs, 2nd ed., Marcel Dekker, 1992.
3. D. Johnson et al., “Mobile Emulab: A Robotic Wireless and Sensor Network Testbed,” Proc. 25th Ann. IEEE Conf. Com-puter Comm. (Infocom 06), IEEE Press, 2006.
4. J.P. Sheu, P.W. Cheng, and K.Y. Hsieh, “Design and Imple-mentation of a Smart Mobile Robot,” Proc. IEEE Int’l Conf. Wireless and Mobile Computing, Networking and Comm. (WiMob 05), IEEE Press, 2005, pp. 422-429.
Yu-Chee Tseng is chair of the Department of Computer
Science at National Chiao Tung University. His research interests include mobile computing, wireless communica-tion, network security, and parallel and distributed com-puting. Tseng received a PhD in computer and information science from the Ohio State University. He is a member of the ACM and a senior member of the IEEE. Contact him at [email protected].
You-Chiun Wang is a postdoctoral research associate in the
Department of Computer Science at National Chiao Tung University. His research interests include wireless commu-nication and mobile computing, QoS management and wireless fair scheduling, mobile ad hoc networks, and wire-less sensor networks. Wang received a PhD in computer sci-ence from National Chiao Tung University. Contact him at [email protected].
Kai-Yang Cheng is a master’s student in the Department of
Computer Science at National Chiao Tung University. His research interests include wireless communication and mobile computing, mobile ad hoc networks, and wireless sensor networks. Cheng received an MS in computer sci-ence and information engineering from National Chiao Tung University. Contact him at [email protected].
Yao-Yu Hsieh is a graduate student in the Department of
Computer Science at National Chiao Tung University. His research interests include wireless communication and mobile computing, mobile ad hoc networks, and wireless sensor networks. Hsieh received a BS in computer science and information engineering from National Chiao Tung University. Contact him at [email protected].
www.computer.org/join/
Complete the online application and get • immediate online access to Computer • a free e-mail alias — [email protected]
• free access to 100 online books on technology topics
• free access to more than 100 distance learning course titles • access to the IEEE Computer Society Digital Library for only $118
Join the IEEE Computer Society online at
Read about all the benefits of joining the Society at
