Lien-Wu Chen†, Jen-Hsiang Cheng‡, Yu-Chee Tseng‡, Lun-Chia Kuo∗, Jen-Chieh Chiang∗, and Wan-Jung Lin∗
†Department of Information Engineering and Computer Science, Feng Chia University, Taichung, 407, Taiwan, R.O.C.
‡Department of Computer Science, National Chiao Tung University, Hsinchu, 300, Taiwan, R.O.C.
∗Information and Communications Research Laboratories, Industrial Technology Research Institute, Hsinchu, 310, Taiwan, R.O.C.
Email: [email protected];{jhcheng, yctseng}@cs.nctu.edu.tw; {jeremykuo, jaychiang, wanjunglin}@itri.org.tw Abstract—LEGS is a Load-balancing Emergency Guiding
Sys-tem using a wireless sensor network. In LEGS, we design a load-balancing guiding scheme and derive an analytical model in order to reduce the total evacuation time of people. The proposed guiding scheme can provide the fastest path to an exit based on the evacuation time estimated by the derived analytical model.
To the best of our knowledge, LEGS is the first system which takes the corridor capacity and length, exit capacity, and people distribution into consideration for analyzing evacuation time and planning escape paths. Through LEGS, the congestion of certain corridors and exits can be released to significantly reduce the evacuation time of people. Analytical and simulation results show that LEGS outperforms existing works, which can prevent people from following the local optimal guiding direction with the longer evacuation time in total. LEGS thus demonstrates an efficient emergency guiding system for public safety.
Keywords: Home Security, Navigation, Pervasive Com-puting, Wireless Communication, Wireless Sensor Net-work.
I. INTRODUCTION
The recent progress of wireless communications and embed-ded microelectromechanical systems (MEMS) technologies has made wireless sensor networks (WSNs) more attractive.
Existing works have been made for vehicle security and tracking [1], emergency guiding and monitoring [2], and cooperative collision avoidance [3].
For emergency guiding purposes, reference [4] deploys a large number of active RFID tags in a building. People use personal digital assistants (PDAs) connected by RFID readers via Compact Flash interfaces for indoor localization and emergency guiding. However, reference [4] guides people to the nearest exit without taking people distribution into account.
Thus, there may be serious congestion in the nearest exit due to uneven people distribution and unbalanced emergency guiding.
In addition to a large number of active RFID tags, reference [5] deploys a few Bluetooth devices in a building for indoor positioning. Similarly, people distribution is not considered in [5] so that the total evacuation time may become longer due to the local optimal guiding, which is also occurred in [6].
In this work, we design a Load-balancing Emergency Guiding System (LEGS) in a 2D indoor environment using a wireless sensor network that aims to guide people to exits as
Y.-C. Tseng’s research is co-sponsored by MoE ATU Plan, by NSC grants, 97-3114-E-009-001, 97-2221-E-009-142-MY3, 98-2219-E-009-019, 99-2218-E-009-005, and 100-2219-E-009-001, by ITRI, Taiwan, by III, Taiwan, by D-Link, and by Intel. This research was supported by Information and Communi-cations Research Laboratories (ICL), Industrial Technology Research Institute
Fig. 1. System architecture of LEGS.
soon as possible when emergencies happen. To the best of our knowledge, LEGS is the first system which takes the corridor capacity and length, exit capacity, and people distribution into consideration for analyzing evacuation time and planning escape paths. In LEGS, a load-balancing guiding scheme is designed to find the fastest path to an exit for people based on the evacuation time estimated by the derived analytical model. The congestion of certain corridors and exits can be released to significantly reduce the evacuation time of people.
In particular, LEGS can prevent people from following the local optimal guiding direction with the longer evacuation time in total.
II. SYSTEMDESIGN
Fig. 1 shows the system architecture of LEGS. Sensor nodes (i.e., black and white circles) are deployed in a 2D indoor environment, which form a multi-hop ad hoc network. One node serves as the sink of the network, and it is connected to the control host, which issue commands and config the network. To support emergency guiding services, sensors are classified as normal sensors (i.e., black circles), exit sensors (i.e., white circles), and boundary sensors (i.e., nodes with one or more neighboring sensors belonging to different emergency guiding trees). Next, we will show how to find the fastest escape path leading to an exit as detecting an emergency event, whereas the detail for how to construct an initial guiding tree rooted by an exit and to prevent people from crossing hazardous region can be found in our previous work [2].
A. Load-Balancing Guiding
In our system, each sensor knows the capacity and length of each corridor, the capacity of each exit, and its own location in the 2D plane. In addition, the number of people around a sensor will be detected by RFID or image recognition technologies and periodically reported to the sink by each
sensors, it will broadcast the people distribution information to all sensors. As detecting an emergency event, all boundary sensors will execute the following steps:
Step 1: For each initial guiding tree TI without hazardous region RH, boundary sensors will calculate the evacuation time Texit of TI by the analytical model proposed in Section II-B.
For each TI with RH, the sensors and corridors inside RHwill be removed first. Then, the sensors outside RH losing their parent sensors inside RH will be guided to an exit using the shortest path so that a new guiding tree TN is formed. For calculating Texit of TN, there are two possibilities as follows:
(a) If there is an exit inside RH, all people inside RH will be guided to the exit inside RH using the shortest path. Texit
of TN can be estimated by the analytical model proposed in Section II-B.
(b) If there is no exit inside RH, the sensor detecting the emergency event and its corridors will be removed and all other sensors inside RHwill be guided to the sensors outside RH using the shortest path so that a new guiding tree TS is formed. Texit of TS can be estimated by the analytical model proposed in Section II-B.
Step 2: The total evacuation time Ttotal is decided by the emergency guiding tree TMAX with the longest Texit. After Texit of all emergency guiding trees are estimated by boundary sensors, boundary sensor i in TMAX will reset its guiding direction to the neighboring guiding tree TMIN with the shortest Texit and recalculate Texit of TMAX and TMIN. If Ttotal decreases, i will notify its neighboring sensors b(i) in TMAX that b(i) become boundary sensors.
Assume that TMAX has m boundary sensors between TMAX
and TMIN, where boundary sensor nj is sorted in decreasing order by the number of people in nj, for j = 1, 2, . . . , m.
First, n1 calculates Texit of TMAX and TMIN after removing and adding n1, respectively. If Ttotal decreases, n1 will reset its guiding direction to TMINand notify its neighboring sensors b(n1) in TMAX that b(n1) become boundary sensors. Second, n2 calculates Texit of TMAX and TMIN after removing and adding n1 and n2, respectively. If Ttotal decreases, n2 will reset its guiding direction to TMIN and notify its neighboring sensors b(n2) in TMAX that b(n2) become boundary sensors.
Similarly, nm calculates Texit of TMAX and TMIN after re-moving and adding n1, n2, . . . , and nm, respectively. If Ttotal
decreases, nm will reset its guiding direction to TMIN and notify its neighboring sensors b(nm) in TMAX that b(nm) become boundary sensors. Finally, new boundary sensors b(n1), b(n2), . . . , and b(nm) will repeat Step 2 to determine whether they should reset their guiding directions to TMIN.
Step 3: Assume that there are N emergency guiding trees in the 2D plane. As N = 2, the total load-balancing guiding can be finished by Step 2. As N = 3, the load-balancing guiding of Step 2 will be first done between the emergency guiding tree T1 with the longest Texit and its neighboring guiding tree T2with the shortest Texit. Then, the total load-balancing guiding can be finished by Step 2 between T1+ T2 and the third emergency guiding tree T3. As N = 4, the load-balancing
guiding tree T1 with the longest Texit and its neighboring guiding tree T2 with the shortest Texit. At the same time, the load-balancing guiding of Step 2 will be done between the remaining guiding trees T3 and T4. Then, the total load-balancing guiding can be finished by Step 2 between T1+ T2 and T3+T4. Similarly, as N ≥ 5, Step 2 and 3 can be repeated to finish the total load-balancing guiding.
B. Evacuation Time Analysis
Given an emergency guiding tree TGrooted by an exit sen-sor, we derive its total evacuation time considering the corridor capacity and length, exit capacity, and people distribution.
Assume that there are n sensors in TGand sensor 1 is the root.
sensor 2, sensor 3, . . . , and sensor n are sorted in increasing order by their hop counts to sensor 1. Below, we first introduce some notations for sensor i, its parent sensor p(i), and its child sensors c(i) in TG, for i = 1, 2, . . . , n:
• Ti: the time to evacuate from i for the last person.
• Di: the time to move from i to p(i).
• Ni: the number of people in i as emergences happen.
• Ci: the corridor capacity from i to p(i).
Assume that there are m child sensors in c(i). We calculate the evacuation time Ti of the subtree TG(i)rooted by i for the last person. According to whether there is congestion occurring in i as the last person in TG(i)evacuates from i, the estimation of Ti can be classified as follows:
Case 1: There is no congestion occurring in i as the last person in TG(i) evacuates from i. The evacuation time Ti is the sum of the time to evacuate from c(i) for the last person and the time to move from i to p(i). So
Ti= maxj∈c(i)(Tc(i)j ) + Di.
Case 2: There is certain congestion occurring in i as the last person in TG(i)evacuates from i. First, all sensors j in c(i) are sorted in increasing order by Dc(i)j , for j = 1, 2, . . . , m, such that Dc(i)1 ≤ Dc(i)2 ≤ · · · ≤ Dmc(i). Second, we find the smallest k such that Cc(i)1 + Cc(i)2 +· · · + Cc(i)k > Ci. If there is no such k existed, it represents that the corridor capacity between i and p(i) is large enough to be passed concurrently by the people from c(i). In other words, there will be no congestion occurring in i and Ti can be estimated by Case 1. Otherwise, there are two possibilities as follows:
(a) If Dc(i)k < NCi
i, it implies that there is a high probability of congestion occurring in i as the people evacuates from c(i) to i. Ti is modeled by summing the time to evacuate from i for all people in TG(i) and the time to move from i to p(i) as
Fig. 2. Comparisons of evacuation time under different numbers of people.
(b) If Dc(i)k ≥ NCii, it implies that some people in sensor 1, 2, . . . , and k− 1 have evacuated from i as the people in k arrive at i. Ti is modeled by summing the time to move from k to i, the time to evacuate from i for the remaining people in TG(i), and the time to move from i to p(i) as follow
In the analytical model, we adopt the maximum time estimated by Case 1 and Case 2 as Ti since Ti is the time to evacuate from i for the last person in TG(i), for i = 1, 2, . . . , n.
Thus, the total evacuation time of TGis equal to T1that can be obtained by calculating Tn, Tn−1, . . . , and T1 in order, where D1= 0 and C1 is the exit capacity.
Fig. 2 shows comparisons of total evacuation time under 50, 60, . . . , 1590, and 1600 people. We deploy 6× 6 sensor nodes in a 2D grid plane. There are two exit sensors located on the bottom left and bottom right corners, and the remaining sensors are normal sensors. The corridor and exit capacities are randomly selected from 2 to 6 people/second, and the moving time for each corridor is randomly chosen from 10 to 15 seconds. We first randomly assign 50% people to all sensors and then randomly select 9 hot-spot sensors for assigning the rest 50% people to them. We compare our scheme against the Smallest Altitude First (SAF) method [2] that guides people to the neighboring sensor with the smallest hop count to an exit and the Fastest Flow Speed First (FFSF) method [6]
that guides people to the neighboring sensor with the fastest moving speed.
From Fig. 2, we can observe that both SAF and FFSF suffer from the local optimal selection problem, which SAF may select the escape path with the shortest distance to a exit but longer evacuation time, and FFSF may select the escape path with the faster moving speed currently but slower later.
In particular, while SAF has the longer evacuation time than FFSF under more than 800 people, FFSF has the longer one
Fig. 3. Hardware components and the prototype of LEGS.
comparisons of simulation and analytical results. Each simu-lation is repeated 1000 times and we take the average value.
As can be seen, the simulated and analytical results are quite close, which justifies the correctness of our derivation.
III. PROTOTYPEIMPLEMENTATION
In our prototype, the sensor is equipped with a TFT LCD panel controlled by ATmega128 [7], a light sensor TSL2560 [8], and UI buttons as input devices. The front side and back side of the sensor are shown in Fig. 3(a). The SAF method [2], the FFSF method [6], and our approach are implemented in Jennic JN5139 [9], which has a 16MIPs 32-bit RISC processor, a 2.4GHz IEEE 802.15.4-compliant transceiver, 192kB of ROM, and 96kB of RAM. In particular, JN5139 allows the flexibility of supporting mesh networking and packet routing inside a building.
For the demonstration of indoor people evacuation, we use a projector to simulate that people escape from an office as emergences happen, as shown in Fig. 3(b). The evacuation simulator is developed by Processing [10] to create images, animations, and interactions. The light sensor can be high-lighted by a laser pen to trigger an emergency event and three emergency guiding methods can be selected for evacuating people to exits. The animation of people evacuation is shown in the projected screen and the evacuation time is counted until all people escape from the office. From the screen, we can compare the evacuation time and guiding directions of SAF, FFSF, and our approach under different people distribution.
REFERENCES
[1] L.-W. Chen, K.-Z. Syue, and Y.-C. Tseng. A Vehicular Surveillance and Sensing System for Car Security and Tracking Applications. In ACM/IEEE Int’l Conf. on Information Processing in Sensor Networks (IPSN), Apr. 2010.
[2] Y.-C. Tseng, M.-S. Pan, and Y.-Y. Tsai. Wireless Sensor Networks for Emergency Navigation. IEEE Computer, 39(7):55-62, July 2006.
[3] L.-W. Chen, Y.-H. Peng, and Y.-C. Tseng. An Infrastructure-less Framework for Pre-venting Rear-End Collisions by Vehicular Sensor Networks. IEEE Communications Letters, 15(3):358-360, Mar. 2011.
[4] L. Chittaro and D. Nadalutti. Presenting evacuation instructions on mobile devices by means of location-aware 3D virtual environments. In Proc. of the Int’l Conf.
on Human Computer Interaction with Mobile Devices and Services (MobileHCI), 2008.
[5] S. R. Gandhi, A. Ganz, and G. Mullett. FIREGUIDE: Firefighter guide and tracker.
In Proc. of IEEE Int’l Conf. of Engineering in Medicine and Biology Society (EMBC), 2010.
[6] P.-Y. Chen, Z.-F. Kao, W.-T. Chen, and C.-H. Lin. A Distributed Flow-Based Guiding Protocol in Wireless Sensor Networks. In Proc. of Int’l Conf. on Parallel Processing (ICPP), Sept. 2011.
[7] TFT LCD microcontroller, ATmega128. http://www.atmel.com.
[8] Light sensor, TSL2560. http://www.taosinc.com/.
[9] Jennic, JN5139. http://www.jennic.com/.
[10] Processing. http://www.processing.org/.