1. 1 Sensor networks for traffic monitoring Pravin Varaiya et al

2. 2 Outline  Challenge  Sensor networks for traffic applications  Pedamacs MAC protocol  Signal processing

3. 3 Challenge  Accuracy and low delay  Biggest cost is deployment and maintenance  lifetime (power consumption) will determine economic feasibility

4. 4 Sensor networks for traffic  Nodes generate data, report to access point  At intersection, vehicle detection must be reported in 0.1 s; also 30- sec periodic data  Nodes are power- and energy-limited; access points are not Sensor node Access point Freeway < 100 m Intersection

5. 5 Current traffic monitoring technology  Loop detectors is the standard; loops last 10 years  Closing lane to cut loops in freeway pavement is very disruptive  Alternatives today are microwave radar, video cameras  Installed cost is $600-$1000 per detector (lane) per year  Can sensor networks compete?

6. 6 Sensor networks with two special characteristics 1.One distinguished node, Access Point or AP; sensor nodes or SN periodically (eg. 30 s) generate data for transmission to access point 2.SNs are power- and energy-limited but AP is not: Consequently Transmission AP  SN is one-hop Transmission SN  AP is multi-hop  Two conditions satisfied in traffic applications Freeway < 100 m Intersection

7. 7 Pedamacs vs random access networks Pedamacs networks:  Access point discovers network topology; nodes discover next hop  Access point computes and broadcasts transmission schedule to all nodes (TDMA data)  During data phase, node sleeps if it is not scheduled to listen or to transmit Random access networks:  Access point and nodes discover next hop  Nodes randomly transmit and constantly listen for incoming packets  Refinements proposed to reduce node ‘listening’ time

8. 8 Comparison of random access and Pedamacs networks  Comparison via TOSSIM, a TinyOS simulator  Need to select critical parameters for comparison –Backoff-listening random access scheme  Back-off window, listening window –Transmission range  Nodes randomly distributed inside unit circle

9. 9 Power consumption in PEDAMACS vs random access  50 kbps; one packet every 30 sec; vertical scale is log 10  Listening in random access uses 1000X more power, and receiving uses 10X power than in Pedamacs

10. 10 Lifetime of PEDAMACS vs random access network  Two AA batteries: 2200 mA at 3 V  Pedamacs network lasts 600 days, need 5X improvement  Random access network lasts 10 days  unsuited for traffic control

11. 11 Pedamacs vs random access delay  Random access delay excessive for traffic application

12. 12 Detecting vehicles  Experiments  Spatial and temporal resolution  Speed  Vehicle classification

13. 13 Data Set 1: motes in middle of lane 1 [Mic] 2 [Mic] 3 [MagXY] 4 [MagXY]

14. 14 Ford15_x0_1.dat Wind disturbance

15. 15 Ford27_x0_1_track_at_end.dat Noise from truck

16. 16 Ford15_x0_1.dat ford27_x0_1_track_at_end.dat Magnetic signature for classification

17. 17 Ford25_x0_2.dat

18. 18 Ford_acc_x0_1.dat Vehicle accelerating going over the mote

19. 19 ford_stopB4mote1_1sec_acc_x0_1_otherCars.dat From another car Car stopped before mote 1,3

20. 20 Summary  Sensor networks offer a promising alternative  Acoustic signal is corrupted by noise--more filtering and processing needed for robust detection  Magnetic signal depends on orientation  Work needed to implement TDMA protocol  Signal processing for speed, vehicle classification  Deployment, reliability