You are on page 1of 35

WIRELESS ACROSS ROAD : RF BASED

ROAD TRAFFIC CONGESTION DETECTION

Swaroop Roy, Rijurekha Sen, Swanand Kulkarni,


Purushottam Kulkarni, Bhaskaran Raman, Lokendra Kumar Singh

Department of Computer Science and Engineering


Indian Institute of Technology, Bombay

http://www.cse.iitb.ac.in/synerg

1
Traffic is everywhere !

2
Traffic is everywhere !

Non-Laned Laned

3
Types of Traffic

Heterogeneous Homogeneous
4
Losses due to road traffic congestion

• Fuel wastage
• New Delhi loses Rs 11.5 crores per day ! [4]
• 30 lakh litres of fuel wastage per day ! [4]

• Longer and unpredictable travel times


• 56 hours a year in Los Angeles, nearly half of the typical
annual vacation time of three weeks [3]

• Reduces the overall efficiency of the road transport system

5
Traffic is a real pain !

6
Problem Definition

To develop an accurate traffic congestion detection system


• Low cost
• Low installation overhead
• Can handle chaotic traffic
• Requires minimal participation from commuters and
vehicles
• Automated traffic classification

7
Comparison of existing techniques

Dual loop Image sensors GPS Acoustic Wireless


detectors based sensors Across Road
Works for
non-laned NO YES YES YES YES
traffic
Cost HIGH HIGH LOW LOW LOW

Disrupts
traffic during YES NO NO NO NO
installation
Vehicle / User
participation NO NO YES PARTIAL NO
required

Propose a low cost solution that can handle chaotic traffic


8
Wireless Across Road

Hypothesis : Wireless Link Characteristics = f (traffic)

9
Wireless Across Road

Hypothesis : Wireless Link Characteristics = f (traffic)

1. Why wireless links ?


2. Empirical testing of hypothesis
3. Traffic classification 10
Why wireless links ?

11
Sensitivity of 802.11b links to LOS [1]

High RSSI

Low RSSI

Decrease in RSSI by 25dbm !

802.11b link quality suffers without line of sight (LOS)

12
Sensitivity of 802.15.4 links to LOS[2]

20% errored packets


at -90dbm
Almost no errored
packets at -90dbm

Airstrip (LOS) Dense foliage (NLOS)

Wireless Link quality suffers without LOS

13
Empirical testing of hypothesis

14
Traffic States

Free Flowing Traffic

• Vehicles move at desired speed


• Results mostly in LOS

15
Traffic States
Free Flowing Traffic

• Vehicles move at desired speed


• Results mostly in LOS

Congested Traffic

• Vehicles forced to move at


crawling speed
• Results mostly in NLOS

16
Measurement Setup

17
Measurement Setup

Ground Truth
Free flowing : < 6:30 pm
Congested : > 6:30 pm
18
Measurement Setup

Ground Truth Measured Metrics


Free flowing : < 6:30 pm Received Signal Strength Indicator
Congested : > 6:30 pm Link Quality Indicator
Packet Reception Rate 19
Is RSSI sensitive to traffic conditions ?

Congested
traffic

Traffic in
transition

Free
flowing traffic

20
Is RSSI sensitive to traffic conditions ?

Congested
18dbm
traffic

Traffic in
transition

Free
flowing traffic

RSSI is sensitive to traffic changes 21


Is Packet Reception sensitive to traffic conditions ?

Congested
traffic almost 25

Traffic in
transition
phase

Free flowing
traffic

Packet Reception is sensitive to traffic changes 22


What about narrow (2 lane) roads ?

10m

Hypothesis : Wireless Link Characteristics = f (traffic)


Does it still hold ?

Conducted similar measurements on the 10m road as on the 25m road


23
Is RSSI sensitive on narrow roads ?

Congested
Free flow

NO separation between LOS and NLOS 24


Alternate Setup :
Increased sender-receiver
distance

25
Alternate Setup :
Increased sender-receiver
distance

High separation
d’’

Low separation

26
Technique will work for narrow roads too
Traffic classification

27
Training Set for Classification
Attributes
lqi_90thpercentile (1)

lqi_80thpercentile (2)
lqi_70thpercentile (3)

lqi_60thpercentile (4)

lqi_50thpercentile (5)
lqi_40thpercentile (6)
lqi_30thpercentile (7)

lqi_20thpercentile (8)

● 8 values of RSSI
● 8 values of packet reception
Total 24 attributes per
instance

● 1 instance in training set from each 5 min log WEKA open source
● 1 class label per instance -> 0 (free-flow), 1 (congestion) package
● 12/13/10
Class label assigned based on manual observation for machine learning tools
Classification Accuracy
● 9 hours 20 mins data for narrow road -> 112 logs
of 5 mins each
lqi_20thpercentile <= 94 ?
● 112 instances with attributes and class label

yes no
● Randomly order instances, input to WEKA first
102 as training data and last 10 as test data, a
decision tree model output by WEKA Congested Free-flow

● Repeat procedure 50 times

Results

Simple threshold based classifier


12/13/10 based on rule (lqi_20thpercentile <= 94 : 1) will probably
suffice
Concluding Remarks

Goal : To develop a low-cost congestion detection


system for developing regions

Exploited “wireless link behavior suffers in NLOS” to devise a


setup to detect road traffic congestion

Developed a decision tree which has an accuracy of 97%, from


the data collected from deployments

30
Future Work

• Conduct similar measurements using other wireless technologies


to test their feasibility

• Aim to build customized hardware after finding the technology that


works best

• Queue length measurement using arrays of sender-receiver pairs is


another direction of interest

Logs, scripts, videos can be found at


http://www.cse.iitb.ac.in/~puru/wirelessacrossroad
31
THANK YOU

Questions ??

32
References
[1] Arunesh Mishra, Suman Banerjee, Vivek Shrivastava,
Dheeraj Agrawal
Understanding the limitations of transmit power control for
indoor wlans
IMC, October 2007

[2] Naveen Madabhushi Dattatraya Y Gokhale Phani K Valiveti


Bhaskaran Raman, Kameswari Chebrolu and Dheeraj Jain
Implications of link range and (in)stability on sensor network
architecture
WiNTECH, September 2006

[3] Eco-Economy: Building an Economy for the Earth

[4] http://timesofindia.indiatimes.com/city/delhi/Delhi-wastes-Rs-115cr-in-traffic-jams-daily/
12/13/10 33
articleshow/5125771.cms
Cumulative probability of the LQI measured every 5 mins
for 90 minutes on 25m wide road

Corresponds
to congested
traffic

Corresponds
to traffic in
transition
phase

Corresponds
to free
flowing
traffic 34
Cumulative probability of the LQI with the diagonal distance
between the sender and receiver 30m

Corresponds
to congested
traffic
Corresponds
to traffic in
transition
phase
Corresponds
to free
flowing
35
traffic

You might also like