I am pursuing PhD under Prof. Bhaskaran Raman. We seek to develop an Intelligent Transport System (ITS) for developing regions. The horror of road traffic conditions in developing regions can be understood only through experience, but some representative videos can be seen at MumbaiTraffic

The key to obliterate the traffic woes is adding to the road infrastructure, building flyovers at key places, widening roads etc in the same rate as that of traffic growth. But infrastructure growth is slow in developing regions due to lack of funds, space and bureaucratic issues. ITS are parallel efforts, that seek to alleviate the problems, using technology. A typical ITS deployment has the following components -

  • Sensing - sensors are placed on the road (dual loop detectors, magnetic sensors, image sensors) or in a set of moving vehicles (GPS, cell phones) to sense various kinds of road data
  • Communication - the data is communicated to some central server using GPRS or some other connectivity
  • Processing - the data is processed at the central server. Processing might involve classifying the data into various traffic conditions (free flowing/ congested) based on some machine learning algorithms, making travel time predictions etc.
  • Application - the processed data is communicated back in the form of some useful information to commuters on the road. What is the congestion scenario at important junctions? Which is the most efficient route between a given source and destination? What is expected travel time between two places? With the boom in cell phone usage, user applications, that can download such information from the central server, are easy to design and deploy.

Overview of work done

  • Literature Survey - We have identified three requirements for a traffic sensing technique to work in roads of developing regions. (1) They have to handle chaotic and non-lane based traffic (2) They should not be underground as that would need reinstallation every time a road is relaid, as pavement lifetime in developing regions is low. (3) Installation and maintenance costs and power requirements should be low.

    Magnetic sensors and loop detectors definitely don't meet the second and third requirements.Whether they meet the first requirement is open to question as they are generally installed in every lane and detect vehicles per lane. Image sensors meet the second requirement but jointly meeting the first and third requirements seem difficult for them as real-time image processing for chaotic traffic might need costly cameras, more computaion and power. Sensors in vehicles (probes) meet all the three requirements, but the questions of penetration of GPS and smartphones in developing regions, incentivising participatory sensing given energy drain and costs involved in sensing and communication are open. We wrote a workshop paper on these observations in NSDR, 2009.

  • Sensing - We have developed two new traffic sensing techniques that meet all three requirements and doesn't have the penetration and incentive issues that probe sensors have.

    Our first technique is based on sound. Chaotic traffic of developing countries is very noisy. The core of our technique comprises a pair of road side acoustic sensors, separated by a distance. If a moving vehicle honks between the two sensors, its speed can be estimated from the Doppler shift of the honk frequency. In this context, we have developed algorithms for honk detection, honk matching across sensors, and speed estimation. Using over 18 hours of road-side recordings, we show that our speed estimation technique is effective in real conditions. Further, we use our data to characterize traffic state as free-flowing versus congested using a variety of metrics: the vehicle speed distribution, the number and duration of honks. Our results show clear statistical divergence of congested versus free flowing traffic states, and a threshold-based classification accuracy of 70-100% in most situations. We wrote a paper on this in Mobisys, 2010.

    Our second technique is based on exploiting the variation in wireless link characteristics when line of sight conditions between a wireless sender and receiver vary. Our system comprises of a wireless sender-receiver pair across a road. The sender continuously sends packets. The receiver measures metrics like signal strength, link quality and packet reception. These metrics show a marked change in values depending on whether the road in between has free-flowing or congested traffic. We have experimented with off-the-shelf IEEE 802.15.4 compliant CROSSBOW Telosb motes. From about 15 hours of experimental data on two different roads in Mumbai, we show that we can classify traffic states as free-flowing and congested using a decision tree based classifier with 97% accuracy. We wrote a workshop paper on this in WISARD, 2011.

  • Communication & Processing - For our sound-based technique, we have also examined whether computation intensive acoustic signal processing, be implemented on an embedded sensor platform, to be used for on-road sensing? Can the sensing and processing be done in near real time? Will the cost be low enough? These are some implementability issues. Will the system be able to detect congestion on a wide variety of roads? Will the traffic classification model vary from road to road? In that case, what will be the training overhead of our system on a new road? Can we do without training using unsupervized learning? These are some usability issues.

    We have developed an acoustic sensing hardware prototype which has been deployed by the side of the road. This unit samples and processes road noise to compute various metrics like amount of vehicular honks and vehicle speed distribution and sends the metrics to a remote server every alternate minute. Data from deployment of this prototype in six different Mumbai roads, validated against manually observed ground truth, shows feasibility of per minute congestion monitoring from the remote server. K-means clustering gives on average 90% accuracy to group unlabeled data on a new road into two clusters of congested and free-flow. Deployment data from one road for six days shows the temporal variation in traffic state for that road. Our prototype has a moderate cost of $160 and is easy to install and maintain on road-side lamp-posts. We wrote a paper on this in SECON, 2011.

Publications

    PhD Project:

  • Rijurekha Sen, Vishal Sevani, Prashima Sharma, Zahir Koradia, Bhaskaran Raman, “Challenges In Communication Assisted Road Transportation Systems for Developing Regions”, 3rd ACM Workshop on Networked Systems for Developing Regions (NSDR'09), a workshop in SOSP'09, Big Sky, Montana, USA, 11 Oct 2009. paper, ppt
  • Rijurekha Sen, Bhaskaran Raman, Prashima Sharma, "Horn-Ok-Please", The 8th Annual International Conference on Mobile Systems, Applications and Services, Mobisys'10, San Francisco, USA, Jun 15-18, 2010. paper, ppt
  • Swaroop Roy, Rijurekha Sen, Swanand Kulkarni, Purushottam Kulkarni, Bhaskaran Raman, Lokendra Singh, "WirelessAcrossRoad: RF based Road Traffic Congestion Detection", The 5th Annual Worshop on Wireless Systems: Advanced Research and Development (WISARD'11), Bangalore, India, Jan 4-5, 2011. paper, ppt
  • Rijurekha Sen, Pankaj Siriah, Bhaskaran Raman, "RoadSoundSense: Acoustic Sensing based Road Congestion Monitoring in Developing Regions", 8th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks, SECON'11, Salt Lake City, Utah, USA, Jun 27-30, 2011. paper

    BTech Project:

  • Pampa Sadhukhan, Dr. Pradip K. Das, Rijurekha Sen, Niladrish Chatterjee, Arijit Das, "A Middleware-Based Approach to Mobile Web Services", The 5th Asian International Mobile Computing Conference, AMOC-2007. paper
  • Pampa Sadhukhan, Rijurekha Sen, Pradip K. Das, "A Middleware Based Approach to Dynamically Deploy Location Based Services Onto Heterogeneous Mobile Devices Using Bluetooth in Indoor Environment", The 2nd International Conference on Advanced Communication and Networking. ACN'10, Miyazaki, Japan, June 23 - 25, 2010. paper

Posters

  • Rijurekha Sen, Prashima Sharma, Bhaskaran Raman, “Horn-Ok-Please”, Microsoft Techvista PhD Poster Session, 2010, Banglaore, 22 Jan. We won the 4th prize, worth Rs. 25,000. pdf,ppt
  • Rijurekha Sen, Pankaj Siriah, Bhaskaran Raman, Swaroop Roy, Swanand Kulkarni, Puru Kulkarni, “Fighting Chaotic Road Congestion”, Microsoft Techvista PhD Poster Session, 2011, Pune, 21 Jan. pdf

Awards

  • Microsoft Research India gives five PhD fellowships every year to graduate students in India based on resume and three recommendations. There were 50 applications in 2010 and mine was one of the selected five. In course of my PhD, I am eligible to get a laptop, a monthly stipend and Rs. 2.5 lacs for foreign trips. This will also ensure increased interaction with the researchers at MSR India and taking part in the various events organized by them. More details of the award are at msr_phd_fellowship. You can see a photo of mine, taking the award from Mr. Prithviraj Chauhan, Chief Ministerof Maharashtra.

Reports

  • I gave my first Annual Progress Seminar (APS) on 27 Jan, 2010. report, ppt.
  • I gave my second Annual Progress Seminar (APS) on 27 Jan, 2011. report, ppt.