RoadSphygmo: Using barometer for traffic congestion detection

Road traffic congestion is a worldwide problem that continues to increase, resulting in adverse environmental and health consequences in addition to wasted fuel and individual productivity loss. Detecting congestion and communicating it in a timely manner may enable a number of mitigation strategies such as re-directing traffic as well as individuals adopting alternate routes or transportation methods. We seek to crowd-source road congestion information detected on individual user's smart phones as a low-cost approach for such detection. To encourage users to participate in such crowd-sourcing, it is critical that the information is collected conveniently, with as little user action required as possible, and the use of low-cost sensors that would be ubiquitously available on smart phones. Moreover, unlike the use of GPS, the use of these sensors should consume very little energy, thus resulting in minimal drain on the phone's battery. We propose using the barometer sensor present in mobile phones to detect traffic congestion. Roads which seem perfectly flat to naked eye actually vary in altitude at different points along the way. The barometer sensor is sensitive enough to measure these altitude changes. On a congested road, a vehicle covers a much shorter distance over a time period. A vehicle experiences a significantly larger number of altitude changes on a free flowing road compared to a congested road. This forms the basis for our traffic congestion detection. We extract features based on the altitude change and use a Support Vector Machine (SVM) as a classifier to initially classify into two broad categories of vehicular state: still and in motion. The sequence of vehicle states is then used to determine the traffic condition. Traffic condition is categorized into 3 states: `stuck', `congestion' and `moving', based on chosen thresholds for the number of still/motion states. To validate the state determination by our RoadSphygmo1 algorithm, we compared it with the GPS speeds during the same time period. While the thresholds we chose are not exact measures of traffic state, they nevertheless provide useful information about the congestion on the road. Field experiments conducted on the roads in Chandigarh and Mumbai in India show promising results.