TrigSense: Accelerometer triggered audio sensing for traffic condition monitoring

Rate of honking combined with the average vehicular movement can coarsely estimate the traffic condition of a street in an urban area of developing nations. Recently a lot of work is done on participatory sensing using users´ mobile phones. This leads to a concern of battery usage in the mobile phones. In this paper, we propose a multimodal approach for triggered sensing, where the low power sensors are used to trigger power hungry sensors for traffic condition estimation. Instead of turning on all the sensors, the low power consuming sensors like accelerometer is used to trigger the high power consuming sensors like microphone by applying a predefined logic. Feature extraction of the audio is done on the mobile device. The feature data and a simultaneously generated metadata xml file containing the location related information and time-stamp are posted to a backend server for further processing. A decision tree based approach is proposed for deriving the traffic condition using horn detection. Based on those decisions, application for traffic condition monitoring are created and updated for public access. Experimental results show the effectiveness of triggered audio sensing in terms of energy savings (upto 50%) on the mobile device and also in the reduction of the overall processing at the server.