Participatory sensing based traffic condition monitoring using horn detection

Rate of vehicular horn is an important parameter to estimate the traffic congestion of a street in an urban area in developing nations. In this paper we propose a participatory sensing based approach for traffic condition monitoring using horn detection employing inbuilt sensors of smart phones. Feature extraction is performed on the audio captured in users' phones. The features are then sent to a backend server for horn classification and decision making. For feature extraction we use a Modified Mel Frequency Cepstral Coefficient method, which modifies the conventional mel filter bank structure according to the spectral properties of car horn. Experimental results indicate that the feature can handle variations due to various positions of the mobile phone. Transmitting of the features instead of the whole audio data to server and performing the classification in the server preserves users' privacy as well as ensures less power consumption in mobile phones.