Understanding Self Honking Patterns of Individual Riders using CNN Model

Chronic stress and anxiety can be exacerbated by prolonged exposure to loud and non-homogeneous vehicle honk, particularly that caused by two-wheeler, especially in crowded urban areas in developing nations. Honking has recently become a significant problem due to the laxity of traffic rules and the extensively increasing of two-wheeler in emerging countries like India. Both efficient urban design and driver profiles that enable us to take appropriate action based on honking are necessary to limit noise pollution. A smartphone-based system has been developed to identify non-homogeneous honks produced by two-wheeler. The proposed system can also classify the non-homogeneous honks as discrete and continuous honks using deep learning techniques. Two CNN models are being used, one to detect self-honk identification and another for honk classification. For the first CNN model, we get 98% training accuracy and 96% validation accuracy of self-honk detection in a controlled environment and more than 85% training accuracy and 82% validation accuracy of self-honk detection in a real-time environment. On the other hand, we create another CNN model to classify the self-honks into discrete and continuous honks. Here we achieve 94% training accuracy and 61% validation accuracy.