Towards Acoustic Event Detection for Surveillance in Cars

Abstract:
The protection of cars from being stolen or damaged is a major issue for automotive industry. Therefore several techniques exist to alert the owner of the car or stop the thief from driving away. In this contribution we describe an acoustic event detection (AED) framework for surveillance in cars, providing information about what is going on in or around the car. The proposed system operates on a continuous audio stream that may be acquired by the existing incar hands-free telephony microphone. The evaluation on a database containing 51 different event classes in combination with in-car street noise provides an accuracy of more than 70% in continuous audio over a wide range of signal to noise ratios (SNRs) in a close-to-realistic parking scenario. The accuracy obtained on segmented single events reaches even more than 90% over all examined SNRs.