Hough Transform as a Tool for the Classification of Vehicle Speed Changes in On-Road Audio Recordings

• Published in: New Frontiers in Mining Complex Patterns Vol. 11948; pp. 137 - 154
• Main Authors: Kubera, Elżbieta, Wieczorkowska, Alicja, Kuranc, Andrzej
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2020; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Audio signal analysis; Hough transform; Speed changes detection
• ISBN: 9783030488604; 3030488608
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-48861-1_9

Spectrogram is a very useful sound representation, showing frequency contents as a function of time. However, the spectrogram data are very complex, as they may contain both lines or curves corresponding to partials (harmonic or not), whose frequency changes in time, as well as noises of various origin. In this paper, we address the extraction of line parameters from spectrograms for audio data, recorded for cars passing by an audio recorder. These lines represent pitched sounds, and the frequency along these lines is usually related to the vehicle speed. Our goal is to detect whether the vehicle is slowing down, speeding, or maintaining approximately constant speed. However, the lines may be broken, they bent when the car is passing the microphone because of the Doppler effect, which is strongest when very close to the microphone, and they are on the noisy background. Our goal was to elaborate a methodology, which extracts a simple representation of parameters of these lines (possibly broken, curvy and in noise), and allows detecting the behavior of drivers when passing the measurements point, e.g. near the radar. Audio data can be very useful here, as they can be recorded at low visibility. The proposed methodology, together with the results for on-road recorded audio data, are presented in this paper. This methodology can be then applied in works on road safety issues.