Haralick feature extraction from timefrequency images for automatic detection and classification of audio anomalies for road surveillance

Asim, Muhammad · 2018 · Crossref

DOI: 10.5339/qfarc.2018.ictpd877

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical public safety issue of road traffic accidents, specifically focusing on the delay in emergency response times caused by the lack of immediate accident detection. While visual surveillance systems are common, they often fail under adverse weather or cluttered conditions. To mitigate this, the author proposes an audio-based surveillance method capable of automatically detecting hazardous events, such as tire skidding and car crashes, amidst background noise. The motivation stems from the need for more reliable, multi-modal surveillance systems that can function where visual data is insufficient, thereby enabling faster alerts to police and medical teams. The methodology involves transforming one-dimensional audio signals into two-dimensional time-frequency (TF) representations, which are then treated as images for feature extraction. The author utilizes the Extended Modified-B Distribution (EMBD) to generate quadratic time-frequency distributions (QTFDs) from the audio streams. This approach leverages the instantaneous frequency information inherent in TF domains, which outperforms classical time-only or frequency-only analysis for nonstationary signals. From these TF images, Haralick’s texture features are extracted to serve as image descriptors. These features characterize the audio signals, allowing for their classification into multiple categories, distinguishing between normal background activity and abnormal events. The experimental validation was conducted using a large, open-source database of sounds. The dataset included various types of background noise, specifically road and traffic jam sounds, upon which the target events (accidents and skidding) were superimposed. The study compares the proposed TF image pattern recognition approach against a recent study that utilized the same complex dataset and experimental setup. The results demonstrate that the proposed method offers significant advantages over standard signal classification methods. Specifically, the approach achieved an accuracy improvement of up to 6% compared to the baseline study, confirming its superior performance in classifying audio anomalies. The significance of this work lies in its demonstration that treating audio signals as textured images via time-frequency distributions provides a robust alternative to traditional signal processing methods. By improving the reliability of audio-based surveillance, this approach contributes to the development of more effective road safety systems. The findings suggest that integrating audio analysis with existing visual surveillance can enhance overall system reliability, particularly in environments where visual cues are compromised. This method offers a viable pathway for reducing response times in emergency situations by ensuring accurate and timely detection of road accidents.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• auditory warnings