Acoustic-Based Emergency Vehicle Detection Using Convolutional Neural Networks

Tran, Van-Thuan; Tsai, Wei-Ho

doi:10.1109/access.2020.2988986

Cited by 75 publications

(36 citation statements)

References 22 publications

(56 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…[101] A CNN-based ensemble model was proposed to classify traffic soundscape to noise, siren sounds, and other vehicle sounds. [102] The method demonstrated 96% accuracy for even short 0.25 s samples to correctly classify emergency sirens.…”

Section: Voice-based Interactionmentioning

confidence: 93%

“…[99] The microphones typically present within the car can also be placed on the exterior to detect sirens, vehicles in proximity, and even pedestrians. [100][101][102] Other interesting use-cases of acoustic sensors include learning algorithm for audio-only odometry that only measured the acoustic signals from external microphones with good prediction accuracy. [103] This system was not affected by the scene's appearance, lighting conditions, and structure.…”

Section: Auditory Technologiesmentioning

confidence: 99%

See 1 more Smart Citation

Intelligent In‐Vehicle Interaction Technologies

Murali

Kaboli

Dahiya

2021

Advanced Intelligent Systems

View full text Add to dashboard Cite

Autonomous vehicles (AVs) will soon become a reality with almost every major automotive original equipment manufacturer (OEM) such as Tesla, BMW, Daimler, and big technology giants such as Google's Waymo and Apple Car pushing toward the full self-driving objective. According to the SAE J3016 taxonomy [1] of autonomous driving, there are six levels of automated driving systems ranging from level 0 (completely manual) to level 5 (full selfdriving [FSD]) automated systems that are expected to function in all geographic locations, all weather conditions, and under all conditions. The proposed benefits of intelligent vehicles include fewer road accidents, enhanced safety, reduced traffic congestion, effective commute time usage, and importantly enjoyable and comfortable ride. With increased autonomy, the drivers also assume the role of mere passengers who are engaged in nondriving activities and are unavailable to participate in traffic interactions. This would increase complexity in a mixed-autonomy traffic environment as interactions with pedestrians and cyclists are based on visual cues with the driver. Therefore, intelligent vehicles also need to autonomously interact with other traffic participants such as pedestrians and cyclists and other vehicles.Human-vehicle interaction (HVI) is closely related to the field of human-robot interaction (HRI). It entails the problem of understanding and shaping the interaction dynamics between humans and vehicles. Particularly, the domain of interaction involves the study of sensation, perception, information exchange, inference, and decision-making. [2] With increasing levels of automation, the driver will have more time and choice to perform various tasks other than driving and this opens new avenues for interaction. As a result, a growing number of sensors are being incorporated into the vehicles to understand the driver's and/or passenger's actions, emotions, and personal choices to offer the precise functionalities and services for an enjoyable ride. [3] Figure 1 shows the schematic of in-vehicle interaction consisting of various interactive interfaces and sensing modalities present in the vehicle as well as some modes of interaction such as gestures, speech, and eye gaze.The advances in HVI have been presented in previous review articles, which have mainly focused on the detection of particular characteristics for driving assistance such as driver attention detection, [4] emotion recognition, [5] drowsiness detection, [6] mental workload, [7] and so on. A few other articles have reviewed the interior designs of AVs to better support and interact with drivers and passengers. [8,9] Likewise, the user interfaces (UIs) and user experience (UX) of vehicle interiors have been studied in context with manual as well as automated driving. [10] A number of previous works have also studied the interaction with the external

show abstract

Section: Voice-based Interactionmentioning

confidence: 93%

Section: Auditory Technologiesmentioning

confidence: 99%

Intelligent In‐Vehicle Interaction Technologies

Murali

Kaboli

Dahiya

2021

Advanced Intelligent Systems

View full text Add to dashboard Cite

show abstract

“…However, it should be noted that the acquisition quality and the number of sound types will affect the accuracy to some extent. [48] Siren sounds of ambulances, police cars and fire engines, car horns, urban noise.…”

Section: ⅵ Discussionmentioning

confidence: 99%

An Acoustic-Based Recognition Algorithm for the Unreleased Braking of Railway Wagons in Marshalling Yards

Zhang

Liang

2020

IEEE Access

View full text Add to dashboard Cite

The abnormal braking of wagons is a challenging safety problem for operators at railway marshalling yard. This paper develops an acoustic-based technology to detect the unreleased braking of wagons during the uncoupling operation. Experiments have been conducted to collect the acoustic waves of wagons abnormal braking, as well as the background sounds like train whistling and wheel vibration. Before data collection, a wayside recording system and an experimental train composed of 5 different wagons have been prepared in the marshalling yard. The recognition algorithm consists of the fast Fourier transform (FFT), feature extraction, template matching and support vector machine (SVM) classification. Based on the sample data of different acoustic waves, the FFT is firstly performed to obtain the frequency spectrum from original time-domain signals. Then the major spectrum features of different sounds are carefully extracted for SVM training through a newly-devised algorithm, where the features include the spectrum center, spectrum flux, energy peak and corresponding frequency. During the SVM training, classifiers are designed under oneagainst-one strategy to guarantee the recognition accuracy. Given a test data, at most 3 SVM classifiers will be activated according to the decision matrix of template matching. Meanwhile, rules have been made to regulate the classification result considering different activation cases. Finally, a case study of all 12 sound categories has been performed to illustrate the application of proposed algorithm. Results show that the acoustic-based recognition algorithm is indeed reliable to identify wagons unreleased braking, with the global warning accuracy over 98%.

show abstract

“…However, these traditional machine learning methods are not suitable for training large-scale samples. Recently, with the development of deep learning methods in the field of pattern recognition, CNNs have also been widely used in voice recognition systems [8,9,10,11,13]. We divide the research of CNN in environmental sound recognition into three categories.…”

Section: Introductionmentioning

confidence: 99%

Environmental Sound Classification Method Based on Two-Stream Lightweight Convolutional Neural Network

Fang¹,

Yin²,

Ji³

et al. 2021

Preprint

View full text Add to dashboard Cite

Neural networks have achieved success in the task of environmental sound classification. However, the traditional neural network model has too many parameters and high computational cost. The lightweight networks solve these problems by compressing parameters, but reduce the classification accuracy. To solve the problems in existing research, we propose a two-stream model based on two lightweight convolutional neural networks, called TSLCNN-DS, which saves memory and improves the classification performance of environmental sounds. Specifically, we first used data patching and data balancing to slightly expand the amount of experimental data. Then we designed two lightweight and efficient classification networks based on the attention mechanism and residual learning. Finally, the Dempster-Shafer evidence theory is used to fuse the output of the two networks, and the two-stream model is integrated. Experiments have shown that the model has achieved a classification accuracy of 97.44% on the UrbanSound8k dataset, using only 0.12 M parameters.

show abstract

Acoustic-Based Emergency Vehicle Detection Using Convolutional Neural Networks

Cited by 75 publications

References 22 publications

Intelligent In‐Vehicle Interaction Technologies

Intelligent In‐Vehicle Interaction Technologies

An Acoustic-Based Recognition Algorithm for the Unreleased Braking of Railway Wagons in Marshalling Yards

Environmental Sound Classification Method Based on Two-Stream Lightweight Convolutional Neural Network

Contact Info

Product

Resources

About