Posted by Admin: System Admin
Driving safety has drawn much public attention in recent years due to the fast-growing number of cars. Smoking is one of the threats to driving safety but is often ignored by drivers. Existing works on smoking detection either work in contact manner or need additional devices. This motivates us to explore the practicability of using smartphones to detect smoking events in driving environment. In this paper, we propose a cigarette smoking detection system, named HearSmoking, which only uses acoustic sensors on smartphones to improve driving safety. After investigating typical smoking habits of drivers, including hand movement and chest fluctuation, we design an acoustic signal to be emitted by the speaker and received by the microphone. We calculate Relative Correlation Coefficient of received signals to obtain movement patterns of hands and chest. The processed data is sent into a trained Convolutional Neural Network for classification of hand movement. We also design a method to detect respiration at the same time. To improve system performance, we further analyse the periodicity of the composite smoking motion. Through extensive experiments in real driving environments, HearSmoking detects smoking events with an average total accuracy of 93:44% in real-time. Machine learning is an important component of the growing field of data science. Through the use of statistical methods, different type of algorithms is trained to make classifications or predictions, and to uncover key insights in this project. These insights subsequently drive decision making within applications and businesses, ideally impacting key growth metrics. Machine learning algorithms build a model based on this project data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of datasets, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks.
Driving state detection using smartphones. With the increase of public awareness about road safety, many works on driving state detection using smartphones emerge to improve the quality of daily driving. SenSpeed [11] is a system for accurate vehicle speed estimation, which can estimate vehicle speed by integrating the readings of accelerometers in smartphone. D3-Guard [13] proposes a drowsy driving detection system, which leverages audio sensors in smartphones, to detect drowsy actions and alert drowsy drivers. TEXIVE [14] uses smartphones to distinguish drivers from passengers and detect texting operations during driving according to irregular and rich micro-movements of users. V-Sense [9] develops a vehicle steering detection middleware that can run on commodity smartphones to detect various vehicle maneuvers, including lane-changes, turns, and driving on curvy roads. Various kinds of works indicate the powerful capability of smartphones and embedded sensors. However, research about smoking detection in driving environment using smartphones is absent. This motivate us to propose HearSmoking to detect and alert drivers’ smoking behavior. Smoking detection in contact manner. Technologies and studies on smoking detection using specialized devices, e.g., smart bracelets, smartwatches and chest belts, have been developed for some time. HLSDA [15] is a smoking detection algorithm, which collects various sensor data from a smartwatch and recognizes smoking behavior. PACT [16] is another wearable sensor system based on support vector machines. It detects smoking events by monitoring cigarette-to-mouth hand gestures in a contact manner. By capturing arm movements and breath puffs from 6-axis inertial sensors worn on two wrists of the user, puffMarker [17] builds a model based on 10-fold cross-validation to detect cigarette smoking. Another study [18] investigates the differences in brain signals of craving smokers, noncraving smokers, and non-smokers. This study uses data from resting-state EEG devices to train predictive models based on residual neural networks, and can distinguish the three groups. These works are all based on contact manner that need users to wear additional devices. Thus, they either cost high price to deploy or suffer inconvenience in daily using. Non-contact and device-free methods are needed for the smoking detection. Smoking detection in non-contact manner. Non-contact methods are proposed by using civil cameras, gas sensors, Wi-Fi devices, etc. Smokey [8] is a smoking detection system that depends on Wi-Fi infrastructure. It leverages the smoking patterns leaving on Wi-Fi signals to identify the smoking activity even in the through-wall environments. A self-determined mechanism [19] is proposed to analyse smoking related events directly from videos by combining color re-projection techniques, Gaussian mixture models and hierarchical holographic modeling framework. Besides cameras, gas sensors are widely used. UbiLighter [20] detects cigarette smoking by using a gas sensor embedded in lighters to capture the gas from burning tobacco. A smoking monitoring method [21] uses a microphone to distinguish smoking breath from non-smoking breath. However, due to the great influence on accuracy that ambient noises would bring by only using the microphone, the system is not suitable to be used in cars. The methods based on computer vision heavily depend on lighting and weather condition. Moreover, companies are not allowed to use the cameras and other recording devices due to privacy regulations in some areas [4]. Other methods based on specific sensors are costly or difficult to be deployed in cars. HearSmoking. Different from existing works, HearSmoking detects drivers’ smoking behaviors only using smartphones. HearSmoking can be applied in many ways. It can help to supervise the drivers of nosmoking vehicles, such as taxis and buses. In particular, HearSmoking is very suitable to be used in Uber and Lyft, since if a passenger complaint a smoking driver, it is easier for Uber and Lyft to obtain evidence from HearSmoking. It can also work with other systems to improve driving safety. For example, it can be integrated with the ubiquitous driving modes and navigation systems on the smartphones. Furthermore, if the detection results are uploaded to the transportation department, the police can further understand the driver’s state when dealing with traffic accidents. Disadvantages ? An existing system that doesn't focus on improving the quality of daily driving by using smartphones emerge in quantity. ? An existing system doesn't focus on multiple body movements when a driver is smoking during driving, e.g., steering with one hand, holding cigarette with another hand, putting up and down the cigarette, inhaling and exhaling smoke with chest expanding and shrinking.
we propose a smoking detection system, named HearSmoking, which only uses acoustic sensors on smartphones in driving environments. We first analyse the smoking behaviors of 17 drivers, and find the typical smoking steps of drivers. To perceive motions, we let the smartphone speaker sends designed acoustic signals. The acoustic signals are reflected by surrounding objects and then received by the smartphone microphone. To get distances between reflectors and the smartphone, we calculate Relative Correlation Coefficient (RCC) of the collected data. Further, we get a set of sequence profiles from RCC profiles. Each sequence profile describes distance changes between moving objects and the smartphone over a period of time. According to our observations, when a driver is smoking, his/her main moving parts are hands and chest, so HearSmoking focuses on detecting movements of hands and chest. For hand movement detection, we innovatively transform a sequence profile into a two-dimension image, and then send the image to a carefully designed Convolutional Neural Network (CNN) to identify whether there is a movement that matches the smoking hand movement pattern in the sequence profile. For chest movement detection, we perform Fast Fourier Transform (FFT) to find out waveforms in sequence profiles that fit human breath rate. Then a major breath path is selected to eliminate multipath interference. We analyse the amplitude and period of the waveform to determine whether there is a breath similar to smoking breath. If both hand movement and breath pattern fit the characteristic of those in a smoking event, we then analyse the periodicity of the detected composite motion to improve system performance. Finally, we get an analysis result whether the driver is smoking or not. To meet realistic demands, we collect training data using smartphones for 5 months to build the system model. We implement HearSmoking on different versions of Android platforms and comprehensively evaluate its performance in various environments. Experiment results show that HearSmoking is reliable and efficient in real driving environments. _ We study the unique patterns of smoking behaviors during driving. Based on our findings, we propose a smoking detection system, HearSmoking, which uses acoustic sensors embedded in smartphones to detect smoking events of drivers. To the best of our knowledge, we are the first to design a smoking detection system by only using smartphones. _ We divide the smoking detection into hand movement classification and respiration identification. We innovatively combine acoustic signal processing with CNN-based image classification into HearSmoking. After that, we design the methods of composite analysis and periodicity analysis to obtain the final detection result. _ We conduct extensive experiments in real driving environments. HearSmoking achieves an average total accuracy of 93:44% for smoking event detection. Advantages ? The system proposes a smoking detection system, named HearSmoking, which only uses acoustic sensors on smartphones in driving environments. ? The system analyses the amplitude and period of the waveform to determine whether there is a breath similar to smoking breath. If both hand movement and breath pattern fit the characteristic of those in a smoking event, we then analyse the periodicity of the detected composite motion to improve system performance.