F2008-08-037
Driver Sleepiness Detection by Video Image Processing
Driver sleepiness detection by video image processing
The number of traffic accident casualties has been a serious concern although it has been decreasing lately thanks to the infrastructure improvement and passive safety technology of vehicles.
The accident analysis shows that driver distraction and inattention is the major cause. As NHTSA reports, to reduce fatal traffic accidents, reduction of human-error related accidents, which represents more than 90% of all traffic accidents, deems to be urgent.
As you know, driver operates a vehicle by `driving,ī `turningī and `stoppingī, through performing cognition, judgment, and operation sequences appropriately. Accident risk is known to increase when this procedure doesnīt work well, which results in the delays of cognition, judgment and/or operation.
We　focused on the active safety technology to assist drivers in the cognitive/judgment field, where not enough research has been conducted, and are developing a driver monitoring system that detects driverīs unsafe activities. This system continuously picks up the driverīs face image with a camera installed in front of the driver, and processes the image to estimate the driverīs state. It allows detection of inattentive/drowsy driving, and awakes the driver by giving warnings, etc. as necessary.
This paper discusses the result of our sleepiness detection research while driving.
Our driver monitoring system is premised on installing to mass production vehicles, and the system is configured with a near-infrared camera (VGA, 30fps), a near-infrared projector (870nm) and an image-processing unit. It can monitor non-contactly the driverīs eyelid behavior and estimates the sleepiness.
Regarding the sleepiness detection, many researches similar to ours were reported in the past. The correlation between our sleepiness criteria adopted for this research and the conventional one didnīt seem to be strong, showing variation due to individual differences. For this research, we took the average of the objective evaluation by three or more skilled observers, who saw the video of subjectīs face and judged the sleepiness level based on the 5-rank criteria arranged from KSS. It was used as the sleepiness criteria.
To improve the accuracy of our estimation, we repeatedly experimented in drowsy driving with a driving simulator, and extracted several features through carefully observation of driverīs eyelid states. After picking out the eyelid features which can be detected by image processing, We derived a estimation function form these features using by multiple regression analysis method.
At the same time, we made additional experiments, and detect driverīs sleepiness level by applying the function to 10 subjects. As a result, we clarify that their sensitivity and specificity of sleepiness level are 75% and more on average.
Our future approach will be focused on the verification of this result through experiments using actual vehicles.
This abstract is supplemented by a PDF, which can be viewed here.
Session: HMI & Safety