DMS: Don't fall asleep at the wheel – I'm watching you!
Driving for long periods of time at high speeds can make a driver drowsy. The drowsiness is worsened by the hypnotic effect of monotonous road conditions combined with the warm afternoon sun. The driver gradually becomes sluggish, his vision blurs, and in just a few seconds, an accident can happen.
According to statistics, drunk driving accounts for about 2.5% of traffic accidents, and drowsy driving accounts for 20% of traffic accidents – or eight times as many as drunk driving. What’s more, drowsy driving accounts for 40% of major accidents, making it the main cause of traffic accidents.
In recent years, the development of smart driving technology has picked up speed, and various auxiliary driving functions have emerged in quick succession. However, the main responsibility still falls on the shoulders of the driver. Until humans can comfortably "hand over the steering wheel" to machines, "human-machine co-driving" will dominate the automotive market. A recent development in this scenario is the Driver Monitor System (DMS), a technology that prevents drowsy driving by intelligently judging and monitoring the driving behavior of drivers.
When DMS was born, the driving state of the vehicle was mostly monitored through the steering wheel and torque sensor. However, these passive DMS systems have complex models, high costs and a high false alarm rate. Furthermore, they cannot be linked to a smart cockpit. Their inferior intelligence also made it impossible to accurately judge the true intention of the driver, effectively making them useless. Meanwhile, the rapid development of AI vision expedited the emergence of active DMS. This highly intelligent visual DMS technology, based on facial recognition, infrared technology, and visual images, quickly became the first choice of major car manufacturers.
Typically, DMS solutions use infrared imaging technology that can adapt to various light sources and has a strong anti-obstruction ability. The infrared sensors deliver high-quality imaging even in challenging lighting conditions, such as at night and when backlit. As they are not obstructed by various types of lenses, blocked eye information can be accurately captured.
In DMS solutions, the infrared camera lens consists of an infrared light-source transmitter, a lens that can identify the light reflected from objects, the image sensor and other components. By installing infrared camera lenses on the steering wheel, odometer, or the A pillar near the windshield, image or video information regarding the driver's eye status, head posture, smoking behavior, and manner of making phone calls can easily be captured. Camera lens configuration usually requires 1MP to 2MP, and uses 940nm, near-infrared light which is imperceptible to the human eye. With its high intelligence, non-contact functionality, high reliability and low cost, it’s no wonder that DMS based on near-infrared technology is dominating the automotive market.
Recognition accuracy is one of the core indicators of driver monitoring systems, especially when it comes to monitoring whether the driver is fatigued, distracted or drowsy. Measuring and quantifying fatigue is naturally the top priority for DMS. With the facial recognition algorithm model, dynamic video is recognized and judged, and decisions are made, frame by frame. DMS then sends the results of data analysis and processing back to the terminal display and can also exchange information with the user through seat belt vibrations (somatosensory) and smell (olfactory sensation). However, this level of functionality requires manufacturers to make massive investments in terms of manpower, material resources, and financial resources to organize personnel tests, collect facial patterns of people in states of fatigue, distraction, and other abnormal states, and create a database for the accumulated reference data. By referencing the accumulated data in the database and analog training, statistics and analysis are performed on key physiological indicators to draw the boundaries between awake, micro-fatigue, semi-fatigue, and deep fatigue. Then model structure parameters can be optimized, fatigue physiological models can be established, physiological thresholds can be set, and trigger conditions can be determined. In addition, efficient engineering implementations such as instruction-set optimization, multi-thread optimization, and real-time algorithmic scheduling are required to optimize and improve the system response speed.
Experiments have proven that, in terms of the algorithm for measuring the driver's fatigue, PERCLOS (a term that refers to the percentage of eyelid closure over the pupil within a specific timeframe) is the most reliable indicator of the driver's fatigue level. In particular, PERCLOSP80 (the percentage of time the eyelid covers more than 80% of the eyeball) has the most accurate correlation with the driver’s level of fatigue. Hence the PERCLOS detection method is favored by the industry.
Endnote:
Until the arrival of "completely autonomous driving technology", the monitoring of drivers will probably become the best way to avoid the risk of drowsy driving, and DMS will become more and more integral to driver safety. Currently, Cadillac, Tesla, and other automotive brands have woken up to the benefits of DMS and deployed DMS products in their vehicles. With the introduction of smart vision technology, DMS will inevitably be incorporated into broader application scenarios. In the foreseeable future, it will almost certainly become a standard feature of smart automobiles and become an increasingly important interface for the establishment of human-machine interaction between drivers and vehicles.
