08/06/2018

An industry review on driver fatigue systems for fleet operators

Why is driver fatigue a problem?

Every year, more than 1 million people die in traffic-related accidents around the world andresearch shows that driver fatigue (or drowsiness) contributes to about 25-33% of fatal and serious accidents. This is equivalent to 300,000 lives or more lost every year. For people aged between 15 and 29, road traffic accidents are the most common way to die.
In order to tackle this issue, automotive technologists have developed the so called driver fatigue monitoring systems, which are able to detect driver fatigue and hence reduce the probability of traffic accidents.

This is particularly relevant in the public transport sector and last year Transport For London announced a fund of £500,000 for bus operators to develop safety improvements, fatigue monitoring being a key part of this initiative.

What is driver fatigue?

Driver fatigue refers to a driver’s mental or physical disorder due to lack of sleep or lengthy time on the road, which results in decreased control of the vehicle.

This is not an accurate definition and reflects the complexity and enormity of tackling fatigue within fleet operating company [3]. In fact, fatigue, let alone driver fatigue, does not have a scientific definition.

A complete but subjective measure of fatigue is given by the Karolinska Sleepiness Scale (KSS), which describes 9 levels of drowsiness. This is a subjective scale, which can be used to self-assess fatigue. Although it has been tried in the past, this methodology can not be effectively used to continuously monitor individual levels of alertness.

Technologists have come up with a number of alternative ways of monitoring fatigue. These solutions are based on the relationship between the fatigue/drowsiness and body temperature, electrical skin resistance, eye movement, breathing rate, heart rate, brain activity or even driving behavioural parameters. Let’s take a look at these systems in more detail.

1. Physiological systems

When a driver is tired, his physiological response becomes slower, his body's response to stimulation delays and his physiological indicators deviate from a normal state. Therefore, by collecting a driver's physiological parameters one can determine if the driver is tired or not. There are multiple systems in this category (ECG, EEG, EMG, Pulse beat and breathing frequency).

ECG is the only non-intrusive method in this category. It's an effective monitoring system but computationally difficult and very sensitive. There are two applications in the market within this category of solutions: Warden or CardioWheel.

2. Driving facing camera systems

These systems are based on the fact that facial features of a fatigued driver differ from that of an alert driver. The latency between the visual stimulus and its response is one of the main measures to determine a driver's consciousness. This latency is known by a parameter called Psychomotor Vigilance Task (PVT) that shows the speed of response of a person to his/her visual stimulation. Research has shown that there is a very close relation between PVT and the percentage of closed eyes in a period of time (also called PERCLOS). Driver facing camera systems use this relation to estimate driver fatigue/drowsiness.

PERCLOS is a key parameter but other facial features such as head position detection, gaze direction detection, eye blink frequency detection, microsleep and mouth state detection are also important.

There are four leading brands in the market with operator connected driving facing camera solutions:

3. Driving behavioral systems

When alertness reaches a low level, the driver’s ability to observe the surroundings, judge the situation and control the vehicle declines. This reduces the precision control of the driver, which is normally reflected by abnormal driving parameters. This can be harsh braking, sudden acceleration or lane departure for example. The acquisition of this data from the vehicle is easy and it is possible to indirectly detect the driver's level of fatigue through the modelling of these data.

Popular solutions include steering wheel angle detection, steering wheel grip detection , vehicle speed detection, vehicle offset detection, brake pedal force detection and accelerator pedal force detection. This area is most developed for the car industry, due to its simplicity but it has the disadvantage of only detecting an advanced fatigue state. Providers of these solutions include most car manufacturing brands like Mercedes or Volvo, and technology providers such as Bosch. A long list of solutions within this category can be found in wikipedia.

Comparison of driver fatigue monitoring systems.

Detecting fatigue can be a good way to reduce accidents. We have learned that the industry has developed 3 main families of driver fatigue monitoring systems, each with multiple solutions and approaches, with their own advantages and disadvantages (see table below).

The main advantage of the fatigue detection based on driver physiological parameters is that it can objectively and accurately reflect the degree of driver fatigue. The disadvantages are that data acquisition equipment is complex and expensive; the most accurate solutions require the equipment to contact the driver’s body directly, which affects the normal operation of the driver. Therefore, apart from ECG applications mentioned above, this method has some limitations.

The advantage of the fatigue detection based on driver face and eyes is that it can accurately determine the degree of driver fatigue. However, the recognition algorithm is complex, the feature extraction is difficult, and the detection results are easily affected by illumination and occlusion. Deep learning technology can make a remarkable achievement in this area and there are signs in the market to believe that this type of solution will be mainstream in the near future.

Fatigue detection based on vehicle driving parameters has the advantage of being simple. The disadvantages are that its analytics are easily influenced by personal driving habits, the weather, traffic conditions and other external factors, the accuracy is not enough; moreover, this method can detect fatigue only when the driver is about to have an accident.

Conclusion

As we have learned about KSS, fatigue is not black or white, but a dynamic array of greys. In fact, fatigue has to be understood as a risk, that is, the probability of having an accident. And as with any risk, there is low, medium and high risk. While technology has made a remarkable improvement over the years, it is still a challenge to detect a "particular type of drowsiness".

Fatigue has to be understood as a risk, that is, the probability of having an accident.

Little research has been conducted in the area of fatigue metrics as a risk. Because of the difficulty to define fatigue, most trials to date have focused on detecting higher levels of fatigue. While this brings some confidence about the technology, we are somehow abandoning the benefits of detecting lower risk events. Is this desirable or acceptable?

More emphasis could be put on detecting different levels of fatigue, including lower risk events. Most importantly, these systems should be able to translate fatigue into actionable information for fleet operators.

From this point of view, I think developers can gain interacting with commercial and public transport operator leaders. Together we could aim to develop and agree fatigue related risk assessment metrics and integrate real-time and historical data with operator fatigue management systems.