A global problem that costs lives
Driver fatigue is one of the most underestimated risk factors in commercial transportation. According to the World Health Organization (WHO), traffic accidents cause approximately 1.35 million deaths globally each year, and it is estimated that between 10% and 20% of these crashes are directly related to driver fatigue. The International Labour Organization (ILO) notes that transport workers face significantly higher accident rates than the average across other industries, largely due to extended shifts and irregular sleep patterns.
For companies with commercial fleets, the impact of fatigue goes beyond human tragedy. Every accident generates costs from material damage, downtime, increased insurance premiums, and potential legal liabilities. Preventing these events requires technology that acts before the driver loses control — and this is where artificial intelligence makes the difference.
How a DMS camera with artificial intelligence works
A Driver Monitoring System (DMS) is a camera equipped with artificial intelligence processing that analyzes the operator's face and behavior in real time. Unlike mechanical or vehicle-based systems, the DMS directly observes the driver to detect signs of cognitive impairment before they result in an accident.
The DMS camera uses infrared illumination that allows it to operate with the same precision during daylight and in total darkness. The infrared sensor is invisible to the driver, so it does not cause discomfort or distraction during nighttime driving.
68-point facial mapping and the PERCLOS algorithm
The technical core of the system is a computer vision algorithm that identifies and tracks 68 reference points on the driver's face. These points include the contours of the eyes, eyebrows, nose, mouth, and jaw. By analyzing the relative position and movement of these points frame by frame (typically at 30 fps), the system can determine the driver's alertness state with high precision.
One of the key indicators is PERCLOS (Percentage of Eye Closure), originally developed by the United States' NHTSA (National Highway Traffic Safety Administration). PERCLOS measures the percentage of time the eyes remain closed or occluded within a given period. When this percentage exceeds a calibrated threshold — typically between 70% and 80% closure within a specific time window — the system classifies the event as fatigue or drowsiness.
In addition to PERCLOS, modern DMS algorithms analyze blink frequency, eyelid closure speed (slow eyelid closure), head position (nodding, lateral tilt), frequency and duration of yawns, and facial micro-expressions associated with loss of concentration.
Alert types: beyond fatigue
While fatigue detection is the primary function, modern DMS systems with artificial intelligence can identify multiple risk behaviors:
- Fatigue and drowsiness: classified into severity levels (mild and severe) based on the duration and frequency of indicators. Mild fatigue generates a preventive alert, while severe fatigue triggers maximum urgency alarms.
- Microsleep: very brief episodes (1 to 5 seconds) in which the driver falls asleep with eyes closed. At 100 km/h, a 3-second microsleep means traveling 83 meters without vehicle control.
- Cell phone use: detection when the driver holds the device to their ear or looks down at the screen. According to the National Safety Council (NSC), using a cell phone while driving multiplies accident risk by 4.
- Visual distraction: prolonged gaze away from the road, conversation with passengers, or any sustained visual attention diversion. Classified into two severity levels.
- Smoking: especially critical in vehicles transporting hazardous or flammable materials.
- Camera obstruction: alerts when the camera is intentionally or accidentally blocked, ensuring the operational integrity of the system.
Real-time processing: from data to action
The DMS artificial intelligence processes each video frame locally on the device, without relying on internet connectivity for detection. This means alerts are generated in milliseconds, even in areas without cellular coverage — critical for mining operations or remote routes.
When an event is detected, the response is immediate and multi-level: first an audible alarm activates in the cabin, followed by activation of a vibration motor installed in the driver's seat to provide physical stimulation. Simultaneously, the alert with video evidence is transmitted via 4G (when coverage is available) to the fleet management platform, where the supervisor can intervene in real time.
Accumulated data feeds scoring algorithms that evaluate each driver's performance over time, identifying recurring fatigue patterns that may be associated with factors such as rotating shifts, specific routes, or seasonal conditions.
Integration with fleet management platforms
An isolated DMS system has limited value. The real impact is achieved when fatigue alerts integrate with a centralized fleet management platform. This integration enables correlating fatigue events with GPS data, creating dashboards showing trends, and generating driver rankings based on safety performance. For a complete view of DMS and ADAS as complementary systems, see our article on Endeavant smart cameras.
The combination of real-time detection with historical data analysis transforms safety from a reactive approach (investigating accidents) to a predictive one (preventing accidents before they happen).
Driver privacy and ethical considerations
A legitimate concern in DMS implementation is driver privacy. Modern systems are designed to analyze facial patterns without permanently storing identifiable biometric data. Processing is performed locally on the device, images are processed frame by frame without facial reconstruction, and only video clips are stored when a safety event is detected — there is no continuous recording for surveillance purposes.
Organizations such as the European Commission and the NHTSA have established regulatory frameworks that distinguish between safety monitoring (behavior-oriented) and workplace surveillance (person-oriented). Fleet DMS systems fall into the first category when implemented correctly.