Motor carriers use a variety of Internet of Things (IoT) devices and applications to monitor driving behaviors. Most of these technologies are designed to alert management if drivers exceed safe thresholds for speed, acceleration/deceleration rates, g-forces and other vehicle operation and sensor data.
Many can also score driver behaviors and create risk profiles. To get picture-perfect visibility of behaviors and metrics that accurately predict accidents, a growing number of fleets are using video-based driver risk management systems.
With forward and inward-facing event recorders, the systems capture risk factors due to weather, driver distractions, short following distances, traffic violations and other bad habits that cannot be detected by sensors alone.
Some technology providers offer managed services where their analysts will review video clips of critical events to capture such risks. Additionally, some have devised powerful algorithms to correlate driving behaviors with collision risks.
The algorithms translate a complex set of driving behaviors into easy-to-use scores and drill-down reports to manage risk.
The SmartDrive video-based safety system has a scoring algorithm that combines recent observations of risky behaviors with the predictive value of these observations. Its algorithm normalizes individual driver scores for risk exposure based on hours or miles driving.
The SmartDrive Safety Score identifies specific skills and behaviors that individual drivers need to improve on to reduce collision risk. Fleets use these scores alongside video clips to coach drivers and change behaviors.
Postal Fleet Services, a major mail hauler that primarily contracts with the United States Postal Service, was not using any technology to monitor driving behaviors before it deployed SmartDrive.
“We did it old school,” says Jeremy Collins, director of business and safety development. The St. Augustine, Fla.-based carrier was assessing risk by looking at MVRs, employment history, logbook records and road tests, among other manual practices.
[related-post id=”149207″/]As a result, management was “reactive” to information that came in from CSA inspections, citizen complaints and accidents, he says.
The company piloted the SmartDrive system and within a few weeks saw its SmartDrive Safety Score decrease by 85 percent. The company has since deployed the technology in its fleet of 700 trucks, he says.
The system gives Postal Fleet Service visibility of risky driving behaviors like texting, not using seat belts and many kinds of distraction. Collins specifically remembers an instance where the system discovered a driver had a bad habit of drifting to the left.
“It was a small behavior that he was not aware of,” Collins says. “That is the level of detail we can get to.”
As fleets adapt new safety technologies on the path towards autonomous vehicles, some risky driver behaviors will go away but others may emerge.
On a monthly basis, Lytx, which provides the video-based DriveCam safety system, analyzes the correlations its algorithms find for behaviors and risk to see if “anything has significantly changed,” says Michael Phillippi, vice president of software development and operations.
Lytx’s algorithms assign point values to risky behaviors from analysis of a database of more than 70 billion driving miles. The Lytx score increases as drivers accumulate risks the system identifies from driving events that trigger video capture.
The risks include observed behaviors and scored combinations of behaviors. An example of a behavior combination is a driver being distracted by a handheld cellphone while having less than two seconds of following distance. This combination would have more risk than the sum of the individual behaviors. In other words: 1+1=3.
Lytx says its algorithms move past causal relationships of risk, such as drowsy driving, and identify correlative behaviors that multiply the risk. Not wearing a seat belt is an example of a correlative behavior. Its data show that an unbelted driver is 3.4 times more likely to get into a collision than an average driver.
The overall goal for the Lytx score, Phillippi says, is to provide fleets and drivers with easy-to-use information that makes coaching events as effective as possible.
On the fast lane to an accident
According to the theory called Heinrich’s Pyramid, for every 300 acts of risky behavior there are 29 acts that lead to a minor incident or scary near-miss, and one act that leads to a serious injury or fatality.
The risks are much higher for driving, according to analysis by Lytx. By analyzing over 50 billion miles of driving data, Lytx found that for every 190 acts of risky behavior there are 6 minor injury and/or property damage collisions and one fatality accident.
Taking a closer look at the Lytx data from DriveCam-captured events, about 50 of the 190 acts of risky behavior are drivers not wearing seatbelts, or being distracted by hands-free/handheld cellphone use, or food and drink in the cab.
Combinations of behaviors are even more serious. Lytx data reveals the riskiest combination of behaviors as:
- Not looking far enough ahead combined + insufficient following distance
- Not looking far enough ahead + handheld cell phone
- Not scanning intersection + entering intersection with a stale yellow light
- Cell phone use (hands-free or held) + poor lane-keeping
Making real-time predictions
Omnitracs offers seven predictive models that tailor to different types of fleets and industry challenges. For larger carriers, the company builds custom models that use any data the customer is willing to share — safety, operations, finance, demographic data and more.
For smaller carriers, Omnitracs offers two predictive models that both use hours-of-service data for different purposes. One detects turnover risk and the other accident risk. By using hours-of-service data, the predictive models have a standardized data set for analysis.
Ram Renganathan, senior data modeler for Omnitracs, says the company is working on projects that will be able to make real-time predictions, and deliver predictions to managers and to drivers. He notes the company has already invested in the infrastructure necessary to handle big data and real-time analytics.
As an example of what is possible, imagine a driver getting a message that recommends pulling over to take a nap when a system detects a high accident risk due to fatigue.
Omnitracs is looking for opportunities to create such real-time predictive models. As one step in this process, its Critical Event Video product could be updated with algorithms that identify traffic signs and objects. This would make it possible to capture drivers’ reaction to speed limits and stop lights as well as monitor their following distances, he says.