Xerox counts on machine vision for high occupancy enforcement
One of the installations in the Halifax trial
Machine vision techniques can provide solutions to some of the traffic planners most enduring problems
With a high proportion of cars being occupied by the driver alone, one of the easiest, most environmentally friendly and cheapest methods of reducing congestion is to encourage more people to travel in each vehicle. So to persuade people to share rides, high occupancy lanes were devised to prioritise vehicles with (typically) three of more people on board and in some areas these vehicles are exempt from tolls or qualify for a discount.
As such, high occupancy vehicle (HOV) lanes have been a feature of many major roads in North America since the 70s and subsequently in Australia, New Zealand and Indonesia and are starting to be seen in Europe. However, one of the major drawbacks to the operation and efficacy of HOV lanes is enforcement as these lanes are often under-utilised and driver-only vehicles tend to use them on the grounds they won’t get caught. Now the situation has become even more difficult as some authorities are allowing driver-only vehicles to use HOV lanes on the payment of a toll (called High Occupancy Toll or HOT lanes).
Now, Xerox has used machine vision techniques to devise a system that automatically determines the number of passengers in a vehicle, enabling authorities to detect non-qualifying drivers using the High Occupancy Vehicle (HOV) and (HOT) lanes. Traditionally HOV/HOT enforcement has entailed local police visually confirming each vehicle has the required number of occupants and chasing offenders to issue a ticket. “Today, officers must park on the shoulder of a highway and quickly merge into traffic to chase down the violator, putting both the officer and the public at risk,” says Mark Cantelli, vice president and chief technology officer of Xerox’s Government and Transportation Sector.
Consequently, enforcement of HOV and HOT lanes has been sporadic, tempting some drivers in non-qualifying vehicles to use the restricted lanes and risk detection. Such behaviour degrades the effectiveness of HOV/ HOT lanes, making them less advantageous for drivers with a genuine case to use the restricted roadway.
To counter these problems Xerox has developed a Vehicle Passenger Detection System (VPDS) which automatically counts the number of passengers in vehicles with better than 95% accuracy.
According to the company, seeing inside a vehicle poses a variety of problems ranging from line of sight into vehicles of different heights, window tinting and vehicle geometries, to the more traditionally difficult weather, lighting and vehicle lane tracking conditions. It says it has developed algorithms which can detect a windshield regardless of the vehicle’s shape and is able to accurately process the images of vehicles travelling at any speed from stop/start to 160km/h (100mph).
Cameras for the passenger detection system can be mounted either alongside or above the high-occupancy lane
The camera itself can be mounted on existing infrastructure alongside or above the lane. Having captured an image, video analytics and geometric algorithms are used to distinguish between empty seats and those occupied by humans. According to Xerox, pets, luggage, groceries or blankets will not be misidentified as ‘occupants’.
Additional testing for ‘inflatable passengers’ is underway but Xerox is confident that the combination of VPDS and a manual image review means few, if any, dummies will have accurate enough representations of human faces to fool the system. Cantelli is keen to stress, however: “We do not employ facial recognition.”
Having determined the number of occupants, if that doesn’t match the setting on the HOT lane transponder, the system will send a real-time violation alert to the enforcement agency. “This detection system automates the process and improves safety through the use of high-quality images and the generation of an evidence package,” says Cantelli, adding: “any driver thinking of moving into a HOV lane without the required number of passengers needs to think again.”
According to Xerox’s solutions director Richard Harris, a trial of the system in the Canadian city of Halifax saw some 40,000 vehicles passing the cameras and the accuracy of the occupant count was more than 98%. To ensure privacy, the system does not connect the HOV transponder IDs with location information; data regarding offenders is encrypted and stored locally while data on non-violators is not saved.
Because different countries and authorities have varying requirements for prosecuting violators, if required possible offenders can be reported in real time to allow the enforcement agency to make a visual confirmation.
Simultaneously an evidence fi le is created which can include time-stamped images showing the interior of the vehicle to illustrate the number of people on-board although the faces will be blanked out. Other images will show the vehicle and the registration plate while the date, location and direction of travel will automatically be included as will a contemporary record of the over-lane display.
The system is still under trial and Xerox expects authorities adopting the technology to follow a similar procedure as with photo enforcement of speed and red light violations, with pictures of the offence posted on a restricted website for the ‘defendant’ to view. “It’s a case of sitting down with the authority and discussing what they need,’ says Cantelli.
Other benefits of the system include allowing agencies to determine how many people are travelling and not just how many vehicles, which can inform roadway management and traffic planning decisions. While the system is available for purchase Xerox is currently concentrating on setting up pilot schemes to allow agencies to become comfortable with the technology. “It takes people a while to get comfortable and write the policies around how they use the technology,” says Cantelli.
Harris believes that with the technology now available to enforce HOV and HOT lane requirements, authorities that had not previously set up lanes may now look to introduce them as part of their strategies to combat congestion. To that end he is considering setting up trials and demonstrations to areas outside of North America, including Europe.
This is perhaps a perfect example of a machine vision solution to traffic planners’ problems.