By Joseph Averkamp, senior director of Technology, Policy, and Technical Strategy for Xerox Transportation

What is the best way to automate enforcement in high-occupancy vehicle (HOV) and high-occupancy toll (HOT) lanes? To answer that question, we ran pilot tests on the Xerox Vehicle Passenger Detection System (XVPDS) over the past 18 months. We’ve uncovered some impressive results.

Subscribe to email updates to receive the latest transportation industry trends and tips.

HOV and HOT lanes encourage carpooling, which can reduce traffic congestion in our cities. The obvious weakness is when motorists use these lanes even when they don’t have enough passengers in their cars. Our passenger detection system uses video analytics technology to count the number of passengers, and automated tools that deliver citations to people who violate the rules. During our pilots and presentations, we’ve been asked a lot of the same questions over and over again. They are great questions, and we present the answers here.

How do you manage privacy?

Facial images are redacted for privacy purposes, and we do not keep images of non-violators. There are two ways the images could be redacted: permanent or temporary. The option depends on the client’s goal and preference. All data is encrypted and stored locally.

Does XVPDS perform well in inclement weather?

We have noticed that the system’s 98.5 percent accuracy falls slightly in inclement weather. This is because of rain or other precipitation on the windshield that blocks the camera’s line of sight. For instances such as these, you have to ask, what is the alternative? The other option would be to have human counters standing outside in the cold or the rainy weather, trying to identify the number of passengers inside a vehicle. Through our pilot tests, we’ve noted that human counters are accurate only 35 percent of the time in good weather. (It’s not an easy task.) In our recent trial under rush hour conditions, human observers were challenged to score approximately 660 vehicles per hour, or 11 per minute. This means the human counters had to evaluate a vehicle every 5 seconds or so without a break for a three-hour stretch.

Clearly, this is a task that is performed better by a machine. Xerox VPDS is able to handle this level of activity without breaking a sweat. Compared to the 35  percent accuracy figure for humans, XVPDS is still the way to go. A compromised image would likely be caught in the review process and a fine would not be issued.

Could vehicles go by faster than the camera can catch them?

No, not really. When the system is properly configured, we can capture every vehicle whether they’re in stop-and-go traffic or moving 100 mph. Don’t try: It would be extremely dangerous to try and out-speed our system.

What if there’s a baby in the backseat, in a rear-facing car seat?

Our system identifies passengers by the line of sight, which means the system wouldn’t be able to see a baby in a rear-facing backseat. It would also be difficult for a human to see a baby in this position, too. An image like this would likely be caught in the back office by the reviewers and a fine would not be issued.

Can it detect the difference between a dummy and a human?

Yes, the system has done well detecting this trick in pilot tests, however we haven’t performed a dedicated study to determine its accuracy. During our pilots, we noticed that using a dummy, or another inanimate object, to try and trick the system is a rare occurrence. We’re always making improvements to XVPDS based on what we’ve learned in real-life scenarios. One thing we know for sure: Compared to human counters, the computers usually have the advantage. The system would truly be maximized for success if we combine the strengths of humans and machine by  transferring  human roadside counters to a position of XVPDS image reviewers.

Do you have a question regarding Xerox Vehicle Passenger Detection System? Send us an email at and we’ll give you an answer.

Subscribe to this blog and receive email updates when we publish a new article.