Repo men are pushing the bounds of personal surveillance by scanning billions of license plates—perhaps even yours.
Only Scott Toth knew exactly how this would go. He deftly hooked his tow truck to a gleaming white 2012 Chevy Cruze in a matter of seconds. A young woman in the passenger seat jumped out and screamed toward the convenience-store entrance: “Somebody get — ! He’s inside!” Then Toth began securing straps to the wheels and hoisted the Chevy onto a hydraulic boom that resembled a giant spatula. The car’s owner emerged from the store, advancing angrily. Other drivers pulled out of adjacent spaces like cowboys abandoning their bar stools.
At 5’11” with wide shoulders and a buzz cut, Toth—a repossession agent out of Cleveland—looks a little like a drill sergeant, though he has considerably better people skills. To calm older drivers, he’ll turn the conversation to his four daughters, projecting the image of a beleaguered 34-year-old father with a house full of teens. But with the Chevy’s glaring driver—a guy of about 20 with an athletic build, camo baseball cap, and carefully considered facial scruff—this clearly wouldn’t work. Instead, Toth lit a cigarette. The driver, seeing the gesture, paused and sparked up too. For a moment, the two stood like country gentlemen enjoying the evening light. “Sounds like you haven’t made a few payments,” Toth said amiably, looking out at the horizon.
“That’s about right,” the man said.
“Can I have the keys?” Toth asked. Keys would help get the Chevy out of park and into neutral, making any getaway much smoother.
“I ain’t giving you s—,” the driver responded. His female companion stood to one side, anxiously chewing her fingernails.
With that, Toth nonchalantly climbed back into his Dodge 3500 and gunned the 6.7-liter Cummins diesel, performing what’s known in the industry as a “drag,” yanking the Chevy across the convenience-store lot, leaving parallel black scars on the pavement behind him. About 50 feet away, he pulled into an empty lot, unhooked the Chevy’s back bumper, and drove around to the front so he could tow the car freely on its rear wheels. Before setting off, he uploaded a few photos of the vehicle’s condition to the company intranet through the truck’s encrypted WiFi hotspot. Meanwhile, back at the store, the car’s owner ranted loudly to his friends, then hustled into a white pickup. Still within earshot, Toth didn’t look up. “He’s just mad,” he said, “and there’s nothing he can do about it.” Toth has performed this maneuver more than 4,000 times, and over that period, he has watched as his repo firm, Relentless, grew from a one-truck shop into a high-volume “collateral recovery agency” with 20 trucks and thousands of dollars in digital equipment. He has also seen his business transform from a simple one (find car, take car) into one that involves the sophisticated coordination of data and vehicle logistics.
Toth didn’t just happen upon the white Chevy, after all. His truck is customized with tens of thousands of dollars’ worth of cameras and image processors and can scan plates even while tearing down a highway. Earlier that day, he had received a tip: The bank that held the auto loan sent an electronic packet of information on the Chevy, gleaned from previous plate scans, that contained the car’s historical coordinates. Toth matched the data with the driver’s personal details, and with a laptop as his co-pilot, he used it to predict the Chevy’s location.
In the world of repossession, Toth’s ability to pull together strands of data into a coherent story is becoming the norm. In just a few years, companies like Relentless have quietly captured billions of GPS-stamped license-plate scans—accounting for more than half of all cars on the road today. That information is stored in a handful of centralized databases, managed by companies few people have ever heard of. There are, of course, more sophisticated ways to pinpoint someone’s location—GPS coordinates, cellphone bread crumbs, and facial recognition. But those are held in check by technical hurdles or established privacy policies. License-plate capture is both relatively easy and perfectly legal (at least in most states). And that’s had the unlikely effect of putting repo men like Toth at the frontier of personal data and surveillance.
When he finished uploading the images, Toth pulled out of the parking lot with a roar, the Chevy jouncing behind us. I had to admire his efficiency, but I was also left with the nagging question of how I might feel had the car been mine. Life is busy. It is not inconceivable to think of missing a few payments on my auto loan. And the plates on my Subaru station wagon back in New England must certainly be among the billions contained in one of the national private databases. Chances are, they’re cued up and waiting, ready to tell the story of my life.
Repossession agents have been fixtures of the auto industry since the rise of accessible credit in the 1920s. Banks hire them to collect a vehicle, sometimes the day it goes into default. The agent tows the car, and the bank pays between 300 to 800 bucks a claim. Successful agents have always had a head for numbers and facts, assembling a driver profile to predict the location of a parked car. But the ability to learn about drivers rapidly accelerated in the early 2000s. Repo sites such as Skiptracers and Merlin Data opened vast databases to agents—property addresses, military-service dates, electric-company bills, spouse names, criminal histories, bankruptcies—aggregating information with computer efficiency. Then the premier white-collar research company, LexisNexis, got into the field in 2004, with its Accurint site for repossessors and police, offering an even broader set of online tools (with a subscription). To counter that move, rival TLO.com, owned by credit bureau TransUnion, offered much the same data for just a buck per search. When license-plate cameras became available to repo agencies in 2009, many started mounting units on trucks. As an agent drove down the street, the cameras captured plates while computers ran them against cars on several banks’ default lists. By just cruising around, an agent could significantly boost the number of hits. A single truck could scan as many as 8,000 plates in one day.
“Relentless has hired a handful of ‘scouts’ whose sole purpose is to suck up license plates all day.”
By 2010, license-plate scanners had become standard equipment for most urban repo firms, and the number of plates stored in national databases was growing by tens of millions a month. Even though there are about a quarter of a billion vehicles in the U.S. total, cars are often scanned a dozen times or more in different locations. The richer the data gets, the easier it is to make predictions about a driver’s home address, workplace, gym, or favorite restaurant. Digital Recognition Network (DRN) has one of the largest plate-capture databases in the country, with a fleet of more than 2,000 affiliated trucks and upwards of 1.8 billion scans. According to DRN, the technology increases the number of cars repossessed by 14 percent. “It allows repossession agents to work more efficiently and to look at data insights to more effectively predict where the car may be,” says DRN’s CEO, Chris Metaxas, a former vice president of sales at Lexis-Nexis who oversaw its government division.
Spurred by success, repo firms have begun to make data collection an even greater part of their operations. Toth’s employer, Relentless, has hired a handful of “scouts” whose sole purpose is to suck up license plates all day. One such person is Lori Jones. For eight hours a day, six days a week, the suburban mother of four tools around Cleveland in an unassuming Honda Fit. Hidden in its air vents is a $23,000 camera suite—including a 20-millimeter lens to spot cars in motion and a 50-millimeter lens to capture vehicles parked 60 feet up a driveway. Where the back seat used to be, a rack-mounted imaging system extracts plate numbers from a photo and stamps them with the time and GPS coordinates. Jones and three other scouts in the Relentless fleet capture nearly a million images per month in Ohio.
Typically, Jones focuses on large parking lots, apartment buildings, and businesses. When she gets a hit, a sound like an air-raid siren goes off. If the claim is parked, she hops out of her Fit to double-check the VIN number and call in a repo agent like Toth. She even scans during breaks. On an average day, she’ll be at the entrance to a local mall, picking at a Chipotle salad while monitoring a laptop screen with thousands of images flashing over it. “I like to be productive during my lunch hour,” she says cheerfully. It’s hard to imagine a less threatening face for personal surveillance.
As Toth hauled down Route 20 toward Euclid, a succession of miniature-golf courses, crowded soft-serve joints, and late-summer farm fields swept by in a blur. He glanced at his various screens and amiably shared trade secrets: During a long shift, a truck-stop shower is a small luxury at just $12; a pack of hot dogs works wonders in a neighborhood with ferocious canines; and winter is the best time to repo because few people are willing to run down the street in their underpants.
Toth first learned about the field 10 years ago, when his own car was repossessed. He started an agency, then became the field manager for Relentless. With his commissions, he quickly paid off his house; the job, he says, instills a healthy respect for the hazards of debt. As the volume at Relentless has increased, so too have the commissions, at least for top field agents like Toth. “You used to pick up maybe one car a day, though you had almost no information on the driver. Now we get as many as five a day, and I have everything I need right here,” he said, nodding to the electronics on his console. His laptop gives him the orders through a secure repossession portal, displaying the name of the driver, address, and historical plates. From the same portal he can follow a reclaimed car’s condition as it moves through processing back at Relentless (cleaning, taking inventory of the contents, and moving to auction in 45 days).
Outside the truck, four cameras sit on either side of the bed, each capable of collecting up to 1,800 scans per minute. To improve accuracy, each camera is ringed by powerful LEDs that shine infrared light undetectable to the human eye. The light helps illuminate plates in darkness at a distance of up to 60 feet. According to the camera’s manufacturer, Vigilant, it can also defeat license-plate covers meant to obscure scans.
The scan starts out as a low-resolution black-and-white, but image-processing software locates the license plate within it. Character-recognition software then extracts the plate number. Finally, the image, GPS data, time, and plate number are sent wirelessly to the database vendor in Forth Worth, Texas.
For all the technology that surrounds him, Toth is quick to point out that scanning plates remains a small part of the job; the main task is still grabbing cars and managing people under tense circumstances. Data may help him zero in, but it doesn’t do a whole lot when it comes to dealing with the vagaries of human nature. “You just never know what the day will bring,” he said, bearing down on a two-hot-dog lunch at the wheel.
As if on cue, the white Chevy strapped to his truck began to act up. First, it chirped and its lights flashed. Then the horn blared. In the distance, we could see a dirty white pickup accelerating toward us. The Chevy’s erstwhile owner, following us, still had his key fob and was remotely flipping the door locks and activating the horn. Toth checked the rearview and nodded. “He’s just trying to be a jerk,” he said. He had over 50 miles to lose the pickup before he arrived at the impound lot. “The last thing you want is to have people follow you back to home base,” he said. “If they can catch you when the gate is open, then they have the advantage.” Toth has had drivers try to block his path and attempt to retrieve the car by force.
For the next 15 minutes, the white pickup bombed down Route 20 to keep pace with us. When Toth pulled to the side, it passed us but soon reappeared ahead, crawling along the breakdown lane, biding time until we caught up. When Toth taxied into a Walmart lot, it followed and parked at a distance. Toth whipped around and pulled up next to it, affecting the stentorian tone of a high school principal: “Is there a problem?” For a split second, no one blinked.
Then the man’s face softened, and he shook his head. “No, I was just out . . . driving around,” he offered.
Perhaps the man needed a little more driving time to wrap his head around what had just occurred. Repo has never been easy or pleasant for those in default. But the speed and accuracy with which it can happen today perhaps makes it more jarring still. Having one’s personal data tracked is a fact of life. Credit scores are the most basic example, but in recent years we’ve added others: GPS, facial recognition, Web cookies, store loyalty cards, fitness data, Klout Scores. But in every case, the tracking seems somehow distant—separated either by technical and legal hurdles or by the notion that users can opt out. Right or wrong, those data streams don’t seem to carry real-world consequences. But when someone uses data to guess your favored haunts and reclaim your car, the consequences become very real, very fast.
A few hours after dropping off the Chevy and taking a smoke break, Toth was stalking a yellow 2011 Camaro in a dark parking lot when a message popped up on his phone: All staff were to gather downtown over a case of Red Stripe. We soon arrived at a row of houseboats moored on a squiggly part of the Cuyahoga River, what’s called Collision Bend, and spotted the firm’s CEO, John Ziebro, swinging by his hands from a wooden gate over the river. “You don’t want to end up in that water,” Toth cracked, pointing at the troubled canal that infamously caught on fire in the 1960s. Standing in front of an antique houseboat was John’s brother David Ziebro, a co-owner of Relentless, along with one of the most successful female repo agents in the industry, another co-owner, Amy Bednar.
As everyone dug into cartons of Vietnamese takeout, the three owners encircled me to discuss license-plate scanning, about which they were both proud and a little defensive. Scanning by private industry has become a controversial topic, the subject of active bills in more than 17 state legislatures and a practice outlawed in New Hampshire, Maine, and Vermont. John Ziebro pointed out that the technology saves the banks millions, allowing them to offer consumers more loans under more liberal underwriting policies and, in some cases, with lower interest rates. There’s also the law-enforcement argument: Police departments across the country can order license-plate scans during an active investigation. When the vendor of the Ziebros’ cameras, Vigilant, surveyed more than 500 police departments in 2013, respondents cited 2,180 crimes solved with the help of the data, including homicides, drug trafficking, and in-progress abductions. Almost 40,000 stolen cars have been recovered too. In part because of the law-enforcement benefits of private plate scanning, lawmakers recently overturned a state ban in Utah and tabled a proposed ban in California.
Even so, as we talked I started thinking about the life cycle of databases: When they start, their users tease out a few simple connections between data points. But as they evolve, they can support sophisticated predictive models of consumer habits. When I spoke with DRN’s Metaxas, he talked about the company’s interest in exploring how license-plate-scan data could boost customer service by predicting a consumer’s financial trouble. Banks, he said, are starting to look at the data before a car goes into default. “The real goal is not to repossess the vehicle,” said Metaxas. “If you can look at data to more effectively predict where the car may be, you will help the finance company improve its customer life cycle.”
For example, if a series of plate scans indicates that a certain car no longer parks at the owner’s usual workplace, the bank could infer a change of employment and may make a phone call to offer a lower monthly payment. It would not be much different from how credit card companies call customers when they notice an unusual pattern of transactions. For that matter, license-plate data could have value to customers beyond banking. Metaxas has voiced interest in selling database access to insurers, credit card companies, and nonauto lenders. It’s not hard to foresee a scenario in which the use is broader still: drivers subject to coupon offers and marketing messages whether they want them or not. In the U.K., a motor-oil company illegally scanned drivers’ plates as they passed on the freeway, cross- referencing the make and model of car, and flashed the type of oil the drivers should use on a billboard.
Critics have raised concerns about the security of private license-plate scans, which are less regulated than those captured by police. For Relentless, the scans reside in DRN’s database, which, according to Metaxas, is “managed and maintained to the highest security standards.” The problem, say privacy advocates such as Jennifer Lynch, a senior staff attorney with the Electronic Frontier Foundation, is that such security measures are not subject to public audit. “We don’t know how they’re securing their systems and who has access—we just sort of have to rely on the company’s word,” she says. If a breach occurred, anyone could abscond with bulk scans. That’s not as bad as it sounds: A hacker would need to tie records to real identities in order to misuse them, a task that would require breaking into a state’s locked motor-vehicle registry as well.
In the end, the most profound impact on people’s lives may not be triggered by data breaches but by the legal use of plate scans by private companies. “There are a lot of public places where people still would like to maintain their privacy,” says Lynch, “whether they’re driving to a firing range, church, mosque, abortion clinic, or gay bar.”
“We don’t know how they’re securing their systems and who has access—we just sort of have to rely on the company’s word.”
The French philosopher Michel Foucault had the idea that once citizens believe they’re being observed, they start to internalize a desire to avoid anything that would cause offense to those in power. If bankers and insurers are keeping tabs on plates—and they have the ability to approve loans and offer low insurance premiums—then wouldn’t it behoove a driver to travel through a better part of town or forgo buying an affordable house in a low-end neighborhood?
Part of the problem is that the scans tell someone a story about you, but if you’re not in control of the story or don’t even know it’s being told, it can seem manipulative. Personal data has always been, to a degree, about an exchange of value. Companies like Facebook get your data, but you get something in return (reconnection with high school buddies). With plate scans the trade is decidedly one-way. Auto financers get to mitigate risk—and lock drivers out of the story. A compromise could be making one’s own plate scans as transparent as credit card transactions. A password-protected site with recent scans would help lessen the sting of private surveillance and provide value—at least I could see where my teenager has been driving. But private license-plate-scanning companies have been lobbying for tighter restrictions on this data, arguing that plate scans released to private citizens create greater potential for misused information.
It was past midnight when the discussion on the river finally wound down. With the food and beer dispatched, it was time for the night’s main attraction: testing a camera-mounted hobby quadcopter. David Ziebro demoed the chopper, showing us the aerial video and enthusiastically explaining, “Believe it or not, this will be an important tool for us in the future.” Toth nursed his beer and issued a skeptical glance, but David Ziebro soldiered on, pointing out that more than 30 percent of defaulted vehicles still slip away from repossession agents forever, often hidden behind tall fences on private property. If the Federal Aviation Administration allows it in 2015, a remotely operated vehicle like the one on hand could be a legal way to spot them from public airspace above a person’s house.
As the black quadcopter swooped and dove over neighboring houseboats, it was a reminder that repo’s future lies in its ability to find new ways to push the boundaries of privacy. During the week I spent with them, the repossession agents of Relentless were always rule-abiding, cautious, and respectful, but their overall mission is, of course, to find as many cars as possible. How they can do so without affecting the 99.5 percent of us not in default remains an open question.
57 Years Of Computer Vision (And Counting)
Put enough processing power behind a digital camera and you’ve got “computer vision,” the process by which machines can analyze the visual world. Since the advent of the transistor, systems that can do this have become cheaper, faster, and smaller. Here’s a quick overview of the highs and lows in the technology’s history.
1957: The first computer scanner copies a 2-inch photo of the inventor Russell A. Kirsch’s son.
1964: Defense contractors Woody Bledsoe, Helen Chan Wolf, and Charles Bisson launch a facial-recognition system for an unnamed intelligence agency.
1976: U.K. police invent a license-plate recognition system. The first major installation is in 1993, as a “ring of steel” around London to counteract IRA bombings.
1985: The first autonomous land vehicle, made by Lockheed Martin, Carnegie Mellon, and others, uses video-based imaging to follow a road at three mph.
2004: Mars rovers Spirit and Opportunity land on the Red Planet using computer vision to calculate distance and position on descent.
2008: The first 3-D pizza-sorting system, the “Scorpion,” builds a 3-D profile of 7,200 products per hour using multiple cameras. It automatically culls misshapen pies.
2010: Shortly thereafter, a man hacks the device to track his own nipples for the first time.
2014: Phone processors become fast enough to handle pattern recognition. Apps such as Vhoto pick worthy stills from a video based on action sequences and facial expressions.
This article originally appeared in the July 2014 issue of Popular Science_._