Sep 072015

In 2013, the City of Piedmont, California, spent almost $600,000 to purchase 39 license plate readers covering most of its border with Oakland. With a population of less than 11,000 people, these 39 license plate readers collect photographs and license plates from more than 1,000,000 vehicles every month. The City of Piedmont sends this data to a regional license plate data warehouse at the Northern California Regional Intelligence Center (NCRIC), where it is stored for one year, even if the data does not generate a “hit” as a stolen vehicle, being registered to a wanted individual, etc.

The City of Oakland, with a population of more than 400,000 people, took three years to gather 4.6 million license plate reads and photographs (see We know where you’ve been: Ars acquires 4.6M license plate scans from the cops). In Piedmont, that same amount of data would be collected in less than 5 months.

Using the information gathered by Oakland’s license plate readers, Ars Technica was able to determine where Oakland City Councilmember Dan Kalb worked and lived with his vehicle captured just 51 times between May 2012 and May 2014. Data from license plate readers could reveal churchgoing habits, whether you visit a medical marijuana dispensary or a health clinic, and whether you spend the night with someone other than your spouse.

Until June 2014, NCRIC generated a report on total license plate reads submitted to it by each agency. Combined with the number of hits reported by the Piedmont Police Department, this data shows that 99.97% of the data collected by Piedmont’s license plate readers is useless – it is data collected about people who are not charged with or suspected of any crime. For example, in April 2014, Piedmont submitted 1,420,244 license plate reads and photographs to NCRIC and only generated 400 hits. That is a hit rate of 0.00028 or 0.028 percent. Below is a table showing this information from December 2013 to June 2014:

Month Total reads Hits Percentage
12/2013 1272871 532 0.042
1/2014 1201196 374 0.031
2/2014 1025771 276 0.027
3/2014 1189422 323 0.027
4/2014 1420244 400 0.028
5/2014 1462313 465 0.032
6/2014 1213121 391 0.032


Source documents:


  3 Responses to “Piedmont License Plate Reader Analysis Shows 99.97% of Data Collected is Useless”

  1. The 99.97% stat is stupid. A stat that would matter would be how much did each hit cost?
    How do you calculate cost? The 600k divided by the expected lifetime of the equipment, plus the monthly costs to run the system.
    Also, of the hits, what percentage resulted in arrests? Were any violent offenders, or were they all non violent? Etc.

    • All good questions. Some other questions: Since license plate readers mostly alert on stolen cars (as represented by the license plates), were more stolen cars recovered than before license plate readers? How much more quickly were stolen vehicles recovered (if they were recovered more quickly)?

      Even better: how much value, if any, is there in retaining the data collected by license plate readers for more than a day or two?

      None of these questions can be answered until law enforcement collects better data and is more transparent with its data. Far too often, we’ve seen claims from law enforcement about crime or safety that don’t stand up to the most basic scrutiny. There is no basis for keeping the data for a day, two weeks, a month, or a year.

      So why does NCRIC keep the data for one year? In their own words:

      “The Focus Group agreed that achieving standardization and as much uniformity as possible throughout the region presents the strongest stance possible to both law makers and privacy-minded groups such as the ACLU.”

  2. […] readers is from people who are not suspected of are charged with any crime. That closely matches a similar analysis in 2014 that showed 99.97% of the data collected from Piedmont’s license plate readers did not generate a […]

 Leave a Reply

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>