What You Can Learn from Oakland’s Raw ALPR Data

Print Friendly, PDF & Email


Privacy info. This embed will serve content from youtube-nocookie.com

Pri­va­cy info. This embed will serve con­tent from youtube-nocookie.com

Police cars mount­ed with auto­mat­ic license plate read­ers (ALPRs) wind their way through the streets of Oak­land like a “Snake” game on an old cell phone. Instead of eat­ing up pix­els of food, these cam­eras gob­ble down thou­sands of license plates each day. And instead of grow­ing a longer tail, ALPRs feed into a giant data­base of loca­tion­al data as they con­duct sur­veil­lance on every dri­ver with­in the city lim­its, and some­times beyond.

This is the por­trait that emerged when EFF ana­lyzed eight days of ALPR data pro­vid­ed by the City of Oak­land in response to a request under the Cal­i­for­nia Pub­lic Records Act.

As cities and coun­ties across the coun­try pur­sue new law enforce­ment tech­nolo­gies, EFF is on a mis­sion to use trans­paren­cy as a coun­ter­bal­ance to mass sur­veil­lance. Since May 2013, EFF and the ACLU of South­ern Cal­i­for­nia have been engaged in a legal bat­tle with two Los Ange­les law enforce­ment agen­cies who are refus­ing to hand over a week’s worth of ALPR data. San Diego Coun­ty, anoth­er juris­dic­tion, has sim­i­lar­ly fought efforts by cit­i­zens to obtain access to data that law enforce­ment has col­lect­ed on them using ALPRs. Both claim that the records are exempt­ed under the Cal­i­for­nia Pub­lic Records Act because they are records of law enforce­ment inves­ti­ga­tions. The agen­cies also argue the pub­lic inter­est in main­tain­ing secre­cy in ALPR data out­weighs the pub­lic inter­est in learn­ing how and where ALPR sys­tems are being used.

The rub here is that law enforce­ment agen­cies like those in LA, San Diego, and Oak­land aren’t using ALPR for tar­get­ed inves­ti­ga­tions, but rather run­ning a drag­net on all dri­vers in their juris­dic­tions. As states across the coun­try become more and more con­cerned about ALPRs and take steps to lim­it their use, we believe the dis­clo­sure of a lim­it­ed amount of license plate records will help to inform pub­lic debate on this mass sur­veil­lance tool.

Events in oth­er juris­dic­tions sup­port our posi­tion. After Muck­rock and the Boston Globe obtained Boston Police ALPR data, the city sus­pend­ed the pro­gram in the wake of the pri­va­cy con­cerns raised by the data. When the Min­neapo­lis Star-Tri­bune obtained ALPR data that it used to track the where­abouts of the may­or, it kicked off debate in the leg­is­la­ture about how to bal­ance the pri­va­cy of inno­cent dri­vers against the abil­i­ty of police to fight crime. As a Min­neapo­lis city offi­cial not­ed at a pub­lic hear­ing on ALPRs after the data release, “now that we see someone’s pat­terns in a graph­ic on a map in a news­pa­per, you real­ize that per­son real­ly does have a right to be secure from peo­ple who might be try­ing to stalk them or fol­low them or inter­fere with them.” A state leg­is­la­tor and for­mer police chief not­ed at that same hear­ing, “even though tech­nol­o­gy is great and it helps catch the bad guys, I don’t want the good guys being kept in a data­base.”

Not all Cal­i­for­nia law enforce­ment agen­cies have fol­lowed Los Ange­les and San Diego’s lead in ALPR secre­cy. Where­as Los Ange­les cops have stalled for more than two years, Oak­land pro­vid­ed raw ALPR data in just under two months.

With more than 63,000 data points, it’s a lot of infor­ma­tion to process. We dug into the Oak­land data to show many of the ways ALPR can be bro­ken down and visu­al­ized to help ensure police account­abil­i­ty. It imme­di­ate­ly became clear that with just a few ALPR vehicles—as few as two cars—Oakland is able to cap­ture plate data from across the city, with a par­tic­u­lar focus on low­er income neigh­bor­hoods. The data also shows that police cars pick up license plates when mak­ing the jour­ney to coun­ty jail (that’s the long tail extend­ing to the east). The data does not seem to indi­cate that Oak­land has any ALPR cam­eras mount­ed in fixed locations.

Today we’re releas­ing the data to the pub­lic, with the indi­vid­ual license plate num­bers removed to pro­tect the pri­va­cy of dri­vers cap­tured by these cam­eras. (While LAPD and LASD also claim the public’s right to pri­va­cy as a rea­son for with­hold­ing the records, the data can be anonymized eas­i­ly with a few clicks, either by delet­ing a col­umn for the spread­sheet or replac­ing the plates with ran­dom num­bers.) We’ve also done some pre­lim­i­nary analy­sis of the data, which we present below. (If you just want the raw data, the links are at the end of the post.)

The Numbers

Total number of data points collected by Oakland Police ALPR cameras

Num­ber of unique indi­vid­ual plates cap­tured by Oak­land Police ALPR cameras

Num­ber of vehi­cles that were cap­tured only once

Num­ber of ALPR reads with­in one mile of Oak­land Police headquarters

Num­ber of cap­tured plates that were like­ly assigned to gov­ern­ment vehi­cles (i.e. police cars, bus­es, coun­ty vehi­cles, etc., which gen­er­al­ly receive plates that are sev­en numer­ic digits)

Num­ber of entries that were obvi­ous bad reads (e.g. the cam­eras picked up road signs such as “CAUTION” or the plate had more than sev­en digits)

Num­ber of times the sin­gle most-cap­tured plate was hit (a gov­ern­ment vehi­cle, like­ly a police vehi­cle, cap­tured mul­ti­ple times at the same loca­tions over a short peri­od of time)

Aver­age num­ber of times an indi­vid­ual plate was captured



The data indi­cates that Oakland’s ALPR pro­gram may mir­ror the nor­mal work­day, pick­ing up like clock­work around 8 am, wan­ing slight­ly at lunchtime, then pick­ing up again in the after­noon. Plate cap­tures dropped off sig­nif­i­cant­ly dur­ing the overnight shifts, with ALPR vehi­cles most­ly going dark between 4 am and 7 am.

ALPR by Frequencyfrequency_0

This chart shows how fre­quent­ly indi­vid­ual plates were cap­tured mul­ti­ple times. The vast major­i­ty of plates were seen only once.

Heat Mapsentire_heat_map

Click to enlarge. The shad­ed area shows the bound­ary of the City of Oakland.

Your plate is more like­ly to be caught on cam­era in a few spe­cif­ic loca­tions. For example:hotspot_downtown

Down­town: Oak­land PD head­quar­ters is locat­ed near the cor­ner of 7th St. and Broad­way, so the increased num­ber of hits in this area are like­ly due to patrol cars trav­el­ing to and from police head­quar­ters.hotspot_northwest_oakland North­west Oak­land: Have a car in north­west Oak­land? Pre­pare to be scanned and recorded.hotspot_intl_blvd_and_fruitvale

Inter­na­tion­al Blvd. and Fruit­vale: The same holds true for Inter­na­tion­al Blvd. and Fruit­vale Ave. (though some neigh­bor­ing areas don’t seem to be tar­get­ed at all).

Surveillance and the Census

Using Tableau Pub­lic map­ping soft­ware, we mapped the ALPR data over var­i­ous lay­ers of data from the U.S. Cen­sus Bureau. In each of these images, the dark­er the col­or, the high­er the intensity.per_capita_census_-_square_crop

Per Capi­ta Income: The data indi­cates low­er-income neigh­bor­hoods are dis­pro­por­tion­ate­ly cap­tured by ALPR patrols, with police vehi­cles cre­at­ing a grid of license plates in the city’s poor­est neighborhoods. white_population_-_square_crop

White Pop­u­la­tion: Per­haps unsur­pris­ing­ly, the per-capi­ta data and the white pop­u­la­tion data sig­nif­i­cant­ly over­laps. If you are dri­ving through or park­ing your car in a neigh­bor­hood with a high­er den­si­ty of white fam­i­lies, you are less like­ly to be picked up by ALPR cam­eras, par­tic­u­lar­ly black_population_-_square_cropnorth­west of State High­way 13.hispanic_population_-_square_crop

Click images to enlarge.

Black and His­pan­ic Pop­u­la­tions: Over­lay­ing Cen­sus data for African-Amer­i­can and Lati­no pop­u­la­tions shows the con­verse of the white population.

ALPR Data vs. Crime Dataall_crime

We also filed a Cal­i­for­nia Pub­lic Records Act request to obtain the Oak­land Police Depart­men­t’s crime data for the same peri­od. Each white dot here indi­cates a record­ed crime. It’s not much of a shock­er that ALPR use doesn’t cor­re­late very well with crime. For exam­ple, OPD did not use ALPR sur­veil­lance in the south­east part of Oak­land near­ly as much as in the north, west, and cen­tral parts of Oak­land, even though there seems to be just as much crime.

To see if per­haps OPD was just focus­ing its ALPR use in areas with high inci­dents of auto­mo­bile-relat­ed crime, we decid­ed to map only the auto-relat­ed crime:auto_crime

The result is the same—ALPRs are clear­ly not being used to deter auto­mo­bile-relat­ed crimes.

ALPR and Mosquesmosques2

In fil­ing requests for ALPR data, we chose one week of the Islam­ic holy month of Ramadan to see whether police were using ALPRs to gath­er intel­li­gence on Mus­lim pop­u­la­tions. When we plot­ted out mosques on the map, we dis­cov­ered sev­er­al were near ALPR hotspots, but there was lit­tle in the data to indi­cate that any par­tic­u­lar focus was placed on these places of wor­ship. Future inquiries worth look­ing into might include gun shops, med­ical mar­i­jua­na col­lec­tives, abor­tion clin­ics, and protests.

ALPR Anomalies

Oak­lan­ders aren’t the only cit­i­zens who should be wor­ried about OPD sur­veil­lance. The ALPR data we received also con­tained instances of ALPR col­lec­tion out­side Oak­land’s city limits.alameda

City of Alame­da: Alame­da is the island to the bot­tom of the map, and is an inde­pen­dent city. Appar­ent­ly at least one offi­cer decid­ed to go spy on its citizens.emeryville

Emeryville: Emeryville is a city that bor­ders Oak­land, and is the por­tion of the map out­side the light-pink shad­ed area. Obvi­ous­ly Oak­land PD doesn’t think twice about sur­veilling its cit­i­zens when they cut across it.piedmont

Pied­mont: Pied­mont (the cen­tral unshad­ed area) is actu­al­ly bor­dered on all sides by Oak­land. As with Emeryville, appar­ent­ly Oak­land PD has not been direct­ed to turn off their ALPR sur­veil­lance devices when they take short­cuts across oth­er jurisdictions.hacienda_crossings_mall

Mall Park­ing Lot: Appar­ent­ly an Oak­land PD offi­cer left his ALPR on while tak­ing a trip out­side the city (like­ly to or from the San­ta Rita jail) and stopped at the Hacien­da Cross­ings Mall in Dublin, near­ly 20 miles away from Oakland.

False Positives

ALPRs aren’t fool­proof. For exam­ple, Cal­i­for­nia cur­rent­ly lim­its van­i­ty plates to sev­en char­ac­ters, but many plates with eight char­ac­ters showed up in the data, includ­ing “CROSSWAL,” “ROSSWALK,” “ROSSINGS,” “CAUTIC1N,” “CAUTICJN,” and “DRIVEWAY.” Obvi­ous­ly none of these were actu­al license plates—in fact, 96 of the entries in the data were sim­ply not pos­si­ble due to being eight or more char­ac­ters long. Instead, they were like­ly read (or mis­read) from traf­fic signs.

We also found oth­er like­ly mis­reads from signs, includ­ing “PLUMBING,” “AHEAD,” “PRIVATE,” “PARKING,” “PARKIMG,” “ALLOWED,” “ORTOWED,” “DORTOWED,” “ONLEFT,” “CAUTON,” “CAUT10N,” and four more vari­a­tions of “CROSSWALK” as well as “ONE WAY.” All told, there were 76 entries that were like­ly mis­reads from road signs (22 of which were over-length).

In anoth­er 95 instances, ALPR cam­eras cap­tured the license plates, but failed to record any geo­graph­ic coor­di­nates. Plot­ted out on a map, it looked like Oak­land police were patrolling the ocean off the coast of Africa.

Don’t Take Our Word for It

Want to take a look at the data your­self? Do you have a bet­ter analy­sis method? Want to draw your own con­clu­sions? Please do! You can find the ALPR data … and the crime data … here in a Google Fusion Table.

Spe­cial thanks goes to Ari Isaak of Evari GIS Con­sult­ing for his help man­ag­ing the data. All heatmaps were cre­at­ed using the awe­some open source heatmap.js project on top of Google Maps.

Updat­ed 1/22/2014: After pub­li­ca­tion of this post, we found a cou­ple of off-by-one errors in our analy­sis. A man­u­al inspec­tion also found many more like­ly mis­reads from road signs, increas­ing the num­ber from 134 to 150. The post was updat­ed to reflect the cor­rect num­bers. We have also made small adjust­ments to the text for clar­i­ty that did not affect the facts of the post.

Oak­land PD ALPR Data, July 20–27, 2014
Oak­land PD Crime Data, July 20–27, 2014