London Police Facial Recognition Struggles To Find Suspects
The London Metropolitan police's latest live facial recognition (LFR) deployment found no suspects with one false alert that resulted in a wrongful questioning.
Powered by NEC software and Axis cameras, this marks the second deployment this year with no suspects found. The results were criticized by privacy activists as proof that LFR is a "charade" that "must end".
The Met emphasized the system made three arrests from an earlier deployment this year. An IPVM review of results so far indicates that the ratio of false engagements is improving, although overall the trials have resulted in just three arrests per year.
Outside of PRC China, LFR usage by police is not popular due to privacy and accuracy/bias concerns. In this post, IPVM examines the Met's LFR results so far.
Poor ******* *** ****** **********
*** *** *** *********** *** ***** ****, ********** * ***** ** ** separate *********** **** *** *** **** of "*********** *** ********** ** * policing **** ** **** ********** *** identification ** ******** ** ******** ** a ********** ********".
** **** **, *** *** ********* an *** ***** ** ****** ****** - *** ** ******'* ******* ************* -********** ********* **,*** ***** ******* * watchlist ** *,*** ********. *******, *** results **** **** **** *** **** alerted *** ******: * ***** ***** that ******** ** ** "**********" *.*. police *********** ** ******** ******:
*** **** ** ********** *** *** the **** ********** **** **** *******. On ***** ****, *** ****** ******* ********* **** ******** **********, ******** **** one *** ** ***** ******:
*** ******* ** *** **** ** deployment **** **************** ****** *****, ******** ** ******* rights ***** *** ******* *****:
*** **** ******* **** ******** ******** in*** ***** ****:
****: *** ******* ***** ** **************** * ********** *** ****** *********** ** *** UK.
NEC ******** *** **** ****
*** ******* ** *** *** **** are *** ******** *****-********* ********** ** ******** ** ****, ** **** ** ************* *** ******* *****:
*** *** ********* ** **** **** it ** ********* ***** ********** *********** *********** ******** **** ** ******** company ***, ** ** *** **** doing***** ******** *** ****** *****.
"False ***********" *******
***** *********** ***** **** ****** ******** and/or ****** ****** *** *** ****** flagged ** ********, *** ** *** actually ******* *******, ****** *** ******* privacy *** ****** ****. (**** *********** occur **** ****** ******** ****** ********).
*** ******* ***** *** ********** *** false ***********, **** **** *** **** 14th **********, *.*. * ******* ******* man *** ******* *** *** *** ID *** *********** ********** ** *******:
** ******* ********, * ************* ****** exchange ******* *** ************ ****.
Overall ****** *** ***** ******* ********
*** ***'* ******** ******* ******** **** its * *** *********** ***** **** have ******** ** ** **** *********** at *** **** ** * ***** engagements, ********* ** * *******/"*********":
*** *** *** **** ********* *** results ** *******-**** ******, ******* ***** **** ** ***** engagements *** ** **** *********** ********* in * *******/"*******":
**** ***** *** ***** ** ***** engagements ** **** *********** ** *********, from *.* *** ****-**** ** *.** for ****-****.
*******, *** *** ****** **** *** resulted ** * ***** ****** ** arrests ***** ****: * ***** ** 18 ******* ***** ** ******* ** 3 ******* *** ****, ** **** one *** **********.
******* *** ******** *********** *** ********* to *******: *** ** *** * deployments ***** ****, * *** ** 9 ******* ******** ****** **** *** deployments.
Costs ********
** ****,*** *** **** ********* *** ***** of *** *** ** £***,*** (***** ~$250,000 ** *** ****) *** *** first *** ***********, *.*. ~$**,*** *** deployment:
**** *** *** "******** *** ******** only", ******* ** *****'* ******* ******** and ***** ***** *.*. *** ******* Watch's **************** ************ *** *** ************* ******** ***** ***.
"Gross ***** ** ****** *********"
****** *****, ******** ** *** ******* Watch, **** **** **** *** ******* are * "***** ***** ** ****** resources" ** "***** * ******* ** people *** ***** ** ********** ** other *****":
**'* ******* **************** *** ******* ******* to *** **** ****** *********** ******* in ******* ******, ******** ******** ** thousands ** ****** *n ***** ** ***** * ******* ** ****** who could be identified by other means. It is a gross ***** ** ****** *********, let alone a serious privacy intrusion on a scale only seen in oppressive regimes. Live facial recognition is dangerously authoritarian, highly inaccurate, and is turning our public streets into police line ups. This Orwellian surveillance has no place in Britain at all. [emphasis added]
******** ************* *** ** "*** **** ** works *** *** **** ** *****'* work."
Met ****** ********
*** *** ********* *********** *** **** * **********, ***** resulted ** ***** ******* (*** ******* drug ******* *** *** ****** ****** for *******) ******* **** * "********* result":
*** *** ** **** ****** *********** technology ***directly ****** ** ** ****** ***** ****** *********** *** ******** **** **** **** ** ************ ****** **** **** *** *******. It is a fantastic ****** **** *** ********** and it links to one of the Met’s top priorities of tackling serious violent crime. This innovative technology, alongside our officers, enables us to find people that pose a serious risk to our community so that we can keep the people of London safe. [emphasis added]
*******, *** *** *** *** ******** addressed *** *********** ***** **** **** suspects. ** ******** ** ******** ***** these ***********, *** *** ******** **** to *************** ***** ** ******* ***** **** *** *** *** comment ************.
Poor ******* **** **** ** ********** ** ****
*** ***'* **** *** ********** ********, **** ****** ****. **** **** ****** ** the ******* ** **** ***** **** they *** ********.
******: ******* **** ****** **** *** latest *************. **** *** "****" ***** *** no *********** ** ******* *** ** ~16,440 ***** *******:
*** ******* ***** ******** ****** ************** *** "******* *********" *** "******* failure":
** ** ****, ***** ** *********** incorrect **** ***** **** * *******, the*********** * **** ***** ******** *** police *** *** ****** **** *****.
*'* ********* ***** ****** ***** ** getting *** "****** **********" ****. ***** on *** ******* ** **** ******* they're ******* ****** **** **.**% ********. I ******* **** *** **** **** to ****** ***** *********, *** ***** negatives.
*'* **** **** ** *** **** "other *****" *** *** ** **** to ******* *** *** ****.
* ******* **** *** **** **** to ****** ***** *********, *** ***** negatives.
******* *** ***** ********* ******** * completely ********* ***********.
*.*. *** ***** **** ** ****** a **** ** *** *** **** and **** **** ****** **** **** face **** ******* *** ****** ***. there ** ** ******* ** ****...
* ******** ******* ***** ******* ** all ** ******** ** *** ******** of *** **********, *** *** * or * ***** ********* **** ********* the ********** ** *** *** ** being ****.
** #*'* *****, *******, *** *** police ******* *** **** ***** ******** matches *** *** ****** ********* ** the ****** ** ***** ***********?
** ****, * **** *** *** system **** ******* ******* *** **** face ********. * ** ****** **** there **** ***** ***** *** *** system ********** **** * ****** ******* by ******* * ******** ** * sufficiently **** ***** ** ********** *** the ****** ******** ****** ** *** match *** ********** **** *** ****** were *** ***** ****** ** **** stop *** ****** ******* **. ** we **** / ** **** ***** those *****?
*** *** ****** ******* *** **** false ******** ******* *** *** ****** generated ** *** ****** ** ***** questioning?
***, **** ** *** ********** ******* a "***** *****" *** "***** ***** engagements". *** *******, ** ******** **** 2020, *** *** *** * ********** where **** *** * ***** *******, 7 ** ***** **** *****; ** those ***** *******, * (********) ****** were **********:
**** *** ********** *** *** *** significant *** ***** ***** *** *** system******"* ***** ************** **** ** ** percent", *.*. */*.
* ** ****** **** ***** **** cases ***** *** *** ****** ********** that * ****** ******* ** ******* a ******** ** * ************ **** level ** ********** *** *** ****** officers ****** ** *** ***** *** determined **** *** ****** **** *** close ****** ** **** **** *** person ******* **. ** ** **** / ** **** ***** ***** *****?
***, * ***** **** ** *** difference ******* * "***** *****" *** a "***** ***** **********". ** *** February **** **********, *** ****** **** shows * ***** ****** *** **** 5 ****** **** **********. *** ***** two *** **** *** **********, *** police *** **** ******** **** ****** them **** *** ****** ******* *** faulty *** ** ****'* ***** ************* them ** *** (** **** **** got ****, ******* *****'* ** ******* breakdown.)
**** ***********!
************, *** *** *** ***** ***** rate **. **** ***** ** ****** how ************ **** *** ****** *** system, **** *** **** ** ***** so **** ** **** ** *** side ** ******* **** ******** ****** than ******* ******** ** ** ******** person.
** **** **** ******* *** ** those ************* / ****** *******? * think ******** *** *** ****'* * fascinating *******. ** *** **** ****** one, **** ******* ** **** ******* ~36,000 ***** ******** ** * ********* of ~*,*** **** **** * ***** false *****.
*** ***** ***** * ** ********* is *** * ********* ** ~*,***. How *** **** **** **** **** / *****? ** ** **** ****** that **** ***** **** ** ** the ****? ****** ** **** *,*** people ** * ******* ** ~** million ****** ** * ****** ****** in * ******** ******** (*.*., * in **,*** ** *,*** / ** million).
****** ******'* ******* ******* ** *** **** ******* who **** ** *** *********. *********, they *** "*** ********* ********* **** of * ********* **** *** ***** Watchlist *********** ** *** *** – intelligence, ******** *** ******** ******** *** policing ******** **":
****** *** ***** **** ** *** Watchlist *** ** * ******** ****** on ******** (***** *** ********* *** proportionality **** *** **** **** *** for **** ****** **** ** ** possible ** *** ** * *********), this ** *********** ** ** * very **** **********. ** ** **** crucial ** **** ****the ********* ********* **** ** * ********* **** *** ***** ********* *********** ** *** *** – ************, ******** *** ******** ******** *** ******** ******** ** in line with the MPS’s strategic objectives as set out in the MPS LFR documents. Together, they may justify the necessity and proportionality of the particular Watchlist’s composition and the need to Deploy LFR using a Watchlist designed for the needs of that Deployment. [emphasis added]
*** **** *********, *** ********* ** "driven ** *** ******** ********** *** intelligence-led ***********":
*********: *** ********** ** ***is ****** ** *** ******** ********** *** ************-*** ***********, both of which determine locality and the policing purpose. It is then the locality and policing purpose which determine the composition of the Watchlist. The individuals found on a Watchlist are there because there is a policing need to locate them, and that need fits with the policing purpose driving the LFR Deployment. This may include those aged under 18, those under 13, a person with a disability (as defined in the MPS LFR SOP) or vulnerable adults where there is a policing need and it is deemed to be necessary and proportionate to locate and/or safeguard these people. The MPS LFR Documents outline considerations regarding expectations of privacy, and outlines specific controls and safeguards to mitigate any impact on those with a protected characteristic(s). [emphasis added]
** **** **** ** ****** ** the *** ** ************ *** *** for ******** ** ** ** ** PRC *****-*****, ***** *** ****** **** no *******/**** *********** *** *** ******** towards ********** ** '*** ****** *** better'.
*****'* **** ** ****** ** ****, where ** *** ***'* **** ***** harming ******** **********, *** *** ******* more.
******: ******* **** ****** **** *** latest *************. **** *** "****" ***** *** no *********** ** ******* *** ** ~16,440 ***** *******:
*** ******* ***** ******** ****** ************** *** "******* *********" *** "******* failure":
** ** ****, ***** ** *********** incorrect **** ***** **** * *******, the*********** * **** ***** ******** *** police *** *** ****** **** *****.
** ********* ** ****** ** * similar ***** ***** ********* **** *********** cameras, * ***** ****** *** ** heavy ****** ** ****** ** **********