Liopa Lip-Reading Video Analytics Profiled
Being able to read lips on surveillance video could enable law enforcement to solve significant problems, but are lip-reading video analytics ready? One UK company, Liopa, aims to solve this problem.
For this report, IPVM spoke with Liopa CEO Liam McQuillan about the company, its technology and products, and Liopa's potential.
Executive *******
** ** ******** **** ***-******* ***** analytics **** ** ****** **** ** CCTV / ***** ************ *** ***** to **** ***** *** ******** ********* limitations ** ********* **** *** ********* assessing **** *** **** ***** ** lip ********.
******** ******* ** **** ***** *** spin-out **** *****'* ********** ******* *** the****** *** ****** *********** ************ (****), * ******** ****** ** ***** security *** * ****** ** *** university.
*** ******* **** **** **** *** initial ***** *** **********:
***** *********** *** *** ********* ** a ********* **********. *** *** *******'* lips **** **** **** ***** ** very **** ********, *** **** ****** is **** ******* ******* ***** *** different **********. *** ******** ***** **** been **** ** *********** **** ***** biometrics, **** ** *****-***** ******. ** 2018, *** ******* ****** ****** ********* lip *******.
***** **** ** *** * *********,****** **********,**** ** **** **** ** *********** with *****'* **********. ******* ********* ********** consists ** ******* *********, ********** *********, *** ********* ********.
***** ******* ****** ** *** **** lip-reading ********** ******* *** *** **** searching *** *** ***** ************ *** its **********. *** ******* *** **** recently ******* *** ******* ** ****** in *** ********** ********, ** ******** to******* ******** *** *********** ************ ** "****** ************" *** ****** and ***** *******, "****** ******** *********" for ****** *****, *** ***** ******** development **** (****).
**** *** *****, ***** *** **** many ******** ***** *** ********* ** using *** ********** *** **** ************, e.g.:
- (*) "*** ** **** ** ******* large ******* ** **** ******* ** identify ******** ** ********"
- (*) "*****’* ********** *** ** **** to ******** *** ***** ****** ** surveillance ***** ******* (*.*. ****) ***** audio ** ****** *** ******* ** of **** *******,"
- (*) "***** ********* * ************* ** how *** ********** ***** **** **** keyword ******** ** **** *******."
**** ******* ** ***** ****** ******* **** *****'* ****** **** ************ * ******* ** ********* ***** Constable ***** ****** ** ***** ***** Police, ***** *****'* ********* ****** ******** applications.
Limited *********** *** **** ******
*** ******* ******** ** ******* ** its ******* ** *********** ******** ** an ********* **** ****:
** *** * **-****** *******, *** we **** **** ******** *********. **'** just ******* *** ***** ******* ** market, *** **'** **** ** ******* sale ** *** **. *** ** have * ****** ** ****** ******* in **** *** ** *** ** hospitals. ** **** ************* **** * couple ** ********* ** ******** *******, where ***** **** **** **** ********** in *** **** *******. ***** *** for **** ** *** ********** ** two ********* *******. *************, ** ***'* say **** ****** ** *** *** companies *** ******* **** *** *** incorporation ** *** ********** ** ****** them * *********** *********. ***** *** good *********** *** ** ************ *** they **** **** ** *********** ******* as **** ** ******* *******.
***** *** ******** ***** ***** ***, the ******* ********* ** ********* $*.* ********* ******** **** ******.
** ********** *********** **** *****'* ******* sheets *** *** ****** **********.** *** ***** ** ******,***** ******** ~-$**** ** ****, ********* ~-$840k ** *********** *** ~$**** ** assets.
** ****** *** *******'* ******* ** be ***** *** (******* ****** ** less), ***** *** ******* ********** ********* they ****.
LipRead *** *** ************
*** ******* ********* *** ************ ** its ******* ********** ** *** *******: "keyword ******** ** ****** *****," *** a "**********-***** ********** ***" ****** ****** Recognition *** *** ***** ******** (*****), ***** ***** ******** ******** *********** with **** *** *********.
** *** ****** *********** ** *** reading ***********, *** ********* *********, ** is ********* ** ********* *** ********** of *** ****** ********** ***** ** context.
*** ****** ****** ***********, ** * move ** **** ** * ******* sequence, **** ***** **** * ****** of *****. ***** *** ****-***** ********, but ***** ***** **** ********* ******. All ****** ***********, ******* ***** ** visual, *** **** ** *** **** type ** ********** ** *** *****. There *** ***,*** ***** ** *** English ********, *** ***** ************ *** looking *** * ******.
*** *** ***** ***********,***** ****** *** ***** ************ *** ***** ******* ********, ** NHS-managed ******** ************* ** ***** ***** of ********* *** **********. *** ***** app ** ********* ********* ** *** iOS *** ******* *** ******.
Technological ************* *** ***********
***** ********* ********** *** ******** *** ********* parse *******, *** ****** *** ********* associated **** ****** ******, ******* ** "a ******* ** **** ** ***** in ****** ********," *** ******* ****** on ***** *******.
*** **** ** *** *** *****, primarily *** ***** ********** *********** **** appears ** ** ** *** *&* phase, *** ******* ******* ********** ***** LipRead ********** **** ******** *****-***** ********* speech *********** (***) **********.
***** ***** *** ****** ****** *********** work ******** *** ****** **** * lot ** ***** ********* **** ******* on * ***** *** ** * convertible *** **** *** *** ****.
***** *** ********* **** **** **** the ******* ** ********* ******* ** speech *********, ** ********** **** ** speech:
*** ***** ****** **'** ********* * lot ** ****** ** ** *** moment ** ****** *********. ** ****** is *** ***** **** *** ********** of **** ** ******. ***** *** many ****** ******* *** *****. **** and ***** ***** **** **** ** audio. **** **'** ******* ** ** a ****** *** ** ****** *********. For *******, ** ******* *** * video ** ***** ******* ******** *** they **** *** ***** *****, **'** looking ** ********** **** ***** ******** that *****. * **** ** ******* lip ******** ** * **** **** and ********* *** ****** ******.
Natural ********
*** ********* ********* **** ********* *********** words *** ******* ** * **** complex ********* ** ******* **** ******* there *** ***** **** **** *** number ** ******* **** ***** *** spoken *********.
***** *** ***** **** ** *******, there *** ** *** ** ****** lip *********. ** ***** ** *** audio ********** ********, ***** *** ** or ** ******** ***** ******** *** speech. *****'* ********* **** * ***** of ****. ******* ***** *** **** than ** *******, * ******** ** visemes ***** **** ******* ********* ******. That's *** ************ *** ***** ** important ** ********* ******** *** *******'* words.
** ***** ** *** **** *********** on ****** ****** ***********, *** ******** have ** ********* *** ***** **** so **** ***.
Target *******
********* *** *** ********** ******** *** a ****** ****** *** *****. ******** patients *** ***** ** *** ***** app *** "****" ******** **** * smartphone ****** *** *** *** ********** ** ******* ***** *** ********* to ****.
*** ******* **** *** ********* **** an *********** *** ********, **** *** McQuillan, ****** **** "*** ************ *****, our ********** **** **** ****** *** high-resolution ******* **** *** **** ****."
******, ***** **** ************ ****** ******* presents * ********* *** **** **** users "*******" **** **** ***** **** into * ********** ***. ******** ** surveillance ***** *** ** *** *** to ********** **** *** **** ***** lips, *** *** **** ** ***** context ** * ****** ***** ****** much **** **** *** ***** ** interpretation. ********* ** *** ******** ******, therefore, ***** ******* ******** ** ***** from * ************* **********.
***********
**** ** *** ***** ** *** other ********* **** *** ******** ** commercially ***** ** **** ** ***** lip-reading ***** ********* *** **** / video ************.
***** ** ****** ******, *** **** and ********* ***** ** **** ******, certainly ** *** **, *** *** correct.
#*, **** ******** / *****. ** spoke **** ****** ****** ***** ******* from*** ***** **** ********** **********. ** **** **** ***** ** has *** ***** ** ******* ***** lip ******* ***** *********, ** ******* they ***** ******** ** ****** ** after-the-event **************/***** *******, ****** **** **** cases ******* ** ****** ************ *** reactive *** ******* ********* ***** *******. He **** ****** **** ***** ***** surveillance ** *** ****** ****** ** US ******* ** **** ****, ** is **** ****** ** *** ** and ****** *** **** * ** presenter ** *** ********** *** ****** attendees ** ***** **** ***** *** possibility ** ***** ************.
*** * *** ***** ** ** a ******** *******.
***** **** **** ***** ** ****** to *** *** ******* ********.
(**** ********** *'* * ******* ***)
** **** ********** ***** ***, **'** going ** **** ** ***** ******* masking *********'* ******, ***'* **. ***** laws ******* ******** ******* *****, ** I ***** **** ***** ** ***** about ****.
*** ******* ** ********* ** ******* conversations **********...**** ***** ** *****?
* ***** ********* *** *** ** read *** **** *******, *.*., ** don't ***** **** ** ****** ** be ****** ******* ****.
**** ***** **** * **** ***** technology, **** ******** **** ** *** be *********. *** *** *******, *** assumes **** "*******" ********, ** **********.
** *** **** * *** **** saying? ** *** ******* ************ ** SITE ** **** **** **** ***** to *****?
***, ******* ** ******* ******** **** this **********.
**** (********* ********) ****** *** ** pseudo-famous *** *** *** ******* ****** (primarily ** ******** **** ********) ***** controversial ******** *****. **** ** * words *** * *****.
** **** **** ** ******** *** reading **** **** ***** **** ***** to **** *** ** *** *******:
***, ** *** *** ******* ***** on *** ******* - *** ** sometimes ******** **** ******** (*** ***** clips) *** ******* *****'* ******** ******* from **** ** ****.
***** ****** ** ************* ** **** states ** *** *.*.
* ***** ** ****** ******* ** see **** ******. * ***'* ******* it ******* ** ***, ** ** honest. ***** ** * ****** **** could *** ** ** ******* ******* like "***" ** "****" ** "***" vs "*****". *** ** **** *****, it ***** ******** ** **** ******* to **** *** ***** ** *** table, ***** * ******* ** ****** common ** ******.