Axis Live Privacy Shield AI Analytics Tested
While Axis Live Privacy Shield worked poorly in our 2019 test and was expensive, now Axis has added AI-based masking and made it free, but how well does it work?
We tested Live Privacy Shield in multiple scenarios to see how it performs, including:
- How reliably does it mask people and faces indoors?
- How reliably does it mask license plates, people, and faces outdoors?
- Does it accurately mask subjects in groups and/or crossing paths?
- How do facemasks, sunglasses, and hats affect masking performance?
- Does it produce nuisance masks on unintended objects?
- How does it handle non-license plate text like vehicle lettering?
- How does it handle human subjects in vehicles?
- How easy or difficult is it to configure?
- What Axis cameras does it work with?
Configuration ********
************* ** **** ******* ****** ** simple *** ********* **** ******** ***** to ****** ** *** ****** **** they **** ** **** (******, *****, and/or ******* ******). ***** *** **** additional ******** *** *********** *** **** type, ***** *** ** ******** ** desired, ***** ** ***** ** **** detail ** *** ******** *******.
No ******* ********/********* ** **** ******-* ******
** ********* ** ******** *** **** Privacy ******. ** ********** **** $*** USD *** *******.
**** ******* ******'* **-***** ******* *********** ** ******* **** ****** **** DLPU, ********* *, *, *** * series *******. ******-***** ******* ** ********* on ***** ******.
Executive *******
***** **** **** ******* ****** ****** well ******* *** *** ** *******/******* subjects *** ******* ******, ********* **** masks, ** ******, ** ********* **********, it ******** ******** ****** ** ********, revealing ****** *******
************, ***** **** ****** * ****** unmasked ****** ** ******** ******* *** required, ** **** ******* ****** "*****" masks *********** **** *** ******'* ***** streams *** **** ****** ** *******.
*******, **** **** *** ********* **** applications ***** **** **** ** **** simultaneously **** **** ******* ******, ********* their**-***** **** ****** *********** ***** ********** **** *********.
Technical *******
**** **** ******* ****** ****** **** in ****** ****** ** *** ****, masking ****** *** ***** ********** ** 25-30 ***, ********* ** ******, ** low ***** *** **, **** ********* obstructed, **** ******* **** *** **********. However, ******* ***** **** ************ ******* when ******** **** **** ***** ***** ~30 ***, ********* ***** *******.
********, ***** *** ******** ********** ****** license ****** *** ******* ********, ** consistently ****** ** **** ***** ** subjects ******* ** ****** ** ****, revealing ****** ******* ** **** *****.
************, ***** ******* ********, ********* ******* were ********** ****** ** ****** ** faces *** ******* ********* ** ***** text ****** ** * ******* *****. Lowering *********** ******* ** ********** ***** issues ******* ********* ******* *********** ** our *****.
*******, **** ******* ****** *********** "*****" masks **** *** ******'* ***** ******, so ***** ****** ** ******* *****. Those ********* ******** ***** *** ******** use **** *** * ********* ******, effectively ******** ********* *** ******* ************.
Axis ****
** ***** **** **** *********** **** were ****** ***** **** ******* ****** addressed *** **** ****:
** ****** ** *** ********* ****** strong ******* *** ******* ******** **** (General **** ********** **********) *** **** from ***** *** ******** *********** *** unions ********** ***** ************ *** *******. Even ****** ***** *** *** **** regulations ** *** **, ***** ** examples **** ** ***** (******* *********** Security ********** ***) *** ** *** definitely ****** ** ********* ********* ** these ********* **** **** **.
Motion ***** ******* *********** **** ******
**** **** ******* ****** ****** ***** to *** ******-***** ******* ******* ** AI. *******, ** *** *******, ******-***** masking *********** ** ******* ******* ** our******** **** **** ** **** **** Privacy ******, ******* ********* ******* ********** *** masking *** ****** ***** **** ****** are ****** ** ** ***, ***** here:
**** ******* **** ** *** *********** when ** ** ********* ** ** areas **** ******** ********. **** ***** in *** ** **** **** **** mask *** ****** ******* *** ** designed *** ****** ****** **** **** lighting:
Accurate ******/**** ******* *******
*******, **** **** ******* ****** ********** masked ******** ******* ** ******* ******* the *****. ***** *****, *** **** subject ** ****** ** ** ***** across *** *****:
******* *** ******** ** ******, ** well, ********* **** ******** ********* *** diverged, ***** *****:
** **** ** ** *** ***** (~1 ***, *****) *** **** **** IR (
Slight *********** ********* **** *********
******* *********** *** ******* ******** **** subjects **** **** *****, **** ******* becoming ************ ****** ~** ***.
*******, ** ****** *****/****** ***, ******* masks **** ******** **** **** ******** items **** ****, **** ** * mask, **********, *** ***, ***** ****:
Outdoor ******* ***** *** ****** ******* ********
********, ******* ***** ******* *** *** miss *** ****** ** *** *******.
*******, ** ********** ****** ***** ******** of *** ***, **** ** *** Subaru ****** ** *** ***'* ****** below ** *** ****** ** *** car ** *** ***** *****:
**** **** **** ** * ***** issue **** **** ** ******* *** suggested ********* ***********:
*** ** *** ***** ** ****. The ****** ** **** *** ************ for ******* ****** *** ** ********. In **** ***** *** **** ** the **** ** **** ******* ** the *********** ********* ** ******** ** can ****.
******** *********** ******* *** **** ** the ****** **** *** *** *** fully ****** *** *****.
People ** ******** ********
******** ** ******** **** ********** ******** when ******* *** ******, ***** *****.
**** ******** *** *** ***** ** relatively **** **** (~** *** *****):
***** **** ************** ****** **** ** vehicles, ** ****** ***** *** **** direct ******, ****** **** ****** *** be ****** **** *** *******.
***** ** *****, *** ******'* **** was ******** **** **** *** *** consistently ****** **** *** *****'* ******** light *** **.
**** **** **** ******* ** ********* a *** ***** ** ******* **** issue:
** **** ** ********* * ******* class ** **** ** ******** *** privacy *** ******* ****** *** ******* even *******. ** **** ************ **** it ** *********** *** ***** ******** be ******* ** ********** ***********.
********** *********** *** *** ******* ********** masking ** ******* *** ********* ******** masking, ********* *****.
Nuisance ******* ******
***** ******* *********** ********, ***** **** several ***** ******* ****** ******* *** outdoors.
*** *******, ********, ****** ******** **** sometimes ******, ***** *****. ** ***** examples, **** *** ******* ***** ******* was *******, *** **** ******.
************, ****** **** ****** ** * face:
******* ** ***** ****** ***** ** fully ********** ** ******** ***********, ****** frequency *** *********.
*******, ****** ******* ************ **** ****** as ****** ** *****, ***** *****:
************, * ***** ******* ** *** floor *** ****** ** *** ***** when ****** *** **** ******* **** enabled.
***** ***** ***** ******** *****, **** said:
** **** ************ ******** * ***** sensitive ******* ******* ** **** **** we ****** **** * *** **** false ********** ** *** ** ***** missing *******. ***** **** *********** **** not ******** *** ******, ** **** cases * **** ****** **** ****** this (****** **** ********* *********). ******* it ** *** ****** ***** ******** to ****** *********** ** **** ** less ********* ** ** ** *** disturbing *** *** ****** ******* ** the ***** ******.
***** ****** **** ********** ** ******** sensitivity **** *** ******* (**%) ** ~25%.
Dual ****** ********* ******** *** *********
**** ******* ****** ***** ******* ***** directly **** *** ******'* ******* ***** stream, *** * *** ** ***** software ****** ***** ****** **. *******, those ******* ** **** ** ******** stream **** ****** * ********* ****** (called * **** **** ** **** cameras). **** ********* ****:
*** **** ***** *** ****** *** stream ******* *******. ********** *** * different ****** ***** ** **** *** for **** ******. ** ******* *** streams **** *** ****** ** ******* the ******* ****.
** *** *******, ***** **** ***** were ********** **** ******* *** ********* by *******, ********* ** ********** ******* attached ** *** ******, *** ***** not ** **** **** ******** ******* Center, *****, ** ** *******.
*** *******, *** ***** ***** ***** the ****** ******* ****** ** *** left, **** ** *** ******** **** Area ** *** ***** (*** ****** background ** ******* *** ****** ** this *******).
***** *** *** *********** ********* ** this ******:
- ******* ********* *** *******:***** *** ******* *********** ******** ****** the ********* *** ******* *****, ***** may ****** ********* ***** ************ **** using **** ******* ****** ** ******** systems *** ******** ***** ** *** deployments.
- ** ****** ******:***** ** ** ****** *** ********** to ****** ******* ****** *** ******** streams. ***** ******* **** ** ******** switch ** *** ******** **** **** starting * ******** ******.
Masked ****** ********
****** ******* ***** ***** ** **** face **** ** **** ********.
******* ******* ******* ******* ** **** humans, *****, ***/** ******* ****** ** any ***********.
Mask ***** *****
*** ******* ******* ***** *** **** Privacy ****** ** * ********* ******, shown ** *** ******** *****. *** application *** **** ** ******** ** a ***** ***** ****, ***** ****:
**** **** ***** ****** ******* **** obtrusive *** *** **** * *****'* attention ****** **** ****** *****.
Inverse *******
************, *** *** ******* ** **** Privacy ****** ****** ***** ** ****** the ****, **** ******** ******** *** the ****** ********** ********, ***** ****:
Versions ****
*** ********* ******** *** ******** ******** were **** ** *******:
- *** *******: **.*.**
- **** **** ******* ******: *.*.*
** ******* ** *** ** ********* and **** ********* *** ****** ******* live ** *** ***** **** ******* faces *** **** *** ***-******* ******* inside *** ******.
*** ********** ****** ** *** ****** but ***** *** ******* ********* **** prefer **.
******* ********* ** ******* ** *** US, ****** ****** ** *****.