Scylla Gun Detection System Tested
With the ambition to reach a $1 billion valuation by 2024-2025, Scylla aims to eliminate the human-in-the-loop approach in visual gun detection systems with its AI-based analytics, but how well does it work?
Based on IPVM's months of testing Scylla's on-prem and cloud gun detection system, we detail:
- How did Scylla perform in detecting firearms on the scene?
- How was Scylla's performance with false alerts on innocuous objects?
- What is the alert latency of on-prem and cloud solutions?
- What PPF is required for consistent firearm detection in full light (~300 lux)?
- How does the PPF requirement change for different scenes, such as low light (~50 lux), and WDR?
- How easy or difficult is Scylla's setup?
- What is Scylla's approach to visual gun detection?
- The overview of the system's interface and its mobile app for notifications.
- Differences between the on-prem and cloud solutions.
Scylla is the first in a series of visual firearm detection tests. We have tested one other provider and will test other solutions leading up to rankings. We also examine various manufacturer live demos and claims for baseline comparisons. If you would like to be included in IPVM's visual gun detection testing, email mert@ipvm.com.
IPVM ******* ***********
**** ******* ** ****** ******* ******, where ** ****** ***** ******** ** indoor *** ******* ******, *** ** night. **** ********* ******** ** ***** across *** *****, **** *** ******* various *******.
** ****** *** ****** *** ********* accuracy ***** ****/*** *******, ********* *** prop ******* (**** *** *****), *** prop ******, *** * **** ****. All *** ******* **** *** *****, and ***** **** **** **** ****** to ******* **** ******, *** ** taped *** *** ********. ******* **** detected/ignored **** **** ** ******* **** not ****** ** ******** *******. ***** is *** ***** ********** **** **** weapons:
** ****** *** ****** *** ***** alerts ***** ******* ********* ******* *** body *****. *** ********* ******* ******** cell ******, ******, *******, *********** ******* canes, *******/*******, **** *******, *** ********. We ********* ******* **** ***** **** and ******* ***** *******, **** ****** guns, ** ******* ******* **** **** arms ******** ** ********* ***** ******.
Scylla *****
***** ****** **-******* ****** *** ****** RTSP ******* ** ****** ********** *** frame ****, *** ***** ******* ******** port-forwarding ****** **** *******, **** ****** telling ** *** ****-********** ** **** used *** **** *******, *** ******* VPNs *** *** ** *** **** cloud ***********.
*** ***** ***** ***** *** ****** setup:
*******
****** ****** ******* ********* ** ********* as * ***** ************ *** **-**** deployment. *** ***** ************ ** $** per ****** *** *****, *** **-**** is $*** *** ******* *** ****, with ** ********** ****** **** *******.
*** *****-***** ***********, ** ***** * monthly ************ *****, ***** **-******** *********** are ********* ******* ** ****** ************. The **** *** **-******** ********** ** $250 *** ******* *** ****, *** for ***** ***********, ** ** $** per ****** *** *****. ** ******* exclusive ******* ******* *** ***** ********** Centers, **** ** ***** *****, ******* to ***-********** ********** (***), ***** *** significantly **** ****-********* **** *** ******** licensing ****. *** ******* *********** *** on-premises ********** ** **** ********, ******* cloud *********** **** ** ******* ******* requirement. ************, *** ******** ******* **** K12 ************, ** ***** ********* ********** pricing ***** *** **** ************ *******, reflecting *** ********** ** ********** *********** environments.
*** **-**** ******* ********* **** ****** come **** ***-**** ********, *** *** MSRP ****** ** $*,*** *** **** channels.
**** * ******** ********** *********, ** ******** ********, ********* ******** without *** ********** ** ****** **** for *** ***** *** *****. **** purchase **** *********** *** ***** ** updates *** *******. *** **** *** a *-******* ****** ** $*,***, ***-*********, granting **** ********* ** *** ****** to *** ********. *** ************ ******** with *** ************, ** ***** ************* discounted ******* ******* *** **** ************ program, ************* *** ********** ** ********** and ********** *********** ********** *********.
Three-Layer ********: ******* *********, ********, *** ************
****** **** **** **** *** * three-layer ******** *** ****** ******* *********. They ********* ****** ******** ***** ***** custom ****** ********* *********, ************* ******* the *******, *** **** ****** ***** their "****** *********," ** **-************ *****.
****** ************* ****** **** *** ********** use ** ****; *** ****** ** built ** * ****** *****-**** ************. The ********** ** *** ******-********* ****** detection ******* (****** ***), ******** ** * ***** ** statistical ******** ********* *** *****, *** just ****** ******. *** ***** ********* is *** ****** *********, ** **-***** decision-making ****** ******** *** *********** ** our **********.
Executive *******
** ***** ****** ********* ******** ** detecting ********** ******** ** *** ***** and *** ****** ********* ** ***** alerts, ********* ** ******** *********** **** one ***** ********** ****** *** *** others ******. *********** *** ******* ****** on-premise *** ***** ********, ********* ******* with ******** ** ******** ** ******** firearms. ****, ****** *** ******* **** various ********* ******* (**** ******, ******* canes, ***.) ** ****** *** ******* scenes, ****** ********* ********* ** ***** alerts, ***** *** **** * ****** limitation ** ****** ******* *********.
****** ************ ******** ********** ******** (*** holstered - ********) ** ****'* ******* scene ** **** ***** (~*** ***) when ~**+ *** *** ********* *** long **** *** ~**+ *** *** pistols. ** *** ***** ********** (~** lux) *** *** ******, ****** ******** higher *** *** ********** ********* (~** for **** ****, ~** *** *******). Scylla's **-**** ****** ******** ****** *************** (~1 ******), *** *** ***** ****** took ~* ******* ** ******** ******.
Strong ******* ********* ***********
****** ****** ******* ********* ****** *** a ****** *********** ** ****'* *******, consistently ********* *** ******** ** ******* and **** ****. ****** ******** ** walks ********* **** ******** **** ** various *********, ****** ************ ******** **** when ~**+ *** *** ********* *** long **** *** ~**+ *** *** pistols.
****** ******* ** ******* *** **** guns **** ********, ******* ** * sling, ** **********, **** *** ******** visible ** *** ******* *** ******** PPF.
**** ******* ***** **** ****** **** not ******* ****** ********* *** ********** firearm *********, ******** ** **** ********** behind *** *****.
Virtually ** ***** ******
****** ****** ** ******* ** ****** and ******* ******, ****** ********* ********* no ***** ******. **** ******* ******** hand ********, ****** ******, **** *******, cordless ******, *** *********** *****, ***** have ************ ********* ***** ****** ** visual *** ********* *******.
****** **** **** **** **** ****** no ***** ****** ** **** ******* scenes *** *.* ***** ****** *** camera *** *** ** **** ******* scenes.
*********, *** ****** ********* *.* ***** alert *** ****** *** ***. ** environments **** *** *** *******, *** can ********** ********* ** ***** ******. However, ** **** ******* ********, *** expectation ******* ******** ** ** ******* of *.* ***** ****** *** ****** per ***.
Low ***** *******
****** ****'* *******, *** ****** ****** had *** ***** *******, **** *** on-prem ****** ********** ****** *************** (~* second) *** *** ***** ****** ** ~4 *******. ****** ********* ******** ****** for *** ******** ** *** ***** and, ******** *** "***** ******" ******* grouped ****** ****** * ******* ****** (i.e., ** *******) **** * ****** notification.
** *** *** ****** ****** *** with *** **-**** ********, *** **-**** system ***** ** ** ******* **** the ***** ******. ***** *** ******* push ************* ** ***** ****** **** the **-**** *** ***** ******* *** connected. ***** ****** *** ** ******* latency ** ~** ******* *** ******** the ****** ****, ********* *****, *** a ***** ***** ****.
~30+ *** *** **** ****, ~**+ *** *** *******
** **** *******, *** ****** ****** required ~** *** *** ********** **** gun ********* *** ~** *** *** pistols. *******, *** *** ************ *** provided *** **** *** **** ******* of *** ******** ** *******. **** the ******** *** ******* ******** ** the ****** (***** *******), *** *** requirements ******** ********* ** *** ******'* down-tilt *****.
***** ******'* ******* *** *********** ** lower **** *** *******, ****** **** IPVM **** ** ****** **** ***** expectations *** ** ******-**-****** **********, *** they ********* ~** *** *** **** guns. ******'* ***** ******* *** *** requirements ** *******:
PPF ******** ********* ** *** ***** *** ***
*** *** ******** *** ********** ********* in *** ***** (~** ***) ********* by ~**% *** **** **** **** and *******. *******, **** * ****** PPF *** *********, ****** *** ****** detection ******** *** ***** ***** **********, similar ** *** ****-***** (~*** ***) performance.
** *** ****** (**** *** *****), Scylla ******** ****** *** *** ********** detection.
No ***** ** *** ****
**** ******* ******** **** *** ****** system **** *** *** * ***** operator ** ****** ****** ** ******* the ****** ** ** ***-****** *************.
Image ******* ******* ** ***********
*** ****** ****** ***** **** ******* cameras ** ********* *****/***** ******* ******, providing ********** *** *********. *******, *** optimal *** ** *** ****** ******, it's *********** ** *** *** ******* quality ******* ******** *** *** ******** scenes. **** ******* ********* **** ******* with ******** *** *** ***-***** ************ had **** *********** ********* ** *** requirements ****** ******* ******, ******* ********** of ***** ******* ** ********* ********** analytics *********.
Version ******
****** ******** *****: **.**.*
**#*, ****** *** *** *******. ** tested **** ******** **** ****** (**** in ***** *** ****), *** *** system *** *** ******** *** ***** positives.
*** *** **** ** ** *** detect ** ******* ***** *** *** sideways?
**** *** ****** *** ******** ***?
*** **** *** ****** ********* ******* the ****** **** ** ** ********? Say *** **** * ******, *** a ******* ****** **** ****. *******: We **** * **** *** ** replace *** ******, *** *** ** have * *** ****** ********* ** the ******. *** ** *** **** to **** **** *** ****** ** working ******* ****** ** **** * gun *** ** * ****** ** make **** *** ****** ** *******?
**#*, ***** *** *** *** *******.
*** *** **** ** ** *** detect ** ******* ***** *** *** sideways?
** *** *******, ****** ************ ******** guns **** ********:
**** *** ****** *** ******** ***?
** **** ****** *** ****** **** the **** ******** ***** ** *** report. *******, **** ****** *** ******** stock, *** ****** ******* ** ***** firearms **** *** ***** *** ****** when ******** *** *** *********.
*** **** *** ****** ********* ******* the ****** **** ** ** ********? Say *** **** * ******, *** a ******* ****** **** ****. *******: We **** * **** *** ** replace *** ******, *** *** ** have * *** ****** ********* ** the ******. *** ** *** **** to **** **** *** ****** ** working ******* ****** ** **** * gun *** ** * ****** ** make **** *** ****** ** *******?
** *** ******** ***, ****** ***** deployments *** ** *** **** ******* of ******* *** ******, **** *** last ******** *****:
** **** ********* *** ******** ** the ****** **** *** **** ****** you **** ***** ********.
*** **** ** ** *********** ******** for *** *****-***** *** *********. ** it's ****** *********, **'* **** ****** to *** * ****** ** ***** of *** ****** ** ******* ********* with ******* ***** ** *****, ***. But **** ****, *** *********** ** that **** ****** ** ***** **** yet **** ******** *****, *********** *** accurately ******* ****.
** *** **** ***** ** **** safety ****. ** ** *** ***** about **** **** * **** ***** system, ** ***** ** ** ****** and ********* ***** ****. * ***** like ** ********** **** ************ ***** about ****.
**** ** * ******** *****. * was ******* ***** * ******** ***** like *** *&* ***.
* ***** **** ** *** ** it *** ** ******** ****** ** when ** *****'* **** **** * rifle.
****** *** *** *************. ** **** not ****** **** ******** ******. ** focused ** ****** ******** **** ***** be **** *** ******* **** ***** be ***** ** **** **** **********.
*******, ****** *** ********** ********* *********** with ******* **** ** ****.
* **** ****** **** **** ******* other *********, *** ** **** *** detect **** ******. ***, *** ****** can *** ** ***** **** ******, but *** ***** ***** *** ******* with ** ******.
****, *** ** ***** ******* *** easily ******** ** ******* *** *** over *** ***.
**** ** ******'* ******** ** *** deployment ********:
** **** **** ****** ***** ** do ** * ****** ********* ** install * ******, **'* **** * hidden ******** *************. **** *** ** a ********** ** **** ******** ***** background ****** **** ****** *****, ** no, ** ***** ************. ************ ** the ********* *** * ***** ** red ***** ****, **** *** ***** on ** ** ***** ****** ** cameras *** *******
**** ** *** ********** ** *** Health ***** *** **** ****'* ******* dashboard:
** **** **** * ***** ** the ********** **** *** *******? * know **** ******* *** **** ****** up ** ** *** ******.
** *****,
** *** *******, ** **** ***, 5MP, *** *** ******* *** ******** that ****** ********* ******* ** ****** RTSP ********** *** ***** ****.
*******, ** **** *** ****** ******* with **** **** *** **********. *** example, ** ************ ******** * ****** ****** ******, *** ********* ****** ******* ***** 8MP ******* *** ****-**** ************.
** *** **** *** ***** ******* in **** **** *** ***** *** with * ****** ******* ********* ******?
** **** **** **** ******* **** H4 **** *** ******* **** ********. We **** ******* ** *** ********** in *.***. * ****'* **** ** this ****** *** *** **** ***********.
** ** ***** ** ****** ***** its *** *******?
**#*, **** ** ******'* *****:
*) *** ****** ****** ******** ******?
** ** ***** **** * ***** should ****** ** *******, ** ****** on ****. ******** - ***** **** matters. ****** ***** ***** **** **** features ** *** ******** *** *** that **** *******. ** *** **** set ** **** ******** ****** **** parts *** ******** ****.
*) *** ****** ***** *** *** dataset?
*** ** **** **** *** **** customers, ********* ****, *** **** ** scrapped ********* **** ***. ******** ***** improvements *** **** ** * ********* with **** **** **** *********.
*** *** ******** *** ********** ********* in *** ***** (~** ***) ********* by ~**% *** **** **** **** and *******.
*** **** *** ****** ******* ** nighttime/IR *********** *****? ** *** ** moderately ******** **** * ***** ****** for **************, ****** ***** ** ********** is ***** ****.
* ***** ********* ** * **** critical *********** ****** ******** ** ********** schools, ********** ****** *** *** **** they *** **** *********. *******, ***** are ********* ************ ***** **** ***** be * ******* (***** ********, ********, bars, ****** ****** ************).
**#*, ****** *** *** *******.
****** ************* ******** ******** ******* *** ~50 ***, ********* ** **** ******* our ******* ** ***** *************.
** ****** ******** *******, ** ******** ******* ******* *** casino ****** ** *** ~** *** range *** ******* *** ******* ** those ********* *** ****** ******* *********. Some ****** **** ***** *** ******** include **** (~** ***) *** ************* (~7-10 ***).
*******, ***** *** ********* ************ ***** this ***** ** * ******* (***** theaters, ********, ****, ****** ****** ************).
** ***** *********, ** ***** **** consistent ******* *********, ********* ************ ********* to ******** *********** **** ** **********, increasing ************ ** ***** ~** ***.
No ***** ** *** ****
**** ******* ******** **** *** ****** system **** *** *** * ***** operator ** ****** ******
technology **** ******** **** ** ********. HiTL is not a strength - it is an overt weakness.
** ** *** ********** ** ** that ****** - *** *******-********** *** detection ******** *'** **** **** ***** in ** **** ***** - ** the ***** ** ***** ** ********************* **** *** ****, *** *******, stating **** ***** ***** *****.** **'* ***. **** ** ***** ************ ********.
** ******* ** **** ******* ****, there *** ****** ** **** **** space ******* **** ** **** **** of ******** ***** - ************ **** it ** * ***** ********* **** their **** ***'* ***** *** ***** time *** **** ***** ************** ***** *********** ************.
******* ** ****** ******* ********* ******** HiTL. ***** ** ******* ** *** Lambs ** *** *** ********** *******... but ***** **** ****, **** ***** be ******* ** ** *** **** of ******** *****. ******* ** **** aint.
** ***'* ******* ** ****** ********* analytics **** *** ********* **** ****** humans ** ******** ***** * ********* zone. *** ***** ********* ******** ** imagining **** **** ** ******** ** make **** ****?** ****** ***- ******* ***** ********* **** ******* by ***.they **** ******* ****** ****** ** ****** **** **** *** *******.
******? **** - ***** ********* *** sometimes *****.
*** ** *** **** ***** ********* occur, **'* ******** - ******* ** ruins *** *** ***** ** ****** these ********* ** *** ***** *****.
**** **** *** **** ***** ********* using * ***-**** ******** **** ****** would ** ******** ** ** ********* similarly.
*** ***, * ***'* *********** (** *******) ********* ******** *** ever **** **** ** ******** *********** any **** ** **** ********* ***** at ***...
**** ** *** ***** ******* *** detection **** ** ****, **?
**#*, **** ** *******. **** ** the ***** ******* *** ********* ****. We **** ********** ****** ***** ** these ******* *** ******* ********* ******* (***** ******* / ****** ********* ******** Tested).
** ****** ****** ********* ******* *** detection ***********, ***** ****** **** ** released ***** **** *****.
>*** ***** ************ ** $** *** camera *** *****, *** **-**** ** $250 *** ******* > *** ****, with ** ********** ****** **** *******.
** ** *** **** ***** *** a ******* ****** ***** ** ****? Is ** ****** $*** *** ******* per **** *** * ****** **, say, ** *******?
**#*, ***** **** *** ***** ***** by ******. *** *********/*******, * ********* reaching *** ********.
* ** ******* ** *** **** tested **** ******** ***** ******* * black ***** ** *** ** **** can ******* ***** *********