Deep Science - Active Threat Cloud Monitoring Startup

Published Apr 03, 2017 09:49 AM

Deep learning for $2 a day for a system that automatically detects guns and masks, alerting the police to robberies and threats?

A US startup is launching such a service, based on the work of its founder/CEO, Dr. Sean Huver [link no longer available], and previous work from projects he worked on for DARPA and the DoD. 

In this report, we share feedback from Huver and analyze the cloud-based threat detection service that his company is set to release.

Deep ******* ********

**** *******, *** ******* ******* ** **. Huver, *** **** *** ** **** he *** ** * ****** ****** that *** *** ***** *** ** current ***** ******** *********. ****** **** video ********* *********, ***** **** ******* to ******** ******, *** **** ** central ******** ** ******** ********* *** monitoring, **** ******* ** **** ******** its *** ********** ******** ** ******* and ******* ** ******.

*** ********** *** ********* ** ****** attempted ********* ** *********** ****** *** ******* businesses, ** **** ** *****-***** ********* that ***** **** *** *****.

*** ********* ***** ***** ** ******** of ****** ********* ****** **** **** Science's *********:

 

Management / *******

********* **. ***** ** *** ******* employee, ***** *** ******** ** *** R&D/development, ******* ****** *** ********* *** Vonage ***, **** ****** [**** ** ****** available], ** ****** ** ** ******* to *** *******.

*** ******* ** ********* ******, *** says **** **** ****** **** ****** from *** ** *****, ********** ** remain *********** *** ***, ****** **** not ***** *** ****** ******* ** the ******.

Threat ********* ********

********* ** *** ***, *** ********** primarily **** *** * ****** ********** to ******* ***** ******** ** ****** facial ******** ******* ****, *****, ********, or ******* *******. *** ****** *** also ****** * ****** ******* * weapon. ** ***** *** * ****** of ********** ****** ***** *** ** both ** ***** ******** *** ***, after ***** ** ***** ***** ** sent ** **** *******'* ******, ***** dedicated ***** ****** *** ***** *** take *********** ******. 

*** ********* ***** ** ******* *** currently ********:

  • ********* ******* / ****** ********** ******** or ******* * ******
  • *****-***** *********
  • ***** ****** *** **** *********

Range/Resolution ************

*** ****** ******** ******* ***** ****** on * ******'* ****, ** ** a ******, ** *** ******** **********. This ******* ** ************* **-** ***, which *** **** **** *********** *** stores **** **** ***** ********** ******* in *****, ** *** ******** **** areas **** * ****** ****** (** a ***** ** *********, **** ***** requirements, ** ** ****** ***** **** ***** ~13' ****, * ***** ****** ** to ~**'). 

Cloud ******** - ******** *********, **** **********

* **** ***** **** **** ******* is **** ** **** *******'* ***** servers, ***** *** ***** ********* *** performed. *** ******* **** ** *** work **** **** *******/**** *******, ******* the ********** ************ *** ****** ** larger ******** ********* *******, ************ ** multiple ******* *** ***** ********* **************.

**** *******'* *** ** ************ ***** ****-********** ** **** to ****** ****** ****** ** *** cameras. ***** **** ********** *** **** for ********** **-**** ********, ** *** make *** ******* **** *********** ** cyber ******* ** ******** ****** ******.

Live **********

** ***** ***** *** ********* *** monitoring, **** ******* ****** **** *** better ******** ****** *********, ** **** as ****** *** ***** ***** ****** to **** ********** ******** **** **** the ****** ** ****** ****** ***** alarms. ********* ***** *** ** ******* for ********* ** ****-******* ***** ******* directly, *** *** ******* **** **** to ***** **** ********** ** *** future.

********, ********, *** ********* * ********** facility ** * ******* *********, ***** is *** **** ******** *** ***** companies ** **** ***** ****** ** outsource **** ******** ** ********* **********. However, ***** ********* ******** *** *********** *** **** ***** ******** ***** *** monitoring ******* *** ********** ** ******* events, ** ** *********** (****** *********** *** *** ** *** best ******* ** **** ******** ****** from). *****, ***** ********** ** ***** regulated *********** ***** ******, *** ******* potential ********* ** ** ***** ** mishandled, ****** *** **** ********** ******* potentially * **** ** ***********, ****** than * *******.

Open **** **** 

********* *** ******* ** ********* ** a **** *****, **** ** ***** *** ~40 ******* ***** *********. **** **** phase ** ***** ****** ** ** a ***** ********, ***** ************ *** use *** ***** ********* *** **********, however *** ********** ****** ** **** staffed **** **** ** ********, ***. 

Pricing *** ****** ****

******* ** **** *******'* **** ** ****** ** *** *****, ****** the ******* **** **** **** ** ******** dealer ******** ** ****. ********* ** number ** ******* *** **** *******, prices ***** **** $**-$** *** ****** per *****, ** **** **** $*/*** in **** *****. *** ****** ******* the ***** ********* ** **** ** 24/7 **** ***** ********** ** ******** alarms. ***** *** ** ****** ** number ** ****** *** *****.

**** ************ ** *** ******* ** currently ********* *** **** ****.

Comments (3)
Avatar
Skip Cusack
Apr 05, 2017

This is an interesting approach. Detecting specific articles like masks and weapons obviates the need for a nice high frame rate, which is not practical. I would expect Dr. Huver has done an actuarial analysis that ends up justifying the cost per camera. Another cost is that of false alarms. Where I live, if the cops have to roll up to repeated false alarms then they start billing the originator. 

NH
Nicolas Heyman
Apr 08, 2017

Depending on the approach the detection it is likely to get better over time. There maybe false positives to begin with, but especially with active monitoring, those should get factored down fairly quickly. 

(1)
Avatar
Brian Karas
Apr 05, 2017
IPVM

Another cost is that of false alarms.

True, but they are attempting to address that by manning their own monitoring operation, so police should only be called when there are actual events, not just someone in the winter time wearing a face mask.