Ivy League Grads Present Percepta Shoplifting Detection

By Joey Walter, Published Jun 17, 2020, 12:07pm EDT (Info+)

Ivy League graduates of the University of Pennsylvania presented their startup Percepta at the May 2020 IPVM Startups show.

IPVM Image

****** **** ******:

  • * **-****** ***** **** ******** ********* IPVM *&*
  • ********** ** *** ******* *** *** new *******
  • *** ******** ******* ** ********
  • *** ******** ******* ** ********

************

*** ***** ***** ** ********'* **** presentation **** *** ******* ****, ********* Q&A:

**********

*** **** ********** ** *****:

Percepta ********

******* ** ****, ** *********** ********* ************, *** ***** ******* ******** of **** ********** ** ************ ********.

***** ** *********** **********, ***** ***** based ** **** ******** ** ******* to ****** ********** ******** ********** ****-**** ******. **** **** ***** ** "******** ****" ***** ********** ******* ** ****** behaviors.

Product ********

********'* ******** **** ******** ******* ********* in ****** ****** ******. ***** ******** analyzes **** *******, *** **** ***** based ** ** ******** ********** ******** on ** ******* *****.

*** ****** **** ** **-******* ******** gateway **** ******** *** ******** ******* to *** *****, ***** **** ***** takes **** * *******.

IPVM Image

******** **** ***** ****** *** *********** because ***** ** ********* ** *** cloud. *******, ** ** ***** ** be ******* ***** ** *** ******** upload ********* ** ********, ****** ******* that ******* **-******* **********. **** ********* that *********** ** * ******** ** their ******, ** *** ******** *** simultaneously ******* ********* ** ***********.

Ethical / **** ********** *********

*** ********* **** ******* ** ******** suspicious ******** *** *********** *** ********* as ******* ** ***** ** **** and ******** ***********. ***** ********* ******* raw ******* **** ********** ******* ***** 3D **** **********, ****** ***********, *** gaze **********.

IPVM Image

**** ***** **** ***** ***** ******** methods, ***** ******* ***** * **% true ******** **** *** *% ***** positive ****. ****, ***** "*********** **** demographic ******** ** ********, ******* ********** their ******* *** ********** ****."

Real-time ******

IPVM Image*** ******* ******** ** ***** ******* to ****** ***** ********* **** *********** is ********.

***** ************* *** **** *** ***** mobile ***, *** **** ***** ** is "****-****" **** ******** ** ***** in ***** * *******.

*******

**** *** ********* ******** **** ** release *** ********, *** ****** * $40 **** *** ****** *** *****. This ******** **/* **********, ********* ******, support *** *******, *** *-**** ** AWS *******.

**** ** ********* ******** ** ***** analytics **** **** *******, *** *******, people *** ******* ***** *********.

Key *********

***** *** * *** ******** ******* of ********'* ********:

  • ******* **********:********** ****** ******* **** ********* ******* can **** ****** ******* ***** ***** pushback ****** ** ****** ***********, ** long ** **** *** ********** ****** behaviors **** *** **********.
  • ***** ***********:**** ******* **** * *** ********** stream ** ** **** *** ******** detection ** *** *****, ** *** bandwidth ** ******.
  • **** ******** **************:***** ********* ******* ** ******** ****** installations ** ****** ******. *** ** processing ** *** *****, ********* ***'* need ** ******** ** ********* ********* appliance *** **** ********.

Key *********

*******, ***** *** **** ********* *********:

  • ** ****** ***********:*** ******** ** ******** ** *********** detection *** ******, *** **** *** include *** ************* ********* ***** ** customer **********.
  • ***** ****** ** ***********:* ***** ** ***** ******** ** investigate ********* ******* ** *** ********* footage, ** ***** ** ******* **** it *** *********** *** *** ****** behavior.
  • *** ******* / **********:******** ** * ***** *** ******* with * ********** **** ******** *** of **********. ** ** *** *********** a *******, ** ** ****** * factor ***** *******.

******* **** *********** ****Percepta *

Comments (3)

5 min setup and 6 second incident to alert time? These numbers sound like they are possible in a lab testing environment, however in the real world I find that hard to believe.

Also if a medium-large retail store decides to purchase this technology for 40 cameras, that would mean they would need roughly +40mbps upload capability for just pushing video to AWS. And that is at a pretty conservative 720p @ 10fps per camera.

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny

Hi Christopher,

In regard to resolution/fps, Percepta said the AI is trained at low FPS and resolution:

(16:47)

our AI is trained on just 15 FPS footage at a very low resolution. So it still performs very well in spite of these. In spite of this, you have sparse data. And we're also concerning pricing wise, supporting higher resolution data for higher accuracy while we're [sic] false positives.

In regard to large numbers of cameras, Percepta said it would be used in high value areas:

(12:18)

we've actually spoken with Walmart as well and they basically indicated it's not a blanket coverage of all cameras, it's high traffic items, you know, high value areas think jewelry, think electronics, cigarettes. Exactly. And even our store by store basis, it's not uniformly across the chain. It's, you know, the areas with higher theft, you're not going to put it you know, for nothing if shrink is not high.

Agree
Disagree
Informative
Unhelpful
Funny

This seems like the type of retail meta-analytics that could have some real legs if it works in practice as well as it does in the lab.

Even if real-time use cases could be tricky based on the cloud processing model, reporting could deliver individuals on a BELO list that LP could reference.

Walmart specifically isn't into confronting shoplifters actively, so the need for real-time notifications might not be critical for a lot of retailers use cases.

Agree: 1
Disagree
Informative
Unhelpful
Funny
Login to read this IPVM report.
Why do I need to log in?
IPVM conducts reporting, tutorials and software funded by subscriber's payments enabling us to offer the most independent, accurate and in-depth information.
Loading Related Reports