Lighthouse Deep Learning Camera Tested

By Rob Kilpatrick, Published Dec 07, 2017, 10:41am EST

A Silicon Valley startup, Lighthouse [link no longer available], with a Stanford PhD CTO, has released a deep learning AI camera with 3D sensors for just $300.

The company claims "False alerts, gone forever. Really" and more.

We bought one and tested their claims. Inside, we share our findings on:

  • Adult / Child / Animal Detection
  • Facial Recognition
  • Voice Search / Voice AlertSsetup
  • Home / Away Detection
  • Mobile App Usability 
  • And More

*******

*******, ********** ********* **** ** our *****, **** ******* false ****** **** ****** classification *** ****** *********** above *** ****** **** other ********/*** ***** ********* have ******* *** ****** to *******, **** ****** *********** *******

*******, ***** ****** ****** of ******* ********* ****** which ****** ** *** work (*******, *******) ** well ** ********* ****** recognitions ***** *** ****** nuisance ** ****** ******. 

*******, *** ********** *** is ****** ******* ******** to **** ***** *******, with ******* ******* *** search, ********, *** ******.

Market ******

*** ******* ****** ** Lighthouse's *********** ** ** demonstrate **** * **** can ******* ** ********* / ******** **** ******** camera. ******* ** ** a ******* ***** ** the ******** *****, ** will *** **** *** immediate ****** ** *** broader / ****** ********** market ***** ** ******* their ******** ***** ** more ****** ******.

** *** ******** ******, Lighthouse **** **** ***** technical ***************, *** ** a ***** ***** ** $300 + $*** *** year ******** ************, ** will ***** ****** ** higher ********** ****** *****. On *** ***** ****, there *** ********* ****** well ** ** *********** who *** ****** **** that *** ******* *** grow ******* ******* **** that *****.

Key ********

** *** ******* ***** **** several ******** ******* **** Lighthouse, *** ** ***** are **** ***** ******** cameras *** **** **** commercial *********:

  • ****** ********* ********:********** *** ************ **** to ************* ******* ******, children, *** ****.
  • ****** *********** ******** (**** *******):***** ****** **** ********* reliably **********, *** **** issues **** ***************** ** attempting ** ********* ******** not ****** *** ****** were *******.
  • ******* ******** ********** / voice ******: ****** *** ***** ***** using ******* ********, ********* speech ** ****, ****** well, **** *** ******* multi-criteria ********.
  • ******** **********:********** ************* *******/******** ****** based ** *****' *********, with ** ****** ************* or ****** ******* ********.

***** **** ******* ************, ** well:

  • "*******" ********* ***** ******:***** ********** ****** ***** may ****** ***/** ******* upon ******** ******* **** running, *******, ***., ** our ***** ***** ******** were *** **********, **** only "******" ******* ********.
  • *********** ********* ** ******/******** subjects:********** ********** ********* ** recognize ***** ***** **** turned **** **** *** camera ** ********* ********, resulting ** "******* ******" notifications. **** ****** ******** for ***** *** **** alerts ** ******* ******* since **** **** ** nuisance *************.
  • ********** ******************: ********* ******* **** **** rarely ********** ** ****** or ****, **** ** a ******* **** ***** classified * ****** ** a ***** ********** ** a ***. **** * few ** ***** ****** were **** ** *** tests (
  • ***** **** *******: ***** ****** *****/****** ***** directly **** ***** ** the ***'* ***** ****, meaning **** **** ******** to *** ******** ***** they **** ** ***** by ********* *** **, which *** ** *********** or ********* *** ****** older **** *** ****** past.
  • ** ******** ********:***** ** ** *** to *****, ******, ** fast ******* ***** *****. Instead, ***** ****** **** start ** ****** *** loop, ******* ** ** animated .***.

******* 

********** *** ** ********* directly **** *** ************ for ~$*** ***, ****** preorders *** ********* **** out. ********** **** ******** a $***/**** ************ ****, not ********.

**** ******* *************** ****** **** **** DIY/consumer ******, ********* similar ** ******* *** **, $*** ******.

Physical ********

********** ** ************* ****** than **** ********/*** *******, using * ********/**** **** factor ***** *" ****, shown ***** ******** ** the **** *** **:

** ****** *** ******** features ** *** ****** in **** *****:

Lighthouse ********* ** *** ** *****

********** ********* ********** **** not ********** ********* *** system's ****** ***. ***** it *****, ** ** *** as * ******* ** directly ********* ** *** app.

*** *******, **** ****** that **** ****** **** it '***** **** ******* new ** ****' *** what **** **, ** reality, ** **** ** 'unknown' ****** *** ** practice ** ***** * known ****** **** *** matched, ** ***** *****:

**** *** *** ********* claim ** '*** ****** has *******', ********** **** **** an ******* ************ *** users ***** ****** *** registered ** *** ****** but **** ************ **** not ******* * *******. However, ** *** *** walker's **** ** **********, whenever *** *** ****** is ********** ***** ************* are *******, * ************ will ** **** **** their *******.

Object ***********: *****/*****/*** ***** ********

**** *** ****** ** a **** ** *******, in ** **** ****** as **** ** * typical **** *******, **********'* object *********** ****** ****, classifying *** ************ *****/******* adults, ***** **** ********, and **** **** *** incorrect ***************.

********** ********** *** ******** object ** * ********* color ********* ** **** it **, ***** *****:

Object ************** ******

********** *******, ** *** few ****** ** *********** objects, **-** ********* *************** out ** ***+ ******.

**** ********, *** ****** merged ***** ******* **** moving ******, **** ** the ***** ***** **** to *** ******* *****. Pets **** ************ ****** with ******, ** ****.

************, ********* ****** ******* were ********** ** ******, such ** *** ******* door ***** ****:

****** ***********

********** ********** ***** ******** using ****** ***********, ***** worked ****** **** ** our *****, **** *** missed ** ***** *************** (an ******* ****** ********** as ***** ** **** versa) ** ***-*************** (*** known ****** ********** ** another).

** ******* ***********, ***** may *** ****** ***** Lighthouse ******** *** ** known ***** ** ******* or *********, ***** *****. Note **** ***** ****** do *** ******* ****** for *** ********* ******.

Occasional ****** **************

*** ******* ***** ** missed ************** *** ********** attempting ** ********* ******** which **** ****** **** from *** ****** ** otherwise ********. *** *******, in *** **** ***** the ******* ** ***** as ******* ******* *** camera ****** ******** *** his ****. ****** **** as ***** *** ********** to ***** ****** ***** in *** ******'* ***** accuracy *** *********** ******** unknown ****** ******.

Voice ******

********** ******** ******* ******** processing, ***** ****** ***** to ****** *** ****** or *** ** ****** using ******* ******* ******* of ****** ** ****** multiple ******** *** ****** through ***** ********.

*** ******* ** *** video *****, ** *** voice ****** ** ****** for ***** ** * specific ****** (***** ** facial ***********) ** * specific ********* (*********, ***-****).

*** ** **** *******, we **** ********** ** alert *** **** (***** they **** "*****") **** it **** * ******** person.

Natural ******** ***********

******* ******** ** ******* in **** ** *** show, **** ******* **** specific ******* (******, ********, kids, ****, ****, ****) and **** ******. *********/******** on ******* ***** **** running, *******, ** ****** do *** ****, *** do ******.

Gesture/Action *********** **********

** *** *****, *** only ******* *********** ***** worked ******** *** ******, shown *****:

**********'* *** **** *** waving ** *** * child ** ******* * notification ** ***** ******* who *** **** ******* / *** *** **** tot ** *****.

********* *** ******* ****** showed *** ****** *** to ********** ********** *** movement ** *******:

********** **** **** **** to ******* **** ******* over **** *** *** more ******* *****, *** did *** **** * timeline.

Reliable ********* ***/****** **********

** *******, ********** ************* armed *** ******** ****** based ** *****' ****** phone ********, **** ****** sent *** *** ********** when *** ***** *** away. *** **** ***** required *** **** ******* is ******* *** ******'* location, *** ** ***** configuration ** ****** ****** changes *** ******** ** even *********. **** ********** *** also ** **** ** searches/alerts, ** ***** *** search "***** ** ** unknown ****** **** **** is **** *** * am ***" ** *** example.

 

Test **********

*** ********* ******** ******* was **** ** **** test:

  • ******* *.*.***

Comments (13)

Lighthouse has contacted us with their feedback about future features/fixes with the Lighthouse camera:

  • Video controls is something we'll be releasing early next year.
  • We've seen some very limited example of curtains and sliders triggering alerts - we're working on a fix that should be deployed shortly.
  • Running and jumping aren't fully deployed yet. We fully expected to have those by launch but didn't quite get there. It's on our very near term roadmap to deploy.

I don’t know if it’s the frame rate or what but it looks like a bit of “Charlie Hustle” going on there...

I have been saying for years that the solution to accurate video analytics is 3D with multiple camera views. This is the right step in the right direction. 

I have been saying for years that the solution to accurate video analytics is 3D with multiple camera views.

To be clear, the additional technology used here is a single time-of-flight sensor operating with the same general FOV as the camera.

 

Any plans for a commercial version  with VMS and  PoE support?

Abdelhamid, Lighthouse response to your question:

 
 

We are evaluating a lot of opportunities and feature additions in the enterprise space. I can't comment yet on which ones will be first.

I am interesting to see the recognition in crowded areas. If only one person is walking in front of the camera it is easy .... How about to make some test of bigger room with 5 -10 people inside.

How about to make some test of bigger room with 5 -10 people inside.

No, Lighthouse is clearly and explicitly marketed and designed for home security where 5 - 10 people inside would be atypical.

That said, I do agree with you more generally that higher loads would challenge performance but such loads are not a fair application of this product.

If only one person is walking in front of the camera it is easy

I disagree there. Most products today, commercial or otherwise, have notable issues just ignoring lights, shadows, wind, etc. even if there is just one or no people.

What is the distance that the camera work properly with?

Just to make sure - This is an indoor only camera, right?

What is the distance that the camera work properly with?

In our tests, the camera would start identifying me as a person (blue halo around my body) at ~22'.

Just to make sure - This is an indoor only camera, right?

Correct, this is an indoor only camera.

Update-

 "Lighthouse Marketing vs How It Works" images changed to accurately represent a "new face" and a "known face" notification.

The original "unknown face" image was using the security notification received when an unfamiliar face is seen but no owners of the camera are considered "home". The image was replaced with a notification received when an owner of the camera sets a ping up to have the camera notify themselves of an unknown face and an app user is considered "home" by the Lighthouse. 

The original "known face" image was using a capture of an event and was updated to a capture of the push notification.

Does the Lighthouse user documentation offer any tips on camera placement?

Lighthouse recommends, for best results, placing your camera in a well-lit place with a good amount of traffic, and with no objects blocking the camera's field of view for at least 3 feet.

Seen here under "Additional Tips".

Read this IPVM report for free.

This article is part of IPVM's 6,810 reports, 914 tests and is only available to members. To get a one-time preview of our work, enter your work email to access the full article.

Already a member? Login here | Join now
Loading Related Reports