Camio Natural Language Processing Tested

By: Ethan Ace, Published on Sep 27, 2016

The ex-Googler led team from Camio has advanced its video monitoring offering to include natural language processing.

Camio ingests video, automatically determines what objects are, labels those objects and then allows users to search for them with words, like Google but for surveillance video.

Here is an example of the type of labeling they claim to do:

We tested Camio's performance connected to an Axis camera and a Nest cam using a Camio Box we bought. 

*** **-******* *** **** from******** ******** *** ***** monitoring ******** ** ************** ******** **********.

***** ******* *****, ************* determines what ******* ***, ****** those ******* *** **** allows ***** ** ****** for **** **** *****, like ****** *** *** surveillance *****.

**** ** ** ******* of *** **** ** labeling **** ***** ** do:

** ****** *****'* *********** connected ** ** **** camera *** * **** cam ***** ****** ***** ******. 

[***************]

*******

*****'* ****** / ************** capabilities *** *** ****** anything ** **** **** seen ** ***** ************. This **** *** **** it ** *** '****' (as ********* *** ** done ** **** **** including *** ****, ** license *****, ** **** matching, ** ****** **********, etc. ***** ***** **** not **). *******, *** Google **** **** ****** input *** *** ****** detailed ******** ** ******, colors, ****, ******* ***** and ************* ******* **** lets ** ** ****** not ******** **** ** enterprise *****.

*******, *****'* ********* ** home ***** ** ***********, since *** ******* **** users ** ******** ** get *** **** ******** and **** ****** ** pay *** $**.** *** month *** ****** *** these ******** *********. ** contrast, ** ***** *** Camio's ********** ** ****** attractive / ** ********* differentiator *** ** ********** video ************ ********.

Key ********

**** *** *** *** findings **** **** ****:

  • *******, ****** ******** ****** **** in *** *****, ****** with **** ********* ** mislabeling *** **** ******* not ******* ** ***.
  • ******** **** ******* ** most ******, **** ********* such ** ***, ******* ***, sport ******* *******, *****, minivan, ***. ***** ********** are ********* ********, ****** some ******* ** ***************** occurs, **** **** ******** labels **** ********.
  • ***** ******** ** *** some ******* ******* ** human ********, **** ** bags ** "*******", *** these ****** **** ********** in *** *****.
  • **** ****** *******, **** ** lawnmowers ** *****, **** not ******** ** *** labels ***** ***** *** learned. ***** **** ***** may ** ******* ** more ******** ****** *********.
  • **** ****** **** *** practically ******, **** ** many ******* ***** ***** labeled "*****" ** "****", and ****** ***** ********** "ceiling." ***** ******* ****** in *** ****, *** are *** ****** ** be ****** ** *** to *** ***.

*** ******** ********* ** Camio (********, ******, *****, direction) *** *********** *********, reviewed ***** ******** **** *** *** ** ******** with *** *** **** advanced *************** ********* *****.

*******

*****'* *** ****, ******** *** ******** analytics, ** $**.** *** per ***** *** ******. Their ***** ***** ****, which **** *** ******** analytics/labeling, **** ***** ******/******, is $*.** *** *****.

***** **** *** ***** any **** ***** *** live ********* ** *** and ******* ****** *******, with ** ******* *** search, ** *********. 

Camio ******* ******** ********** ********

*****'* ********* ******* ***** using ******* ******** **********, which ******* ****** ** objects ** *** ***** for ******** *** ******, in **** **** ****** than ******* *********. **** labeling ** *********, ********* completely ** *******, *** human *********.

*** *******, ***** **** analytics ********* * ******* entering * ***** ****, Camio applies **** ******** ******, such ** "***** ******* vehicle", "****", "***********", ***., not **** ** *** moving ******, *** **** to ********** ******* ** the *****, **** ** grass, *****, ** *******. These ****** *** ** searched *****, ** ****** on ******** ******* ****.

**** ***** ******* *****'* labeling *** *** ** may ** ********/****:

****** ********

***** ********* ******* ****** ** all ****** ******, *** and *****, ****** *** outdoor. ** *** ** instance ** ***-***** ******* being ******* ** ******.

Vehicle ***** **************

******* ** ****** *** generic **** "********", ***** may ****** *** ******** types ** ***, **** as "***** ******* *******" or "*******" ** "****** car." More ******** **********, **** as "******* ***** ******* vehicle" ** "**** **** car" *** **** ********.

** *** *****, ***** searches **** ********* ******** (***** below), ****** ***** **** on *** **** ** caution, *** ********* ********** an *** ** * minivan, * ***** ** an ***, ***. ** found **** *** **** specific **********, **** ** "compact ***" ** "**** size ***" ** "******* sport ******* *******" **** less ******** **** *** more ******* *****.

Object *********

** ******** ** ******** objects, ***** ****** ***** to ****** ** ***** and ********* (***********/*********). ** users *** ****** *** "cars ***********" ** "****** wearing ****" (*****).

**** **** ***********/********* ***** to *******' ******** ******** ** *** camera, *** *** ******* motion ******** ** **** other. ** "****** *********** desk" ********* ******* ****** in ***** * ****** approaches *** ****** **** **** ** *******, not ****** ************ *********** a ****.

***** **** **** ****** to ******* *** ***** modifiers ******* ** *********** with **** ***** ** make ********* **** *********, and ****** *** **** of ********* **** ****.

Mislabeling ******* *******

** *** *****, ***** labeled ******* ******* ** human ******** **** * handful ** *****, **** of ***** **** ********.

*** *******, *** ******* carries * ********/***** ** the ***** *****, ***** was ******* ** "***" by *****. ****** ******* ** its ****/***** *** *** way ** ** *******, not ********.

 

***** **** ********* ** apply *** ******* ***** "weapon" ** * ******* carrying * ***** *****. Camio **** **** **** is ****** *** ** the ****/***** ** *** object *** *** *******'* posture, ****** ** **** similar ** * *****.

Objects *** *******

** ******* *********, ****** objects **** *** *******. For *******, ****** *** tests, ***** **** ***** clips ** *********** ****** grass, **** ** * riding ***** *** **** behind. ***** ******* ***** both ** "********", *** with ** **** ******** data, *.*., * "*********" tag.

Some Labels *** *********

*******, **** ** *** labels ******* ** ***** were ****** *** *********** useful ** **** *****. For ********, "****" *** one ** *** ********* labels, *** ********* ** simply ***** ***** ** which **** *** *******. Note **** **** ** not * ******** **** labeled ****, *** ************* tagged.

** *** ****** ******, many ***** **** ******* "ceiling", ****** *********** ***** clip *** ******* ******* in **, ** *** camera ** ** *******/******** mode ** * ********** small ****.

Camio ***

***** ********** ******** ******* setup ** ******* ******* to ***** *******, ********* by ********* ***** ***** or ********* **** *****, with ** ****** ** live *****.

*******, **** **** ********** this ******* **** *** "***** ***", * ***** **** factor ****** ***** ***** the ***** ******* *** cameras *** ******* ***** ** the ***** ******* ****** setup. There *** *** ****** of *** *********:

  • ***:$** ******, ******* ** to ***** ** *******
  • *** ***:$*** ******, ******* ** to ******* ** *******

Comments (16)

I was just saying to my coworkers at ASIS about the next generation of video analytics being machine learning and using human feedback to "teach" the analytics, much like Google's Picassa or Apple's Photos does with faces.

The next day I came across Iron Yun at the rear of the show and was blown away by the capability and how it performed at identifying human subjects stacked deeply in the FOV using web cam video live.

I would be interested to see how the two solutions compare, and more importantly, I'm keeping my eyes open on this type of analytics since machine learning in general is finding its way into many aspects of our lives, like automated assistants. The security applications are going to be interesting...

That is a very interesting technology. I wonder how long until we see it integrated into a normal VMS, not as an add on like BriefCam or other analytic companies.

Less than 1 year. I'm CEO of Camio and we're enabling cameras, NVRs and VMSs to embed this technology seamlessly with non-restrictive licenses, white-labeled services, and an extensible video processing pipeline. That way, manufacturers leap frog the low-level grunge-work to run Machine Learning at scale and at low cost without losing their ability to differentiate with unique video processing capabilities and service levels.

Thats fantastic. Who we talk to about this? About how much CPU per camera does it take for a 1080P 7 frames per second H.264 stream? Does this have a large database that is required to run alongside it?

Great, please contact me or partners@camio.com.

For CPU benchmarking, the 1.5Ghz cortex-A53 cores with 2Gbyte DDR3 SDRAM easily handles three 1080p video streams at 20 fps.

The local firmware doesn't have any large database. Instead, it extracts features from the video event locally in a chatty "metadata-first" exchange with the cloud to learn and rank what's interesting without needing to upload video.

Hello Carter.

I'm interested in this solution. We are a brazilian installer and have a lot of projects in courthouses, universities and hospitals.

Is this sollution deigned to enterprise scenarios with hundreds of cameras?

How can we get a demo license?

Hi Rafael - Yes, hundreds of cameras is no problem.

If a large deployment has limited uplink bandwidth available, then you can leave the video itself on premise using metadata mode; but you still need at least 30 kbps/camera * num_cameras uplink available (e.g. 120 cameras * 30kbps/camera = 3.6 Mbps up). Yes, contact us when your video streams are connected and we'll send you a code for your evaluation.

Very interesting product. Will keep my eye on this for sure. While many of our customers fear the cloud, I do believe something like this may start to tip the bar.

I do agree, quite heavily, that this is misplaced in the home market. If this were scaleable, and at the very least capable of processing in an Enterprise data center with only the metadata being sent to the cloud the potential for uptake in large enterprises is huge. I cannot tell you how many customer sites I see with one or two guards addressing elevator alerts, fire alarm alerts, door monitoring systems, card access alerts, key management, and answering random phone calls all while being expected to watch several hundred or thousands of cameras. Analytics alone isn't quite at the point it needs to be.

I can see placing something like HP Moonshot in a datacenter and sending the data out to the cloud being more acceptable.

I will definitely keep an eye on this company in the future. Thank you to IPVM for updating this as I missed the initial review.

Yes, you control the metadata and media that gets uploaded to the cloud at both the account and camera level. For example, you can leave all video files on the local network but upload only lo-res thumbnails for fast browsing in the cloud.

Regarding the problem of overloaded guards missing things, there are two aids:

  1. alerts triggered by persistent searches (e.g. alert for [people approaching side alley] delivered to the guard's phone)
  2. very fast browsing for human review (e.g. this browser page shows all people this morning on one screen and includes histogram of frequency over time at bottom)
    fast browse for human review

We'll look at HP Moonshot too. Thanks for the pointer.

the problem with home users is that they tend to have low cost equipments which do not produce metadata, and uploading 4-8 HD cameras to the cloud would take too much bandwidth.

4, that's the point / role of the Camio Box that we mentioned in the report.

These examples might help clarify how Camio works with any camera/NVR/DVR that provides its RTSP H.264 video stream.

Camio has open-sourced that rtsp repo under the Apache License to create a shared resource for everyone that needs programmatic access to well-tested RTSP urls across camera makes and models. There's a README with instructions to add your own cameras/NVRs.

I wonder if something like this could extend to internal loss prevention through sophisticated analysis of retail cash register transactions? Recognize your stock, serial register stream mismatches flag potential sweethearting, monitor movement of cash?

Unfortunately, it's very difficult to get that level of precision and tracking of individual objects (like bills) without very high res stereoscopic cameras. But Camio could help more crudely with basic triggers for suspicious *sequences* of events. For example, shoplifting might look like a sequence: 1) "people approaching front entrance" -> 2) "people at high-value display area" -> 3) "people departing front entrance" without the presence of a common intermediate event as 2a) "people at cash register". Camio hooks enable you to create these kinds of rules in custom-coded solutions.

But is it safe in the cloud.  sounds cool, vendor sounds like they have clue (see ISC West discussion article too.)  What's in the cloud, how'd it get there, did getting the data there perforate the site network perimeter, is my data hanging out in the breeze not technically private and/or owned by me, etc.  you know the drill...

@Rodney - Camio Box sits on the local network with your cameras/NVR and uploads only via SSL port 443. All inbound network ports can be closed, so it eliminates the port-forwarding security risks too. Both the transmission and storage of metadata and video is encrypted. The SSL connections even use Perfect Forward Secrecy so that even if the NSA somehow decrypted a single recorded SSL session, they can't use the keys for subsequent sessions.

Also, there are no account credentials stored on the cameras, NVR, or Box. Camio Box uses a special token that permits only uploading of the content. So if someone steels the Box for an offline attack of its memory or SD Card, there is nothing to be hacked and no way to gain account access.

 

Bottom line: all data is owned and encrypted for you only (and any authorized guests you explicitly add to your account).

Login to read this IPVM report.
Why do I need to log in?
IPVM conducts unique testing and research funded by member's payments enabling us to offer the most independent, accurate and in-depth information.

Related Reports

Bezos-Funded Deep Sentinel Tested on Mar 28, 2019
Backed by Jeff Bezos, the Silicon Valley startup, Deep Sentinel, has declared: No One Does Home Security Like We Do Our Surveillance Team has...
Bosch AI Camera Trainer Released And Tested on Apr 09, 2019
Bosch is releasing a highly unusual new AI feature - 'Camera Trainer'. Now, coming as a standard feature in Bosch IVA/EVA analytics, one can train...
The HIVIDEO $31 Face Detection DVR Tested on Apr 25, 2019
Face detection in a $31 DVR? That is what "HIVIDEO" (not to be confused with Hikvision, even if the company intends to do that) was promoting at...
Hikvision DeepinView Camera Analytics Tested on May 08, 2019
Hikvision is expanding its 'deep learning' offerings with a new camera series called 'DeepinView' claiming false alarm reduction and improved...
Hikvision DeepinMind 2019 Test on Jun 06, 2019
In 2018, Hikvision's DeepinMind AI NVR performed terribly, recognizing vehicles, animals, and other objects as humans, misclassifying demographics,...
Wyze AI Analytics Tested - Beats Axis and Hikvision, Now Improved, Now Being Terminated on Nov 25, 2019
When we originally tested Wyze's free person detection deep learning analytics in July 2019, they performed well, with few false alarms, but a long...
Yi Home Camera 3 AI Analytics Tested on Sep 10, 2019
Yi Technology is claiming "new AI features" in its $50 Home Camera 3 "eliminates 'false positives' caused by flying insects, small pets, or light...
Camect "Worlds Smartest Camera Hub" Tested on Oct 18, 2019
Camect is a Silicon Valley startup that claims the "Smartest AI Object Detection On The Market", detecting not only people and vehicles, but...
BriefCam Video Analytics Tested on Jan 06, 2020
BriefCam, acquired by Canon in 2018, is one of the most commonly used video analytics offerings in the West. But how well does it work? We...
Halo Smart Vape Detector Tested on Jan 16, 2020
The Halo Smart Sensor claims to detect vaping, including popular brand Juul and even THC vapes. But how well does it work in real world...

Most Recent Industry Reports

Hazardous & Explosion Proof Access Control Tutorial on Feb 27, 2020
Controlling access to hazardous environments requires equipment meeting specific ratings that certify they will not start fires or will not...
Motorola / Avigilon Drops ISC West on Feb 26, 2020
Motorola Solutions has pulled out of ISC West 2020 effective immediately, because of coronavirus concerns, IPVM has learned. This is done amidst...
Cancel or Not? Industry Split Over ISC West on Feb 26, 2020
The industry is split, polarized, over whether ISC West 2020 should run or be canceled. New IPVM survey results of 400+ respondents show heated...
Coronavirus Hits Sony, Bosch Says Switch on Feb 26, 2020
Sony's fall in video surveillance has been severe over the past decade. Now, they may be done. In this note, we examine Bosch's new...
Video Surveillance Cameras 101 on Feb 25, 2020
Cameras come in many shapes, sizes and specifications. This 101 examines the basics of cameras and features used in 2020. In this report, we...
Favorite Video Analytic Manufacturers 2020 on Feb 25, 2020
Video analytics is now as hot as ever, driven by the excitement of advancing deep learning offers. But what are actually integrator's...
Latest London Police Facial Recognition Suffers Serious Issues on Feb 24, 2020
On February 20, IPVM visited another live face rec deployment by London police, but this time the system was thwarted by technical problems and...
Masks Cause Major Facial Recognition Problems on Feb 24, 2020
Coronavirus is spurring an increase in the use of medical masks, which new IPVM test results show cause major problems for facial recognition...
Every VMS Will Become a VSaaS on Feb 21, 2020
VMS is ending. Soon every VMS will be a VSaaS. Competitive dynamics will be redrawn. What does this mean? VMS Historically...
Video Surveillance 101 Course - Last Chance on Feb 20, 2020
This is the last chance to join IPVM's first Video Surveillance 101 course, designed to help those new to the industry to quickly understand the...