Bosch IVA Video Analytics And Motion+ VMD Tested

By: Ethan Ace, Published on Jun 06, 2018

Bosch's video analytics now ship on nearly every model, from indoor domes to high-end 5MP starlight cameras. 

In this test, we evaluate Bosch's Intelligent Video Analytics, included on their Flexidome and Dinion models, as well as Motion+ VMD, using these models:

  • Flexidome starlight 7000 (1080p, Intelligent Video Analytics)
  • Dinion starlight 8000 (5MP, Intelligent Video Analytics)
  • Dinion bullet 5000 (1080p, Motion+ only)

Testing in indoor and outdoor scenes, day and night, in varying weather, examining:

  • Configuration recommendations
  • Calibration difficulty
  • False alert performance
  • Detection range
  • Analytics compared to camera VMD
  • VMS integration differences

For comparison, readers should also see our recent tests of Axis VMD4, Hanwha Wisenet X AnalyticsHikvision Intrusion Analytics, and Dahua Intrusion Analytics.

*****'* ***** ********* *** **** ** ****** every *****, **** ****** domes ** ****-*** *** starlight *******. 

** **** ****, ** evaluate *****'* *********** ***** Analytics, ******** ** ***** Flexidome *** ****** ******, as **** ** ******+ VMD, ***** ***** ******:

  • ********* ********* **** (*****, Intelligent ***** *********)
  • ****** ********* **** (***, Intelligent ***** *********)
  • ****** ****** **** (*****, Motion+ ****)

******* ** ****** *** outdoor ******, *** *** night, ** ******* *******, examining:

  • ************* ***************
  • *********** **********
  • ***** ***** ***********
  • ********* *****
  • ********* ******** ** ****** VMD
  • *** *********** ***********

*** **********, ******* ****** also *** *** ****** tests ** **** ****,****** ******* * ****************** ********* *********, *** ***** ********* *********.

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

*******

** *** *****, *****'* Intelligent ********* ****** **** **** when ******** **********, **** very *** ***** ******:

  • *** ***** ****** **** common *******:*** ******* ** **** few ****** ***** *******, such ** ***** *******, blowing *****, ******, *** shadows.
  • *********** ********* ** ****:*** *** ********** ****** to ****, **** **** * single ***** ****** ** * droplet ** *** ******'* dome, ********* *** ******* constant ****** ** ******, such ** ****, *****, Hanwha, *** *********.
  • ****** ********* *****:*** ******** ******** ** ~10 *** ** *******, ***** than **** *********** ****** analytics ******, ********* ** longer ********* ****** ***/** wider *** ********.

*******, *********** *** ** a **** ********* *** ******* process which *** ******* ***** and ***** *** *** be ********* *** **** experienced ********** ******* ********, with ****** ********* ********* to ************. 

*******, ******** ** ***, Bosch's ******+ ****** *** performed ******, **** ******** triggers **** ****** ******* which *** *******, ********** moving ******* *** ***** changes.

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

*****'* *********** ***** ********* is ********* ** ****, 8000, *** **** ****** cameras, ***** ******* * second ********* *** ***** analytics. **** ******** ****** such ** *** ***** ********* ********* ***** *** ****** *** *********.

***** ******** ********* ***** Analytics ** ****, ****, and **** ****** ******, which ******** **** ** the **** *****, *** ***** some ********, **** ** face *********, *** ** more *********** ** ******* and ********* ********* *******'* *************.

*** ****** ******* ******+, *****'* camera **** ***.

*********** *******

****** **** ****** ********* in *** **** ******, Bosch *** ******** *********** for ****** *********. **** process ******** ******* *****, angles, *** ******** ******* on *** ******'* ***** of ****, *** ******** measurements ** ***** ******* and *******. 

** * *******, ***** recommends ***** *** ******** markers *** ****** ** a ****** (~*.**/*') ** the ********** *** ********** of *** ****. *******, for ********* ********, ***** vertical *******, ** ***** two ************ ****** *****, and ** ***** *** recommended. 

** ****** *********** ** this *****:

Calibration ******

** *** *****, *********** frequently ******** **** ***** and *****, *** *** reasons:

  • ***** *** **** "****** pieces" ***** *** ************, such ** ****** *****, height, ***** ******, *** calibration ********, ***** *** cause *********** ******* ************ or ***** ************. 
  • *********** ***** ******* ** the **** ** *** field ** **** ** difficult, **** **** ******* in **** ******, *** can **** ******* ****** on *** ******'* ***********, especially **** *** ***/**** of *** ***** ** view ***** * ****** pixel ********** ******* **** or **** ** ********.

******* ** ***** ******, inexperienced ***** *** **** calibration ***********, ** *****, or *** ** **** wildly ********** *********** ****.

** ****** ***** ****** in **** *****:

Outdoor ***********

********, **** *** **** and **** ****** ****** ignored ****** *** ********* of ******* *****/*******, **** false ****** ********* ** a **** ** ***** 1 *** ***.

****** *** *****, ***** animals **** ******* *******, a ******** ***** ***** source *** *********** *********.

*** *** ********** ****** of ***** ****** *** human *******, ***** *****, caused ** ****** ****** the ******.

****** *********** ******, ***** alerted **** ********** ** raindrops ** ***** ******/*****, Bosch *** ******* **** *** ****** *** ******* ******* 3 ******.

 

Indoor *** ***********

*******, *** ******** ****** from ***** *******, **** as ****** ****** **/***, adjacent **** ******, ** displays ***** ****** **/***, shown *****.

******** **** ******* ******* without ***** *******, ** well, **** ** ****** of ******** ******* **** objects ** *** **** or ***** **** ** they **** *******.

Reliable ********* *********

*** ******** ******** ** much ******* ******** **** subjects ********** **** **** subjects ***** ****** *** field ** **** ** moved **** **** *** camera. *** ********, *** starlight ***** ****** ******** at **** ~**' **** subjects ******** **********, *** reliably ******* ******** ** 200'+ ** **** ****** away **** *** ****** or *******.

******* ******** ***** ** PPF **** *********** *** ~4-5 *** **** ********.

Versus ******+ *** ***********

******* *****'* ******+ ****** VMD ** *** **** scenes ** ***, ** found **** ******+ ********** false ******* ** **** sources, ********* *****, *******, small *******, ***** *******, etc. **** ***** ******** *********** to ** **** * default ** **, ******+ still ******* ** ***** changes, **** ** ****** grass *** ******* *******, shown ****:

VMS ***********

***** *** *********** ****** depending ** *** ***, with ******* *** ********* supporting **** ****** *** bounding *****, ***** ***** supports ******, *** ** bounding *** ***********. ******** and ******* ***** ** not ********* *** ****** at ***.

**** **** **** ** this ******* ** ********** new ** *** **** 1-2 *****. ***** *** no *********** ***** *** 2016, *** ******** ***** were **** ********* ** Genetec ** *********** *****.

Exacq *** ***********

*** ** ********** **** ***** by ******* *** ******** ***** to ******** ****** ** Event ******* **** ** trigger *********, *************, ** other ******* ** *** VMS, ****** ******* ***** simply ******** ****** **** no ********** *************.

 

Genetec *********** 

******* ******** ***** ******** events **** ** ************* required. ***** *** ****** for IVA ****** ***** ******** Desk ** **** ***** with *** ***** ***** type.

Milestone ***********

**** *****, ********* ******** rules ** ** ******* for ********* *** ******, similar ** *** ******** handles ***** ****** *********. Bounding ***** *** ********* in **** *** ********, shown *****.

Versions ****

*** ******* **** *** ****** firmware ** ** *** time ** *******: *.**.****

Comments (43)

Solid products, solid pre and post sale support. 

Interesting they chose to have no feet measurement.

Also i don't see this calibration process being realistic for more than couple of cameras job.

I've never quite understood the resistance to calibration - I've almost always seen better analytics results with systems where manual calibration like this was applied. It seems like if you have a reasonably intelligent duo, one walking around, the other on the software (actually the only one needing to be savvy), you can whip through a camera every five minutes. Sure, it's extra effort, but don't you want it to work? 

So 100 cameras @ 5 minutes, 500 minutes x 2 technicians @ $50/hr cost marked up to $95 = ~$1,580, not accounting for walking in between cameras, travel to and from the job over the course of 2 days.

Of course they want it to work, but you do need to sell it/find someone willing to pay for it.

5 minutes is a wishful thinking if you need that other person to stand like this in front each of the camera it's not going to be 5 minutes.

Also this is $1,580 on top of all the normal installation cost of course.

I actually think the training investment to teach people how to calibrate is a much higher cost than the per-project cost. I would honestly tell any integrator to plan on about a day's training to learn to properly calibrate IVA consistently. If you're going to configure it infrequently, I'd say count on retraining each time you touch it.

I'm not saying this analytic is for every camera on every install, but it is noticeably more accurate than other camera analytics immediately. The higher number of false alerts generated by others are easily going to chew through $1,600 in added cost in response time, with the added downside of users starting to ignore alerts because they are too frequent.

"but it is noticeably more accurate than other camera analytics immediately."

I am going to say this before Avigilon dealers start protesting in front of Ethan's house - Ethan is referring to our other tests in this series - Axis, Dahua, Hanwha, Hikvision.

We've tested Avigilon in the past and we are testing Avigilon analytics again. Given their past performance and integrator ranking, we expect them to perform well. On the other hand, Avigilon analytics are increasingly an Avigilon closed offering so Bosch still has value relative to other 'open' analytics.

And we do plan to test more higher-end / specialist analytics, now that we have the foundation of a number of mainstream cameras done.

You assume the first calibration for all the cameras went right and they all work good, we know in the real world it's probably not accurate.

Interesting. Using OV but performing better than their competitors that use the same technology.

Would be nice to compare with Avigilon, Davantis, Xtralis, etc.

What do you mean "Using OV"? Bosch licensed OV's patents to settle the lawsuit against them, but did not license OV's software. 

Sure? :)

Absolutely home grown (well, the smart guys were acquired from the VCS acquisition)

Did they get the ground plane angle configuration to retain its view?

Great writeup, thank you guys for the hard work that goes into these. 

Approximately how long does it take per camera to calibrate and confirm the correct calibration? 

 

The first one, probably 10 minutes. The tenth one, probably 3-5. Not including walk testing you'd do on any analytic system to make sure things are alarming where you want them to and not alarming where you don't.

The problem is not time on such things, its mistakes or problems, i.e., if you do it correctly, it's X minutes but if you make a mistake or there is some weird case where it does not calibrate correctly, it could be mean multiple site visits, etc. Also, the time to learn this, i.e., Bob is trained on this but Bill is not but Bob is busy and Bill is on site, etc.

The more you make the user do, the more likely there is to go wrong. That fact that it works well is good, especially since analytics still suffer from so many problems. But Bosch should strive to reduce calibration or, better, auto-calibrate.

Agreed. It’s still not a mass market option. We know Bosch are good on analytics but from a dollars and cents point of view I think this leaves much to be desired. The ROI would be bad on both sides. One tying up a solid tech on this for hours (though still charging money) and the client paying a lot for trial and error. I am not sure how many customers out there would accept a service rate of X amount spending hours fixing the camera due to scene change, rolling out an unfamiliar tech or issues with the camera itself so then you have to setup the IVA up again. I guess if someone has a really important spot at their location it may worth for one camera with IVA but I know which folks would have sticker shock on this.

 

Concerning XProtect:

 

Via Bosch's Integration Partner Program (IPP), users of Milestone XProtect can take advantage of two plugins:

https://ipp.boschsecurity.com/en/integrated-solutions/partners-solution-advisor/find-a-partner-solution?partner=Milestone%20Systems

Analytic Service Plugin:

"The Bosch Analytics Service for Milestone XProtect acts as a software bridge between the Bosch Analytics Events, generated on a Bosch camera and Milestone XProtect. Generally, the Bosch events are transformed into Milestone Analytics Events that are available in the Milestone rule or alarm engine."

Bosch Analytic Service - Software Context

 

Enhanced Metadata Plugin:

"The Bosch Enhanced Metadata plug-in for Milestone XProtect VMS Products adds advanced metadata rendering to the Smart Client:
– Object shapes
– Object trajectories
– Fire and smoke objects
– Thermal objects are visualized by the plug-in"

 

Bosch Metadata Plugin: Client object color configuration

 

Furthermore, I hope the plugins are a precursor to "forensic" search for Bosch products in XProtect.

Are the bounding boxes burned in the video when used with Exacq, Milestone, and Genetec? Or can they be turned on/off when needed? 

They aren't burned in, they're metadata. You can turn them off anytime. 

First off, nice work on the test, as John B says, you put some serious effort into these and it is appreciated. 

Do you know if these cameras have an alarm output (ie NO/NC) that can be linked to an event?  

Most of them do, yes. The Camera Finder lists 128 models with an output, and scanning it, it looks like about 50% of them include either IVA or Essential VA.

Can you run all available analytics at the same time on camera? Or how many of them can be run at the same time?

You can run up to 8 rules on the camera, or up to the max the CPU will handle, whichever comes first.

There are 8 rule slots in the VCA task setup:

And the CPU gauge can be found in the camera web interface or in Configuration Manager. Web interface shows it on the top right:

What happens when there are too many things to process? As you near or hit 100% professor capability? Does it notify? Alarm? Stop analytics from working?

Please note that for the calibration of newest camera types like FLEXIDOME 7000, you only need to enter the mounting hight of the camera. All other informations are availlable for the camera by its integrated position sensors. That's pretty easy.

Yes, it does have a sensor built in, but in our testing the calibration that was produced using only mounting height was not accurate.

For example, entering only mounting height and loading all other sensor values consistently resulted in the horizon being inaccurately skewed and too low:

At other times the horizon was almost 1/3 of the way down the FOV. In our tests, loading sensor values never resulted in proper calibration. It might work fine in some instances, but for long range and best false alert perform, Bosch recommended full calibration.

What was the mounting height of the camera?  

My apologies, the mounting height was in the calibration video.  IVA is designed to work best at typical mounting heights, 10 feet or higher.  Bosch lists a minimum mounting height in the manual.  That being said, and as you have noted, it works below 10 feet. I’ve calibrated countless cameras and have noticed that when mounting below 10 feet you sometimes have to manually calibrate the cameras for best results.  Above 10 feet the norm is just to enter the height, resulting in the proper horizon.  

As an added note, performance also increases with height.  For those that are astute at mathematics, you might notice that by having the height of the camera, the tilt/roll angles from the sensor, and a flat plane you can deduce a whole bunch of information.  A higher mounting height provides more accurate numbers further away from the camera as the angle is not so shallow.

The mounting height in testing varied, actually. We didn't get it up to 18 or 20 feet, but it was mounted at 12-13'. I believe our building is only 16' high where we were testing, so 12-13 is not unrealistic. We may look at mounting higher and checking calibration.

Ethan- Great job on this report. A few questions.

On a scale from 1-10, how hard was for you to calibrate these Bosch cameras? I know you had to do a lot different scenarios for this report so i know that took hours. My gut tells me the average industry installer can't handle this. 

Also, what did the IR in the bullet model do to the analytics false alarm? Any difference?

Honestly, calibration is a 5 at most. I've been thinking about it this morning, and comparing it to Axis VMD4, Hanwha Wisenet X, Hikvision, and Dahua IVS, I do not think it's actually harder to do right. I emphasis right because out of the box those other camera analytics do not require this calibration process, but you absolutely have to tweak settings based on the scene in order to not get hammered with false alerts.

By contrast, the Bosch process is front loaded. So you're spending more time learning how to calibrate, and doing initial calibration, and may need to tweak it, yes, but once calibrated, we didn't have to go back and adjust anything to reduce false alarms. 

What is the price of these cameras when compared to Avigilon?

I found that it is very easy to train technicians to install Avigilon cameras.

On the other hand, they are expensive. When comparing the 2 and taking into account the long calibration of Bosch which is more cost-effective?

They are very cheap compared with extremely high performance, integration with VMS and ONVIF profile T. For a motorized Bullet 1080p with accelerometer and essential video analytics (EVA), that is almost the same of intelligent (IVA), it works until 50 meters (100 the IVA) and cannot work in marine/lake/river enviroment (for example ship detection) we're selling @ 399€ in Italy.

 

399€: installer cost or customer cost?

When I write a sow, how much detail do you recommend I add describing analytics.  I.e.

1. Should I describe in great detail the analytic expectation for each camera?

2. Should I give a maximum allowable false alarm rate?  What would be a fair false alarm rate for these analytics based on your tests, seems like perhaps 1 a week?

When I read the test, I think of these things as how do I get my integrator to set the cameras to their greatest potential and know they calibrated them effectively?

Based on your finding and the discussion it also looks like to be fair I have to build in some value to my coast estimate for this analytic tuning evolution.

NOTICE: This comment has been moved to its own discussion: End User: How Do I Specify Video Analytics Performance / False Alarm Rate?

Bosch is implementing developments coming from their Automotive Industry regarding video analytics used in drive assistant software. They are way ahead of the competition in this area I suppose.

Also worth mentioning the Forensic Search function, which means that all metadata is stored with video material, so you can not only create 8 preconfigured IVA rules for analytics events, but lookup recordings later by any IVA query you create, even if it was not a preprogrammed rule. It is also working with Genetec as far as I know, if you record on BOSCH VRM system or internal SD card.

Hello,

when you talk about calibration, you don't mention that all the new series (4000i,5000i,6000i,7000) have an integrated accelerometer that automatically sets tilt and rotation of the camera as well, for the models with motorized lens (bullet and dome) the focal lenght. The only parameter that must be set manually is the height from the ground in meters, rendering calibration a matter of seconds

Moreover, in order to exclude small animals, oscillating objects and so on, there are dozens of rules that can do the difference, for example the aspect ratio of the object and its absolute size, the fact that it's moving in a restricted area (think about a tree that is oscillating strongly) and so on; from our direct experience I can guarantee that the false alarm ratio can reach 1/2 FA/Month per camera in complex scenarios (moving trees/bushes, some animal walking around, normal wind and rain).

You did'nt mention VCA Task editor function as well, that gives the installer the chance of modifying (with a little coding) the behavior of every single VCA task in order to limit false alarm rate.

Well said but Bosch failed with the 4000i and 5000i minidome cameras and the use of extremely CHEAP adjustment rings on the minidomes. If you have any latency at all, heck even with an analog monitor hooked up to the analog output it can takes upwards of 10 minutes to focus these. IF you get the camera into focus good luck tightening down the set screws. Don't get me wrong I am a strong Bosch proponent as seen from my many posts but they completely failed with these cameras. I can't justify selling them anymore due to that major defect. It looks really bad to the customer when we spend far too much time focusing these. As far as the analytics they work extremely well. I have one setup to sense loitering in the office managers doorway to send her and email that reads, "Tell them to get back to work and quit bugging you" when someone stands there for more than a couple of minutes. 

when you talk about calibration, you don't mention that all the new series (4000i,5000i,6000i,7000) have an integrated accelerometer that automatically sets tilt and rotation of the camera as well, for the models with motorized lens (bullet and dome) the focal lenght

We do talk about the integrated sensor. It's discussed in videos showing calibration, at length.

The only parameter that must be set manually is the height from the ground in meters, rendering calibration a matter of seconds

That is totally true in theory and performance was pretty good when only doing so. But we found it was never totally accurate without doing full calibration, so if you were looking for long range/low PPF detection, absolutely add measured points and markers and verify the calibration.

I have no doubt you can get false alarms down to 2 per month in some scenes, and I'm aware of the VCA Task Editor. The question is how much time that takes per camera. No doubt, in many installations, it's worth it and there is a budget to do so, but if you have to run a few minutes' calibration, filter aspect ratio or other elements, or do coding to edit the VCA task, it is absolutely more time consuming than other options.

The high level point is that Avigilon, as others have said, produces good results without this process. They may not be as absolutely accurate as Bosch when heavily tweaked and adjusted, but for most users that's not going to matter because they will not put that much time into the adjustment.

 

 

 

Once you know how to Calibrate them the process is quite simple and fairly quick. I would rather have the analytics work then have no setup and the false, and don't work as needed.

Thank you for the article. Do video stream settings impact IVA functionality ? Resolution, compression, etc. ?

 

Good question. Video stream settings don't impact performance. Processing is done before video is encoded, so those settings aren't taking effect until after the IVA is processed.

Read this IPVM report for free.

This article is part of IPVM's 6,380 reports, 860 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

Related Reports

Video Surveillance Architecture 101 on Feb 18, 2020
Video surveillance can be designed and deployed in a number of ways. This 101 examines the most common options and architectures used in...
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...
Camera Analytics Shootout 2020 - Avigilon, Axis, Bosch, Dahua, Hanwha, Hikvision, Uniview, Vivotek on Jan 22, 2020
Analytics are hot again, thanks to a slew of AI-powered cameras, but whose analytics really work? And how do these new smart cameras compare to top...
Leica Launches LIDAR / Thermal / IP Camera on Mar 04, 2020
Swiss manufacturer Leica is launching what it calls a "real-time reality capture device" fusing LIDAR, thermal cameras, IP cameras and more. In...
Video Surveillance Trends 101 on Apr 01, 2020
This report examines major industry factors and how they could impact video surveillance in the next 5 - 10 years. This is part of our Video...
Surveillance Storage 101 on Mar 23, 2020
This guide teaches the fundamentals of video surveillance storage. Inside we cover: Surveillance Storage NVR / Recorder Storage Camera /...
Video Analytics 101 on Mar 16, 2020
This guide teaches the fundamentals of video surveillance analytics. Inside we cover: Why Use Video Analytics Video Analytics Warning Where...
VSaaS 101 on Mar 25, 2020
Video Surveillance as a Service (VSaaS) is the common industry term for cloud video. But what does it mean? How does it all work? Inside this...
Euklis Presents AI Analytics on May 05, 2020
Euklis presented its AI facial recognition, LPR, and object recognition analytics at the April 2020 IPVM New Products show. Inside this...
Embedded Logix Thermal Temperature Detection System Examined on Apr 08, 2020
Embedded Logix has been producing thermal temperature measurement systems for industry and fire detection for over 10 years. Now, they are entering...

Most Recent Industry Reports

Remove Dahua and Hikvision Equipment "Immediately" Or Else Banned From US Government Contracts on Jul 10, 2020
The US government has directed contractors to remove covered equipment, such as Dahua, Hikvision, and Huawei Hisilicon products, "immediately." If...
ZeroEyes Presents Firearm Detection Video Analytics on Jul 09, 2020
ZeroEyes presented its Firearm detection Video Analytics system at the May 2020 IPVM Startups show. A 30-minute video from ZeroEyes...
Directory of 162 "Fever" Camera Suppliers on Jul 09, 2020
This directory provides a list of "Fever" scanning thermal camera providers to help you see and research what options are available. There are...
Clinton Public View Monitor (PVM) Mask Detection Tested on Jul 09, 2020
Face mask detection, or more specifically not wearing one, is expanding amidst the pandemic. Clinton Electronics has added this capability to their...
These Florida Real Estate Agents Are Now Selling "SafeCheck USA" Temperature Detectors on Jul 09, 2020
The "Kakon Brothers", William and Nathan, are self-described "south Florida Power Agents specializing in Luxury Real Estate" who "have closed over...
Avigilon ACC Cloud Tested on Jul 08, 2020
Avigilon merged Blue and ACC, adding VSaaS features to its on-premise VMS, offering remote video and health monitoring that was previously limited...
Hikvision's India Dominance Faces Threat on Jul 08, 2020
While Hikvision has become a dominant video surveillance provider in India, recent tension between the governments of India and the PRC is...
The US Fight Over Facial Recognition Explained on Jul 08, 2020
The controversy around facial recognition has grown significantly in 2020, with Congress members and activists speaking out against it while video...
Sperry West / Alibaba Tablet Temperature Measurement Tested on Jul 07, 2020
In April, we ordered a ~$500 temperature tablet from Alibaba. We set it to the side while doing 18 other temperature screening tests but, after...
Facial Recognition: Weak Sales, Anti Regulation, No Favorite, Says Security Integrators on Jul 07, 2020
While facial recognition has gained greater prominence, a new IPVM study of security systems integrators shows weak sales, opposition to...