Hikvision VMD And Intrusion Analytics Tested

By IPVM Team, Published May 04, 2018, 07:16am EDT

VMD worked poorly on Hikvision cameras, new IPVM testing found.

However, Hikvision ships analytics on practically every IP camera, with intrusion detection included on even their lowest cost models.

To see how these analytics work in real-world scenes, we tested indoors and out, using multiple cameras and fields of view:

  • Smart Series DS-2CD4526FWD-IZH, 1080p, 32-92° AoV
  • Value Plus DS-2CD2025FWD-I, 1080p, 108° and 86.4° AoVs
  • Smart Series DS-2CD4565F-IZH, 6MP, 31-95° AoV
  • Value Series DS-2CD2042WD-I, 4MP, 83° AoV

Inside, see our results on:

  • False alerts - where and how many
  • Indoor performance
  • Outdoor performance
  • Detection range / PPF thresholds
  • Configuration recommendations
  • Comparison to camera VMD
  • Comparison to DeepInMind
  • VMS integration

*******

** *** *****, **** when *********, *********'* ********* analytics ********* **** ***** alerts **** *******, ****** foliage/brush, *******, *** ***** sources. **** ***-*** *****/***** Plus ****** ****** *** Smart ****** ******* ********* these ******, **** **** ********** better ** ******** ************ in *** *****, ********* below.

************, **** ****** ***** ***** rates ********* ************* ***** default ******** *** ****** always ** ********* *** the ******'* ***** *****.

*******, ** ****** ***** series ** ****, ********* analytics ****** ********** *********'* camera *** *** ***** alarm *********, **** *** triggering **** ********** ******** on ****** ********, *******, shadows, *** ***** *******, and ********* ******* ** shadows, ****, *** **** digital *****.

******* ** ****, ** recommend ***** ****** ********* when ** ** ********* by ***** ****** *** or ***, ** ** greatly ******* ***** ******. However, **** **** ****, integrators *** ***** ****** set *** ************ ***** accuracy / ********* ** such ******.

Value/Value **** ****** ***********

** *** *****, ***** and ***** **** ****** cameras ****** ******** ******* with ********* ********, **** no ***** *********** **** light *******, ***********, ********, etc.

*******, ********, ***** ****** frequently ******* ** ******* brush, *******, *** ***** animals, ********** ** ******** used. ******* ****** **** can ** **** ** filter **** ** ***** alerts *** **** ***** reduce ********* ******** ** longer ****** *** ******* as ***** ******* *** be ****** *** ** their ********** ***** ****.

Smart ****** ***********

** ******** ** *** Value ******, *** ***** series ****** ******** ********, with ** ***** *********** from *******, *****, *****, small *******, ** ***** objects. *******, *** *** 4565 ******** **** ******* false *********** *** ** shadows ** ******** *** outdoor ****** ** *** field ** **** (********** of ********), *** *** 1080p **** *** ***.

*******, ***** ****** ******* were ********* ** ****** turning ** *** *** in ***** ***** *** rejected ***** ******* **** as *******, ********, *** reflections.

Limited *** ***********

*********'* ****** ********* **** limited *** ***********. ********* XProtect ******** ***** ****** by *******, *** **** are *** ********* ** Avigilon, *****, ** ******* (***** ******** * '**********' camera ******* *** ********* cameras).

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

*** ******* ********* ******* include ********* ********* *** line ******** *********, ******* in *** "***** *****" section ** *** *** interface. ** *** ***** *********, check "******" *** **** a ********* **** (** line, ********* ** ********). No **** ** ******** in ********* ********* ** default.

Configuration ***********

************* *** ********* ********* ******* between *****/***** **** (****) and ***** ****** (****) Hikvision *******, **** ********** options ** **** *** found ** *** *****:

************, **** **** **** the "***********" ******* ***********, ***** may ** ********* *** users ********** **** **** series:

  • ** ***** ****** *******, Sensitivity ** ****** *** size ** *** ****** which **** ******* *** analytic.

  • ** ********, *** ***** series ********** ** **** convoluted, *** *********** ***** the ********** ** * detected ****** ***** *** enter *** ****** ****** it ******.

***** ****** ******* **** object **** ***** ***/*** size ********, *** ***** in *** ***** ******.

** **** *** *************** for ********* ******** ** each ****** *****.

Value ****** ************* ***************

**** **** ** ******* settings ***** ****** ******* ********** false ******* ** **** sources **** ******* *** outdoors, ********* *******, *******, foliage, ********* ******, ****** switching ** *** ***, and ****. ** ** highly *********** ** ****** settings ** **** *** the ***** ** ****** false ******.

***** ** *** *******, we *********:

  • ******** *********:** *** *****, ********** threshold ** *-* ******* or ****** ******* ***** alerts **** ***** ******* and **** ******* *******.
  • ******** ******* ****:** ***** *******, ****, blowing ******, ***., *** issues, ********* ******* **** may ****** ***** ******, but ****** ** ********* adjusted ** ****** ***** detection ** ********, ** valid ******* *** ** missed.

********* *********** ***** ** difference ** *********** **** using ****-****** ********* *****, as ** ** ***** on *** ********** ** an ******'* ***** **** entering *** ****. **** the **** ** **** screen, *** ******* ** an ****** ******** *** zone ** ********. ** excluding ***** ** *** scene, *********** *** ** adjusted ** *********** ** shadows **** ******* ******* the **** ******* ********* alerts.

Value ****** ********* ***********

**** *** ***** ******** changed, *** ***** ****** cameras' *********** *** ***** indoors, ******** ****** ***** alarms **** ** ****** going **/***, ********* ******, and *******.

*******, ********, **** ***** settings *******, *** ***** series ******* ********** ******* on ***** ******* *** swaying ********** ** *** field ** ****. 

  

Smart ****** ************* ***************

**** **** ** ******* settings, *** ***** ****** camera ***** ********** ***** on ******** *** ******* in *** ****** **** as ******* **********, ***** animals, ******* *** **** dust ******** ** *** field ** ****:

** ***** ***** ******, the **** ********* ********** is ** *********, ******** by ***********:

  • ******** *********:** **** *** ***** line, ********* ********* ** 1-2 ******* ** ****** reduces ***** ****** ** many ******* ** ***** alarms, **** ** ****, debris, *******, ** ***** animals *** ** *** FOV *** * **** ***** time. ** ************ ********* detection ** **** ****** objects **** ** ******** in ****** ****** ** view, ***** ********** *** be *********.
  • ***** ***********:******** *********** ******* *** Smart ******' ********* ** small *******, **** ** dust, ******* ******, ** small *******. *******, ** may ****** ********* ********, with ****** ****** ********** missed **** *********** ******* to ~**. *********** ** the ***** ** **-** is ****** ******* *** most ******.

******* ********** ********* ********* to *********** ** *** Value ****** (*** *****), it **** *** ****** detection **** ***** ****-****** zones ** ** *** in *** *****. 

Smart ****** ***********

***** ********* ********, ****** testing *** ***** ****** camera ********** ******* ** the ******* ** *** field ** **** ** ~40ft ** *** ******* scene, ***** ******** ******** ** common ********* **** ** swaying ***** *** ***** animals.

*******, *** ***** ****** cameras ****** *** *********** issues **** *******, **** as *** ***** ** the *** *****, ********** of ********.

***** ********* ******** *******, Smart ****** ******* ***** ******** alarmed ** ***** *******, underperforming *** ***** ****** camera ***** *** *** have **** *******.

Compared ** ********* ****** ****** *********

******* *** ***** ***** issues ********* *****, ********* detection ******** ****** **** Hikvision's ****** *** ** all ******* ******. ****** testing, *** ******** ** many ** * ** 2 ***** ****** *** minute **** ****** ******** (below), *** *** **** susceptible ** ****** **** floating ****, *******, ******, etc., **** **** *******.

Compared ** *********'* **********

*********'* ********** ** ********* ************ ***** ******** intrusion ********* ******* *** issues **** ***************** ** animals, ********, *** ***** objects ** *****. **********'* false ***** ***** **** much *****, ********* * few *** ***, ******** to **+ *** ***** in **** ********* **** using *********. ********* ** claiming ************ ** ***** rates ** ******** ******** (planned *** "***"), ** well, ***** ** **** test **** ********.

Detection *****

*** ********* ******* ****** consistently ******* ** ******* ** and ** ***, ** *********** ranging **** ***** ** 6MP, *** ******* ****** of ****.

  • **-**********-*** (***** **:*):~**.*' **** *****/~**' ********* *****
  • **-**********-* (***** **:*):~**.*' ****/~**' ********* *****
  • **-********-*** (*** *:*):~***' ****/~***' ********* *****
  • **-*********-* (*** **:*):~***' ****/~**' ********* *****

Tech ******* **** ******** ***************

****** *******, ** ****** tech ******* ***** ******** times ****** *** *************** to ******** ******** ******** for *** ******* **** poor ******* *** ***** times. *** ***** *** techs *********** ******* ********* to ****, ***** ******* increased *** ****** ** false ****** ** **** cameras. *** ***** **** ****** informative, ********** * ********* of *, *********** ** 15, *** ********** ** 10, ***** *** ****** reduce ***** ****** *** **** ******** in ****** **********. ************, the ***** **** ********* ****** detection ** **** ******* of *********, ***** *********** increases ***** ******.

iVMS **** ***********

*********'* ********* ********* **** their *** ****-**** ***, with ******** ***** *********, events ************* ********* ** an ***** ****, *** ****** displaying ******** ***** ** the ********** *****. ***** may **** ******-***** *** event *** ******** *****.

 

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

*********'* ****** ********* *** not ********* ** ********, Exacq, ** *******, *** integrate ** ********* ******** out ** *** ***. Network ***** *** ********* support *** ***** ********* ** version *.*, ********* ***** in ****, ********* *** Q2 *******.

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

********* ********** *********'* ********* out ** *** ***, with ***** ****** ******* to ****** ***** *** specific ********* **** **** to *****/****** ** (*****). Note **** **** ****** are *********, *** ******** boxes.

************, ********* *** ********* a ********* ******** ****** [link ** ****** *********] ***** displays ******** ******** **** as ******** ***** *** alarm *****/*********, *** ** were ****** ** *** it ******* ****** *** tests, **** **** **** support's **********.

Versions ****

*** ********* ******** ******** were **** ****** *******:

  • ********* **-**********-***: **.*.** ***** 170310
  • ********* **-**********-*: **.*.* ***** 170711

Comments (17)

Avatar

Jon Dillabaugh

Pro Focus LLC
05/04/18 07:56am

We have been unable to get Hikvision Tripwire to work reliably in any model of camera, with the exception of cameras connected to a Hikvision 77xx series NVR. It seems their analytics are more reliable when connected to an NVR. I will admit, this was with very limited testing, but it was a casual observation we have made over time. 

Avatar

John Bazyk

Command Corporation
IPVMU Certified | 05/04/18 08:58am

I'd like to see a shootout.

"Worst Analytics" 

Put Hikvision, Dahua, Hanwha, Bosch, Avigilon, Ganz and DW all side by side and see which one's have the most false alarms. I don't even care about actual alarms, just false alarms. That would be a good show. 

Every time a rep comes in and tries to sell me on how their analytics are better than anyone else's, I ask for a free camera to test them before I even consider signing their NDA or signing up for their partner program. Most never call me again.

Here's my conclusion, analytics aren't great. Most of the time they're a marketing gimmick.

I made the decision about two years ago to switch to using Spectrum and haven't looked back. At the same time, our team decided to record 24/7 low res and high res on motion. Using the smart search function has been far more valuable to our customers than most analytics.

We just put in 40 cameras at a high-security building. One pointing right at the entrance (thanks IPVM camera calculator). I showed them the smart search and we looked through a day worth of video. ALL we saw were cars coming in and out of their driveway. 

AI and Machine Learning is the future, I think analytics were a great failed experiment (for the most part...not all analytics totally suck). Now it's time to learn from the failure and move on. 

Michael Miller

In reply to John Bazyk | 05/04/18 02:49pm

We just put in 40 cameras at a high-security building. One pointing right at the entrance (thanks IPVM camera calculator). I showed them the smart search and we looked through a day worth of video. ALL we saw were cars coming in and out of their driveway.

We have been doing this for years with Avigilon's analytics.  The big difference is we can actually search for People or Vehicles, not just motion events. Also now with Appearance search, once you find the person or vehicle you can then have the system automatically search the other analytic cameras in the system for that person or vehicle.

Avatar

Kyle Folger

IPVMU Certified | In reply to John Bazyk | 05/07/18 12:06pm

I would agree with Michael. Analytics may not be the greatest, but I have been impressed with Avigilon's analytics. They do work fairly well. While I install mostly Spectrum, I have installed Avigilon with the built-in camera analytics. 

If you use Smart Search indoors, most of the motion activity will yield similar results as performing a people search in Avigilon. However, if you perform a Smart Search outdoors and are looking for vehicles (not just on a tight shot), it will generally provide vehicles plus anything else that moved such a foliage and animals. Avigilon's analytics shine when searching outdoors for people or vehicles.

I will agree that most of the Analytics built into cameras aren't that great. However, Avigilon was able to purchase one of the best analytics companies at the time, so I would think they have a slight advantage.

Undisclosed Manufacturer #1

05/04/18 01:48pm

John,

Smart Search is great but very limited for proactive measures. In all cases, preventing a crime or dangerous situation is better than reacting to it after it happens. This is where analytics come in as you can immediately notify staff on site and\or trigger lights, sirens, activate third-party text messaging services, or play prerecorded message for deterrence, you can automatically create and tag a bookmark to save time on viewing potentially hundreds of smart search events, you can trigger presets on PTZs or zoom lens cameras to get a much more detailed look at the event and the list can go on and on.

Even in your example, if you could only show events on cars that were a specific color, that could have reduced the amount of investigation time by 99% potentially. What if you were looking for a pedestrian walking through the area where only cars are supposed to go? 

Any rep who promises the world with their analytics, or even uses analytics as a main selling point is a bad or lazy rep or they have a low-quality product in general (or they are tricked by their own marketing and have little real world experience).

A good rep will apply analytics as part of a solution to solve a problem and manage expectations of what the analytics will provide and advise you to add another 5 hours of labor to your install for thorough testing. (false alarm rate in varying conditions, correct camera positioning, activation scheduling, most efficient configuration, advice on policy or procedure to use analytics efficiently, what to do with the data, etc.) Or, they will tell you a physical barrier or device would work much better than a digital one.

A deep understanding of the analytic and having a VMS capable of automated actions based on that analytic event can win you projects against lower-cost competitors and can greatly increase the potential for ROI for your customer beyond just the operational efficiency of Smart Search.

Avatar

Jon Dillabaugh

Pro Focus LLC
In reply to Undisclosed Manufacturer #1 | 05/04/18 01:59pm

UDM#1

I would normally agree with you that analytics have been weak in our experiences, that is, until I recently bought a Hanwha SmartCam D1 doorbell that has facial detection, human detection, abnormal sound detection, and basic VMD.

I can tell you that the Hanwha implementation of human detection has not given me a single false alarm yet in the week it has been in place. It has also given me faces that it has captured for review, which I can ask it to alarm upon recognizing a subject in the future. I can even sort and name these subjects.

My point here is, if a residential doorbell can work this well, with a very wide shot, covering a generally busy area (I live in a condo), then why can't some of these other analytics perform as well?

Avatar

John Bazyk

Command Corporation
IPVMU Certified | In reply to Undisclosed Manufacturer #1 | 05/04/18 02:18pm

Even in your example, if you could only show events on cars that were a specific color, that could have reduced the amount of investigation time by 99% potentially. What if you were looking for a pedestrian walking through the area where only cars are supposed to go?

99% is a bold claim and I will disagree with it. Until Analytics are more reliable in security cameras I can't recommend them to our customers. We have dozens of cameras around our office running on several different VMS platforms. Few of them work reliably enough to be used in a security application. Obviously, that's just my opinion. Hopefully, that will change, I am more interested in AI than traditional analytics and see more potential there than anywhere else. 

I'm not sure who you work for, feel free to contact me. Send me one of your cameras and I'll test it. I personally sell thousands of cameras every year. We are always looking to offer our customers the best platform that meets their needs. Maybe you have something I haven't seen yet. 

Undisclosed Manufacturer #1

In reply to John Bazyk | 05/04/18 02:32pm

200 cars pass through a parking garage entrance to an office building in a day and you are looking for a yellow or green car, most likely there will only be 1 or 2 that show up (I don't know why anyone would take a cab into a parking garage). This is a 99%+ reduction in incidents you need to view.

Avatar

John Bazyk

Command Corporation
IPVMU Certified | In reply to Undisclosed Manufacturer #1 | 05/04/18 03:07pm

I'm sold on analytics. I think the concept is great. I want analytics to work, I want to sell them. I just haven't found a company that brings enough to the table to justify selling them.

IMO they do not work reliably enough to be sold to our clients. What if your analytics don't index the correct colors, let's say there's poor lighting one day due to rain or snow or their are lights are out in the garage, green looks grey or brown or black. You can't tell me that you're indexing colors correctly 100% of the time in real-world applications. When they don't work you're having to search through every car wondering why you spent so much more money on a camera analytics. 

I'll make my offer again, send me a camera. I'll put it up in our parking lot or at one of our client's buildings and test it out. If you're so confident in your system, let me test it, if it works as promised I'll be your biggest advocate. 

Undisclosed Manufacturer #1

05/04/18 02:24pm

I get what you are saying and this is a great support of my point. The application for that doorbell cam is the perfect situation for face detection analytic. The height it is mounted at, the distance faces will be from 2MP, integrated IR, they even configure the aspect ratio in 4:3.

Then you have reps who will just say, "Put this baby in your warehouse and you can tell humans from other forms of motion, then configure your VMS to send a text message when it happens" and the end user wonders why his forklift drivers 80 feet from the camera with strong backlighting from the open bay door shadow out their face aren't detected, but a shelf of gizmos 15 feet away with human faces on the marketing box sends a text at 3am.

Avatar

Luca Fogliati

05/05/18 10:35am

Have you ever talked with someone from HV to see how to properly set VCA on their cameras?

In the first example (outside scene)you are setting it wrong...you just cannot pretend to cover all that area but should draw a corridor in the middle of the area instead

Plus HV is clear about false alarms if you cover bushes,branches,leaves etc...they should just be taken out of the smart area

Also is it my monitor that is gone or you have some color problems with the camera (i see purple everywhere) white balance?ir cut filter?if you have wrong chroma again you cannot pretend things to work perfectly 

In my experience with correct position of the camera, correct lens choice and correct settings of the cameras i find it easier to have false negatives than false positives.. 

Plus i would expect to see a test of vca on thermal cameras since it's 2018...that's the real deal to be used with video analitics...we are using HV thermal with VCA to protect outside areas and get 98% reliability even in "wild areas "

Don't want to sound as an HV commercial but there are rules to follow when it comes to video analithics plus the company is pretty clear about limitations of some features at least HV Italy is and they are taking great care about costumers being teached how to properly set the cameras

John Honovich

IPVM | In reply to Luca Fogliati | 05/05/18 11:00am

Luca, thanks for your first comment!

Have you ever talked with someone from HV to see how to properly set VCA on their cameras?

Did you read the post? From the post:

Tech Support Poor Settings Recommendations

During testing, we called tech support three separate times asking for recommendations to optimize analytic settings for our cameras with poor results all three times. The first two techs recommended setting threshold to zero, which greatly increased the number of false alarms in both cameras. The third tech seemed informative, suggesting a threshold of 2, sensitivity of 15, and percentage of 10, which did indeed reduce false alarms but also resulted in missed detections. Additionally, the third tech suggested motion detection be used instead of intrusion, which drastically increases false alerts.

Maybe Hikvision Italy is way better than Hikvision USA but to answer your question, yes we spoke with Hikvision repeatedly.

Plus i would expect to see a test of vca on thermal cameras since it's 2018...

99.99% of surveillance cameras sold are not thermal. And Hikvision markets their analytics as working with conventional cameras. And we did the same test at the same time with Axis.

That noted, we are happy to talk to Hikvision Italy (john@ipvm.com or ethan@ipvm.com).

Avatar

Luca Fogliati

In reply to John Honovich | 05/05/18 12:56pm

I am not from HV Italy and yes it looks like the tech service in the USA is lousy if they give you these kind of answers....99% of analitycs success comes from proper dimensioning (installation height that is different for each analitych mode you want to use) scene shot and environment...really you cannot pretend VCA to work on that outside scene if you set it like that.....and HV in their tutorials (just as many other producers i work with)are pretty clear about that in case of intrusion detection you should take out of the area threes,bushes,car lights ,etc and draw a corridor as frontal as possible to the camera (camera's vertical installation angle's Paramount as well)

really if somebody would call me for support on bad behaviour of analitycs i would skip the setting part of the camera and first thing first ask for some screenshots from the camera.....

 

Dennis Ruban

In reply to Luca Fogliati | 05/08/18 12:21pm

I'd like to see a VCA tutorial you're talking about. Where can I download it?

Chris Chambers

In reply to Luca Fogliati | 03/09/19 02:56pm

I'm a little late here, just joined.  I can confirm IPVM's experiences, both with Hik false alarms and missed actual events on multiple models, and with horrid tech support.  (I've even taken to copying Jeffrey He on emails to see if that helps.)

I did try switching over to smart events, and that does reduce false positives.  On at least 2 cameras, however, that resulted in video freezing.  On the fairly new 4 MP mini dome PTZ, it was horrible freezing of up to 5 seconds or more, along with ghosting.  Shut off the smart events completely and it went away completely.  I returned the camera for a refund.  Too bad as the feature set and small size (albeit with a horribly insecure twist into place mounting method) is nice for minimally intrusive residential use.

Since I don't record full time, I set for 15s pre record and 10 seconds or 30 seconds post record.  When enabling smart events, the pre and post record settings seem to be ignored.  Thus I find that key details are missed.  Thus far Hik's tech support does not have an answer for this behavior.  Any feedback on ensuring pre and post record is done for all events would be very helpful.

John Honovich

IPVM | In reply to Chris Chambers | 03/09/19 10:11pm

Chris, welcome to IPVM!

I've even taken to copying Jeffrey He on emails to see if that helps.

I am sure, if he could fix these things, he would. It's a corporate / HQ / R&D issue.

There are a new generation of Hikvision products coming out that surely will be better (5 and 7 series cameras, Accusense, new firmware DeepInMind, etc.). We are working on such tests currently. That is the most likely route though how well they work still remains to be seen.

Undisclosed Integrator #2

05/09/18 08:23am

We ran an outdoor test with the newest Axis AVMD and generally it was fantastic.  But running against something picking up all motion (lights, trees, etc.) we found the AVMD missed 4 people walking across a parking lot.  This was without the small size filter.  Yes the other had a hundred events and the Axis was almost perfect, but picking up 95% of the events is a failure more so than picking up too many events.  

That's a problem....you don't know what you miss.  It is not reliable enough for our customers yet.  We also tested SightHound and found the same problem...it missed 2 people at night at an entrance.  Big deal to miss events.  Easy in a doorway or controlled area, but not as much in difficult conditions.

Read this IPVM report for free.

This article is part of IPVM's 6,653 reports, 896 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