Axis Zipstream Claims Average 50% Savings

By: Ethan Ace, Published on Mar 25, 2015

Axis claims that its new Zipstream compression technology can "lower bandwidth and storage requirements by an average 50% or more."

That is quite an audacious claim for such a fundamental element of video surveillance systems.

In this note, we explain how Zipstream works and how it matches up against other bandwidth 'saving' technologies and rival products:

Dynamic Scene Compression - How It Works
Dynamic I Frame Interval - What It Does
Axis Best and Worst Case Zipstream Scenarios
Axis Availability Restrictions
Zipstream vs H.265
Zipstream vs Vivotek Smart Stream and Arecont Bandwidth Savings Mode
Zipstream vs Ambarella
Zipstream vs Dahua and Hikvision
Zipstream Marketing Impact

************** *** *** ********* compression ********** *** "***** bandwidth *** ******* ************ by ** ******* **% or ****."

**** ** ***** ** audacious ***** *** **** a *********** ******* ** video ************ *******.

** **** ****, ** explain *** ********* ***** and *** ** ******* up ******* ***** ********* 'saving' ************ *** ***** products:

******* ***** *********** - How ** *****
******* * ***** ******** - **** ** ****
**** **** *** ***** Case ********* *********
**** ************ ************
********* ** *.***
********* ** ******* ***** Stream *** ******* ********* Savings ****
********* ** *********
********* ** ***** *** Hikvision
********* ********* ******

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

**********: **** ********* ***** ******* *** **** ********* ********** [**** no ****** *********]

Bandwidth ********** ****

********* **********, **** ****** cameras **** *** **** manufacturer **** *** **** CODEC, **********, ***** **** and *********** *** ** huge. **** ** ********* to ********** *** ********** how ********** ** *** Zipstream **** **.

*** ***** ***** ***** two **** *******, *** Q1604 *** *****, **** set ** ****, ** FPS, ************ ** ~**, across * **' ********** FOV. ******* ***** ******* being ************, *** *****'* bitrate *** *** **/* while *** ***** *** *.** Mb/s, ****** ** *** bandwidth.

************, ******* **** ** sharpness, ********, *** ***********, etc., ****** ********* ****** of ******* ** ********* scenes, ** ** **** in *********** ****** ********* **** Results. *** ** ***** variations, ********** *** **** savings ********* **** ******* in *** ***** ********** is **** **********.

Dynamic ***** *********** - *** ** *****

*** *.*** ** **********, and *** ****** ** compression ** ********** ** the *********** / ************ level. ** **** *******, this ** * ***** from * (***** ***********) to *** (**** ***********). This **** ** *.*** quantization ****** ** * to ** (*** **********, see *** ***** ******* / *********** Tutorial).

********, * ****** *********** level ** ******* ** an ****** ***** ******. For *******, * ******* Axis ****** *** * compression ***** ** **. The ***** ***** - whether ** ** * white **** ** * person's **** *** *** same *********** *****.

*** ***** ** *** two *** ********* ********, ******* ROI ******* *********** / quantization ****** ***** **** scene *******. ****** ******* such ** ****** ** vehicles *** **** * quantization ***** ** ** applied, ***** *** ********** of *** ***** **** as ***** *** ******** are **** **********, *.*., a ************ ** **.

*** ******* ** ***** different *********** ****** *** different ***** ** * stream ** **** '****' are '******' ** ******* portions. *** *******, *** generally **** *** *********** on * **** *** high *********** ** * blank ***** **** ** typically **** *** ***** bandwidth / *******.

*** *** ***** **** this ********* ** *** accurate *** ********** ** in ******** **** ***** of *** ***** *** be ******* * **** compression ******* ****** ********** details.

Dynamic I ***** Interval - **** ** ****

** ******** ** ********* compression ***** ** ***** activity, ********* **** ******* I ***** ********. *** example, ** * ***** scene **** ** ******, ** I ***** *** **** be **** ***** * seconds ** ****, ***** no ***** ******* **** occurred ** *** *****. When ******* ****** *** scene *** ***** *******, however, *** ****** ******* the ******** ** * shorter ********, ******* ** I ***** ** **** each ******.

**** ******* ** **** are ********* * "*** *** ****** ************* that ****** *** **** the ******* *** *** be." **** ******** * brake ** ****** **** very **** * ***** intervals ** *** ***** problems *** *** ******* that ****** ** ******* I ***** ********* *** search *** ********.

Axis **** *** ***** **** ********* *********

** ****' ********* **********, here *** ***** ****** for ******** ****** ********* scenes:

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

******** ****, **** *** retail ***** *****, *** hard ** ******** **** more ******* ******** ******* loss, ******* **** ******* lots ** ******* **** of ******* *** ****** walking *******.
***** ****, **** *** parking *** *****, *** easier ** ******** **** without *********** ******* ****, because *** ******* *** relatively ***** / **********.
***** **** ********* ** the **** ****** ** waste ********* *** ** compressing ***** *********, *** therefore, ****** *** ******* savings *********. *******, ***: ******* ********* ** *** Light *** ******: ******** ********* ** Night ** ****.

Axis ************ ************

****** **** ** ****' existing ******* **** ******* Zipstream, ******** *********** *** usability ** *** ***** term.

********* ** ******* ** Axis ******* *************-*, ***** ** ****' newest ******** ****. *** only ******** ******* **** can ** ******** ** Zipstream *** **** *****, **** ***** and *** **** *** Series. *******, **** ** releasing * **** ** new ** ***** ***, of ******, ********* ******** new ******.

******** ** ******* ******** cameras **** *** ********* upgradeable ** ********* *** release ** ***** *, 2015. ** **** ** test *** ***** **** this ** ****.

Zipstream ** *.***

** *** **** ****** for *.*** *** ** not ****** **** ** adopt ** **** ****. To **** ***, ********* could *** ** ************ against ******* ****** ******* out *.***, ***** ** already ********* *** **** surely ********** ********** ****.

**** **** *.*** *** marketing ***** ** '** to ** *******', **** similar ** **** ***** with *.***. ** **** end, ** ********** *** Axis ******** *****, '*** switch ** * *** codec ***** ***** *** support *** ********** ***** of ****** ******** *** use, **** *** *** stay **** *** ***** you ****/***** *** * proven ***** - *.***.'

Zipstream ** ******* ***** ****** *** ******* ********* ******* ****

********* ****** **** ************ with*******'* ***** ******, ** **** *********** adjust *********** ** ******* in *** ***** ** reduce *********. ** *** test ** ***** ******, reductions ** **-**% ** well *** ****** *** up ** **% ** low *****, **** ***** of ****' **% *** 90% *********** ** ***** scenes. ************, ** *** Vivotek ****** ******, ******* bandwidth *********** ******* ***** Stream *** *** ****** than *******, ** ***** these ********** **** ******, bandwidth ******** ****** ****.

*******'* ********* ******* ****, ******* ******* ********* feature, does *** ****** *********** or *** ** ********* does, *** ******* ******* some ******* **** *** scene, ********* ********* *** reducing *** ****** ** colors *********. ** ***** bandwidth ******* **** **% to **% ********* ** the *****, *** ******** image ******* ****** *** to **** ****.

Zipstream ** *********

********* ** *** ******* encoder **** ******** *** high-end IP ******* *** ***** chips (**** *** ** and *********** ** *.***) ******* ***** ********* consumption / ******* *** cameras ***** ****. *********’* ***** ******* ************* they **** “********” **** extend ****** ******* ***** downwards, **** **** ***** to ** ** *** as *** - *** Kbps *** **** ** video. ********* *********, ** course, **** ********* ***** for ***** ********.

*** ********* *********, ***** and *******, *** **** well ** *** **** ** ****** ********* / Storage ********. ** **** ** test **** ********* ******* such *******.

Zipstream ** ***** *** *********

******* ***** ** ********* market *** ******* ********* savings. ** **** ***, Axis *** ** **** to ******** ********* ** a ****** ** *** a ******* *** ***** cameras, ********** *** ***** gap / ******** ** Axis ******** ** ***** biggest ******* ***********.

Zipstream ********* ******

*** **** ** **** any ************ ************, ***** their **** *** ******* numbers, **** *** *** greatest ******* ** **** the ****** **** ***** initiatives (*.*., ***** *** LightFinder). ** **** ***, we ***** **** ********* could **** ****** ** launch ***** *** *********** optimizations / ********* *********. And ** *** *******, as *********** *** *** the ***** ** ****** after *******, ***** ******* them * ********** *********** advantage.

**********, ******, * *********** portion ** *********'* ****** will ** ***** ** how **** ** *** really ******** ********* ******* sacrificing *******. ** ****** to **** *** **** Q1615 **** *** ******** is ******** ** ***** ****.

Comments (22)

Luke Maslen

IPVMU Certified | 03/25/15 10:41pm

Hi Ethan, thank you for a very interesting report. I will be very interested in reading IPVM's test report of bandwidth savings and image quality when the Axis Zipstream feature is released.

I wonder what would happen if Axis used their Zipstream feature in combination with H.265 compression? I'm a little surprised they aren't implementing both technologies now rather than waiting another year to support H.265. It would have put Axis in a better competitive position with Dahua who have recently started to trickle out H.265 devices. It will be interesting to see how the technologies compare as these two companies roll out their new products.

John Honovich

IPVM | In reply to Luke Maslen | 03/26/15 12:52am

Axis has already stated that Zipstream can be used with H.265. The real question is when Axis releases H.265? Is it next month? next year?

Btw, has Dahua formally announced any H.265 camera? I can believe it's coming, I just had not seen a camera one.

Luke Maslen

IPVMU Certified | In reply to John Honovich | 03/26/15 03:39am

While giving some existing Axis camera owners a firmware update to add the Zipstream feature is a welcome idea, releasing new Zipstream cameras without H.265 suggests one or more of the following to me:
• H.265 chips could be too buggy and not ready to unleash on the world.
• short-sighted not to have added H.265.
• lack of engineering capacity and/or capability to implement both Zipstream and H.265.
• cynical upgrade cycle, i.e. if you buy a H.264 Zipstream camera now and want H.265, you'll have to buy another model of camera when it is released by Axis.

Without having seen any independent test results, it appears that Zipstream is a desirable development which should have helped Axis differentiate its cameras from competing offerings. However I suspect the emerging rollout of H.265 devices may significantly counter or crush the Zipstream advantage. As you say, the real question is when Axis will release H.265? Another question is how good will the new H.265 devices perform from other manufacturers? It's going to be an exciting year!

I saw a Dahua H.265 NVR last week following the announcement of two H.265 NVRs on their website. I haven't seen any Dahua H.265 cameras or heard of any formal announcements yet.

Jon Dillabaugh

Pro Focus LLC | In reply to John Honovich | 03/26/15 01:19pm

Yes, they also have some H.265 NVRs.

Undisclosed #1

03/26/15 05:49am

Normally, a single compression level is applied to an entire video stream.

Won't a camera set to CBR dynamically change the compression level?

John Honovich

IPVM | In reply to Undisclosed #1 | 03/26/15 12:35pm

The key difference is that, with CBR, the same level is applied to the whole scene.

For example, 1080p 30fps camera set to 1Mb/s CBR:

(a) white wall, no motion, quantization for the whole scene/stream might be set to 27

(b) white wall, face appears on half the FoV/camera, quantization for the whole scene/stream might go to 38, even if half of the scene is still a motionless white wall.

By contrast, what Zipstream and other dynamic region compression technologies are claiming is that they would set quantization of say 27 for the part with the face but 38 for the other part of the white wall.

Undisclosed #1

In reply to John Honovich | 03/26/15 04:46pm

Thanks, my bad.

So Zipstream is dynamically changing the q in the spatial dimension and CBR in the temporal.

John Honovich

IPVM | In reply to Undisclosed #1 | 03/26/15 04:47pm

Yes.

johan Schorreel

03/26/15 07:54am

When tested, would it be a good idea to compare it to what Panasonic can achieve with their VIQS ?

John Honovich

IPVM | In reply to johan Schorreel | 03/26/15 12:37pm

We tried VIQS a year or two ago, the bandwidth difference savings was in the range of 10 - 20%, which is fairly negligible. Ergo, we had not pursued or considered it since. If Panasonic believes it has improved significantly, we'd consider testing it again.

johan Schorreel

In reply to John Honovich | 03/26/15 02:41pm

John, I'm struggling to find the test. Can you give me the link ?

John Honovich

IPVM | In reply to johan Schorreel | 03/26/15 02:45pm

We didn't publish it, again because it wasn't a very significant result and Panasonic has not marketed it that hard.

Richard Williams

IPVMU Certified | 03/26/15 05:08pm

Another question would be how this and other compressions impact the recording server CPU. There are differences in CPU usage between Baseline, Main, and High profiles in H.264, so I would think it would be different on Zipstream also. How will this negatively impact a limit on maximum number of recorded cameras on an existing recorder? When designing a system, we now not only have to take into account bandwidth and storage, but compression and profile that will be used on each camera to make sure the recorder being spec'd can handle the load and if required, allow for an x factor of expansion. It keeps getting trickier.

John Honovich

IPVM | In reply to Richard Williams | 03/26/15 05:34pm

We will test client / server side load. From what Axis claims and what we can see, most, if not all of the incremental load is on the camera side.

Steve Mitchell

03/26/15 05:58pm

This is a great strategy and I’m glad Axis is commercializing it. We did quite a bit of work on something similar in our lab and I recall seeing another [VENDOR] demo that clearly showed object segmentation with the background at a different frame rate than the moving foreground objects. Using a different Q-value for background is a clear win since the static portions of the scene should compress quite well. And treating the background/foreground with different GOP structures is much like using a different frame rate—also very effective.

One drawback can be subtle artifacts surrounding the moving foreground objects that don’t quite mesh with the background in all frames. But I feel like that tends to be an acceptable tradeoff in our industry (in contrast to entertainment—which is what a lot of the compression technology is geared towards). To me that opens interesting philosophical doors regarding applications and use cases of surveillance video and the trade offs involved for what might be possible over very low bit rates and to optimize storage—constraints that the entertainment industry doesn’t tend to wrestle with as much.

Of course the math involved in object segmentation and background subtraction is computationally expensive and it happens right behind the imager. And a good implementation requires close coordination with the compression hardware. So a hardware based implementation makes sense.

The relationship of this to H264/H265 can be distinct. My impression is this strategy effectively filters noise out of the static background so you’re sending what is mostly the same picture to the compressor every frame. Thus the resulting bit rates are low. But the magic happens “before” the compression. So compatibility with H264/H265 is preserved in the output. To answer the question of why aren’t they doing this with H265 might be because H265 decoders are still not widely available in our market and simply aren’t ready yet. In other words, this can give big wins even within H264, and the industry isn’t ready for H265 anyway..

John Honovich

IPVM | 03/30/15 11:57pm

Axis has added a teaser 30 second Zipstream demo video:

Undisclosed Manufacturer #2

04/06/15 06:58am

Hi Ethan and John.

As I remember Hikvision also claims a sort of bandwith saving. They call it "Smart ROI" http://www.cips.az/2015/?p=news__read&t=exhibitors&q=5&l=en

Undisclosed Manufacturer #3

04/08/15 02:26pm

This reminds me of something we did back in the day at Cernium, circa 2005. The I-frames were encoded at a lower resolution than the P-frames. If you looked closely, targets were moving within a "bubble" of higher resolution. It worked well but looked weird. Customers that noticed were troubled to learn that every video frame was not a complete picture. While this is normal to us, most of our customers don't understand that. Making it obvious to them was a huge mistake.

Undisclosed Manufacturer #4

04/08/15 02:34pm

Does anyone still have "court admissability" issues with compressed video. I remember when MPEG4 or H.264 came out and some people claimed that it wouldn't be admissable in court because of the compression technology. Of course, those same people said that MJPEG was OK - which is also compressed, just not inter-frame referenced compression.

I would think that those that fought against H.264 would hate the fact that the camera is manipulating the video, possibly leading to court issues...

John Honovich

IPVM | In reply to Undisclosed Manufacturer #4 | 04/08/15 02:36pm

No, and I have never heard of anyone who actually did have _real_ court issues.

Mostly this was fear combined with vendor hype. Remember Avigilon before they dumped JPEG2000?

Undisclosed Manufacturer #3

In reply to John Honovich | 04/08/15 03:03pm

I did find this:

Grant Fredericks, a well-known, well-respected video evidence expert, in a response to my email on this topic, had the following comments:

"...cases concerning the issue of inadmissibility of MPEG video are based on myth, not reality. Although experts have a number of legitimate concerns about MPEG (either MPEG 1, 2, 4, and in all their levels and forms), no one can state that MPEG is an unreliable format" (Goldstein, 2009).

Reference

Goldstein, E. (2009, April 30). The admissibility of MPEG encoded video. Canadian Security.

The admissibility of MPEG encoded video.