Test: H.264 I vs P Frame Impact

Author: Ethan Ace, Published on Oct 02, 2013

Codecs like H.264 reduce bandwidth by only sending full frames every so often, mixing them with partial frames only capturing changes in between the full ones. They are called 'I' frames because they are the initial / full frames, followed by 'P', or predictive frames.*

Note: if you are not familiar with codecs, please read our Surveillance CODEC Guide before continuing.

I Frame Questions

Since I frames require much more bandwidth than P frames (frequently 10 or 20x more), some will argue that reducing the rate of I frames will reduce overall bandwidth significantly. For instance, instead of having an I frame each second, reduce it to 1 every 5 seconds.

On the other hand, some will argue that reducing I frames can result in quality problems because it can be harder for the processor to continue to faithfully update and represent the image if it has changed significantly since the last I frame.

We seek to answer these two questions:

  • How much bandwidth savings can you achieve by reducing the I frame interval?
  • How much quality degradation can occur by reducing the I frame interval?

The Tests Conducted

In order to answer these questions, we used five 720p cameras at various price points and performance levels:

  • Avigilon H3 1MP
  • Axis M1114
  • Axis Q1604
  • Bosch NBN-733V
  • Dahua HF3101

We aimed these cameras at a toy train set to create consistent motion, and varied I-frame levels from a default of one per second to as high as five and as low as one every four seconds.

*Some versions of H.264 also support 'B' or bidirectionally predictive frames, but these are less common in surveillance cameras and therefore excluded from this study.

****** **** *.*** ****** ********* ** **** ******* **** ****** every ** *****, ****** **** **** ******* ****** **** ********* changes ** ******* *** **** ****. **** *** ****** '*' frames ******* **** *** *** *******/ **** ******, ******** ** '*', ** ****************.*

****: ** *** *** *** ******** **********, ****** **** *************** ***** *********** **********.

I ***** *********

***** * ****** ******* **** **** ********* **** * ****** (frequently ** ** *** ****), **** **** ***** **** ******** the **** ** * ****** **** ****** ******* ********* *************. For ********, ******* ** ****** ** * ***** **** ******, reduce ** ** * ***** * *******.

** *** ***** ****, **** **** ***** **** ******** * frames *** ****** ** ******* ******** ******* ** *** ** harder *** *** ********* ** ******** ** ********** ****** *** represent *** ***** ** ** *** ******* ************* ***** *** last * *****.

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

  • *** **** ********* ******* *** *** ******* ** ******** *** I ***** ********?
  • *** **** ******* *********** *** ***** ** ******** *** * frame ********?

The ***** *********

** ***** ** ****** ***** *********, ** **** **** **** cameras ** ******* ***** ****** *** *********** ******:

  • ******** ** ***
  • **** *****
  • **** *****
  • ***** ***-****
  • ***** ******

** ***** ***** ******* ** * *** ***** *** ** create ********** ******, *** ****** *-***** ****** **** * ******* of *** *** ****** ** ** **** ** **** *** as *** ** *** ***** **** *******.

***** ******** ** *.*** **** ******* '*' ** *************** ********** frames, *** ***** *** **** ****** ** ************ ******* *** therefore ******** **** **** *****.

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

Key ********

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

  • ********** *-***** ********* ******* ** ******* ***** ******* ***********, *** a ***** ******** ** ********* ***********. ********* ** **-**% **** common **** ******** *-****** **** * ** * *** ******, while ********* ** **-**% **** ******** **** ****** **** * to *.
  • ********** *** ****** ** *-****** *** ****** ******* ** * significant ********* ** ***** ********** ****** *******, *** * **** ***** **** ** *********. ****** **** one *-***** *** ****** ** *** ***** *** ******* ********* bandwidth ** *-**%. **-**% ********** **** **** **** ******** *-****** to *** ***** **** *******.
  • ***** ********* ********* ********* ** *-***** ******** *********, *** **** versa, *** ***** ****** ** **** ****** ****** ****** ************* and ****** **** *** **** ************. 
  • **** ******'* *-***** **** ******** ********** ********** ** *-***** ********.
  • *** ***** ***-**** *** *** **** ****** ** **** **** that *** *** ******** ********* **** *** ****** ** *-******. Instead, *** ****** *********************, ******* *** ****** **** ********. *******, ******** *** ****** of *-****** ******** ** ********* ********** ******* ** ***** *******.
  • *** ***** ****** *** *** ***** *** **** **** *** I-frame *** ******, ****** ****** ***** **** ********.

Image ******* ******

** **** *****, ** **** *-***** ********'* ****** ** ***** quality *** ** ******* ******* *************. ********* ****** *******:

  • * *-***** *** ******
  • *.* *-****** *** ******
  • * *-****** *** ******
  • * *-***** ***** * *******
  • * *-***** ***** * *******

***** ****** **** *** ******* ********* ** ***** ******* ** the ****** ** *-****** ** *******, ** **** ** *** minimal ***** **** **********.

Bandwidth ****** 

**** ********** *** **** ******* * ******, ********* ******** ****** all *******, *** ********, ** ***** ** **** *****:

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

** ********, ********** *** ****** ** * ****** ** **** there **** ******** **** ** * ******, ************* ********* *********:

 

**** ***** ***** *** *********** ** ********* **** ********** **** one *-***** *** ****** ** *.* *** *. ***** **** Dahua *** *** ***** *** ****** *****, ** *** ******** from ***** *****. ****, *** *** ***** ***** *****'* *********/*********** performance.

P-frame *********/****

*** ****** *****, ***** *************, *********** *** ******:

  • *****, **** *** ********** ******** ** *-****** ** **** ***** (moving **** **** ** *****). *-****** *** ***** ** *** taller **** *****, ***** *-****** *** *****/***.
  • ******, *** *-***** **** ** **** ****** ******* ********** ********** regardless ** *-***** ********. **** *** **** ** **** *** relatively *** ****** ***** ****, *** ** * **** ****** scene ******* **** * ***** *****.

Bosch ***-*** *********** 

****** *** ***** ******* ** *** ****, *** ***-**** *** not ******** ********* **** *-***** ******** *** *********, ******* ******* approximately ****** ****** * **/*. ******** **** ****** ** * stream ********, ** ********** **** *** ****** *********** *** *** additional *-****** ************ ************. ******* ************ *** ** *******:

  • * *-***** *** ******: ~** 
  • *.* *-****** *** ******: ~**
  • * *-****** *** ******: ~**

**** **** ************ ******** ********* **** ****** **** *** *-***** per ****** ** *** ***** *** ** **** *******, ****** around **.

Comments (14)

"********** *** ****** ** *-****** *** ****** ******* ** * significant ********* ** ***** ********** ****** *******, *** * **** ***** **** ** *********"

*** **** * ********** *** ***** ******* ***** ******* ********* on *** ******** ** *-****** ? ** **** * ***** is ******** *** ********** **** *** **** * ** * frame, *** **** ******** *** *-****** ****** ***** ******* ? Are *** ****** **** ********** ****** *** ******* ? *** you **** ****** ** ********** ****** ******** ******* ******** *** Constant ******* ******** ?

* ****** *** ********* ***** ******* **** * ****** ***** they're ***** ***** ****** ** *********, ** ******* ** ***** a **** ******* ** * ****** ***. ** ******* *** artifacts *** *** ****** ****** *******, ***** *** ********** ** what ** ******* ** **** * ***** (*** ****** ******(*)) is *********. ** ** **** ***'** *** *** ******** ***** grow ****** *** ******.

** ** ********** ****** *** *******, ** **** ** *** and ***. ***** ***, ********* **** ******** *** ******** ** you ***** *** ***** * ***** *********. ***** ***, *** camera ****** ** **** ** *** * ******** ***** ************ level ** *** ****** * ******, *** ***'** ***** *** artifacts **** * ******.

**** **** ** ****, **** ****** ** *** ** **** straight ;) ** *** ********* **** ***** ** **** *** span ** *** *** ******* **** * ***** ** '**********' what ******* **** *** * *****. *** '**********' ** **** encoded/compressed ***** ** ****** **** ********** *********. ??

**** ** *******. ** ** *** **** * *** ****** through * *****, *** *******, *** ******* **** *** *** car ** * ******. ** ** ***** *** ****** ***** of *** ********** **** *** ******** *****, *** ********* **** the ******* ***** (*** ***, ** **** ** ***** *** car **** ** **, ***** ** *** ***** *****). **** will ***** ** *******, ******* ******* *** ** ******. *** since ********** * ****** ********* *** ******** * ***** *** changes, ** **** ******** ***** ***** *** **** * ***** is ****, ***** *** *** *** ** * "********" ** the *****.

*** **. ***** ****** ******* *** ******* ***** **** ********** before. *** *** ***** ** ********* **** ***** **** *-****** set ** * *** *********.

* ****** * ***** ********* ** ****** *** *** *******, so ********* *** ***** *** *** * ******* *******. *******, some ******* ** ***** *** ** '*****' **** ****** * frame *********.

**** ****** *** ****** *** ** ********* ***** ** *** report **** ***** **** ** ***** **** **** ****** *** comparisons. ***** ******* *** ****** ** ***** ***** *** **** for ***** ************?

** *** *** **** ******* ** ******* *** *** **** flexible ************* ** ******/***** * ***** *********. *'** *** * note ***** *********, ******, ******.

******. ** **** ***** ******* *** *** ********** ** ***** when ******* *** *******.

**** ***** ******* *** *-****** ********* ** ***** ******* *** bandwidth?

**** ****** ** *********** * ****.

*********** ******, *****. *** *** ******** *** ***** ** **** experiment? *'* ********* *** **** ********* (** ***** ** *** usage %) ********** *** ******** *** *-***** ******** ***** ****.

** ****'* ********* ******** **. *** ******* ** ***** *** in ***, **'* ******** ** ****. *** ********** ******* * I-frame *** ****** *** *, *** *******, ** *** ******* CPU *****, ** ****. ******** **** * *** ****** ** 1 ***** * ******* ********* ** ** *** ** ****. Those *** *** **** ******* ********, *** *** ******* *** practically **********.

*********** ********.

* ***** **** ** *** ***-********* *********** ** ****** ******* - **, ** ******. *.*** ********** - ***** ** *** typical ******** **** ********** ** ***** *******. **** ** ******* worse, ** **'* *** ******** **** ******* *** ****** **********. And, **** ***** ***** ***** *** **** **** ****** ******** hours *** ***** *******-******** ************, ** *** ***'* *** *** the ********* - ****** ****** ** *** *** ****. ***, try ** ******* * ******/***** *** *******.

* *** * *** ** '*************' **** *********** *** ***** rates, *** ********** *** *** *** ******** ******** ***** *** user ********* ** *** *** ***** ******* ******** (****) ** (I-frame ********, **
*** ****, ** ******** *** ****** ************ ***** **). ** particular, *********'* ***-********** ********* *** *** **** ** * ******* of * ***** *** ******. **, ** *** ***** ***** - ***, * *** - ****, ****'* ****, *** **** I-frame *** ******, *** * *** ** **** *** **** image. ***** *** ****, **'* ***** **** ******* *** ****** won't *** ** ******!

***** *** ********* (***** *******, *****, *** *******) ** ****** keep *** ****** ******** (** *********) ****** ** *** ***** steam *********. *'** **** ****' '*******' *** **** ** ** (1 *-***** *** ** ******, ** * ********** **) ** 1 *** *** *** ***** ********** *******, *** ****** **** little ****** ****** - ** ********** ******* ****** *******,*** *******. The *********** ** *** ** *** '******' ***** ** *** image, ** *** ****** ** ******** - *** ****** ******* - ** ****** **** ******* ** *** *-*****, **** ** 1 ***. ** ****, * ***** *** * ******** '********' about *** ***********, *** ** ******* ** ** '****-*********' ** subsequent *-******. * *** **** ** *** *-* *** ******* at * *** **** ** ** ******* ** ***-****** *** static ******, *** ******* ** ***-****** ** * *** ** motion *******. ****** ********** - ****** ** ***** **** * T1. ********* *** ********* ***** *** ********* (***** *****), *** other ******** ******** **** *** *******, ** * ***** *** the '**** *****' ** ********* ***********, *** *** *** ****** 1.5Mbs.

******, * **** ** ***** ******* *** ***** ******* **** impacted ** *** ********* (***** *** ****** ************ *******/****** **********) which ********* * ********* *** ** ******** *** * *** bandwidth ***********, (*** **** ****) *** **** *** **** ********** in ** ********.

****-**** **** ** **** **** **** *** *** ** ******** pulling ******* ** *** **** (*-***** *****) *** ***** ** is....

***** ** * *** ** **** *********** **** **** ***** be ******* **** ******* * ****** *** **. * ***** it *********** **** ******** ***-**** ********* *** *** ******** **** *** ****** ** I ******, *** ******** *** ************.*'** ***** ****** **** * ***** ******, *** * ****** if ***** *** ***** ******* **** ******* **** **** ********.

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 on Bandwidth

Winter 2018 Camera Course on Jan 18, 2018
Learn video surveillance and get certified. Register now. Save $50 on the course, ending this Thursday the 18th, plus get access to 2 class times...
VSaaS Usage Statistics 2018 on Jan 18, 2018
VSaaS has been a 'next big thing' for more than a decade. The prospect of managing, storing and streaming video from the cloud rather than...
Multicasting Surveillance Tutorial on Jan 04, 2018
Network bandwidth can be a concern for some surveillance systems. While improvements in video codecs, such as smart codecs for H.264 and H.265,...
2018 IP Networking Book Released on Jan 03, 2018
The new IP Networking Book 2018 is a 228-page in-depth guide that teaches you how IT and telecom technologies impact modern security...
QoS for Video Surveillance on Dec 29, 2017
Along with VLANs, QoS is one of the most misunderstood topics in IP surveillance networks. Many purported "experts" claim it is required in any...
Hikvision NVR Load Testing on Dec 14, 2017
IPVM members recently debated Hikvision NVR's performance under load in Hikvision 30+ Cameras On NVR - Apps And Client Really Slow Down And CPU...
D-Link ONVIF Switch Tested on Dec 04, 2017
D-Link's surveillance switches claim to "enhance ease of use and streamline management" for network administrators, with simplified UIs and...
IP Camera Manufacturer Compression Comparison on Nov 27, 2017
Despite the use of standards-based video compression (H.264/H.265), our tests show that default image quality settings for different manufacturers...
Camera Multi-Streaming Usage on Nov 22, 2017
IP cameras typically support multiple streams, allowing a single camera to transmit multiple streams at different resolutions, frame rates and even...
CBR vs VBR vs MBR - Surveillance Streaming on Nov 21, 2017
How you stream video has a major impact on quality and bandwidth. And it is not simply CODEC choice (e.g., H.264 vs H.265). Regardless of the...

Most Recent Industry Reports

Chinese Government Hikvision Surveillance System On US Government Network on Jan 18, 2018
Hikvision, the Chinese government-owned manufacturer, has publicly claimed that their products are running on a US government network. Moreover,...
Camera Course - Last Day - Save $50 on Jan 18, 2018
Today is the last day to save $50 - register now. Learn video surveillance and get certified. Save $50 on the course, ending this Thursday the...
VSaaS Usage Statistics 2018 on Jan 18, 2018
VSaaS has been a 'next big thing' for more than a decade. The prospect of managing, storing and streaming video from the cloud rather than...
Vivint Streety Video Strengthens Door Knocking on Jan 17, 2018
Vivint is famous (or infamous depending on your perspective) for mastering large scale door to door selling. The company has skyrocketed from a...
Axis: "It’s A Question Of Trust And Who You Want To Be Associated With" on Jan 17, 2018
Who do you trust? Who do you want to be associated with? Axis is raising hard questions to start 2018. In this note, we examine these questions,...
Software House Vulnerability Allows Inside Attacker To Open Doors on Jan 17, 2018
A vulnerability in Software House IP-ACM modules allows an attacker to potentially unlock doors, or perform other actions, on affected systems....
'Defiant' Hikvision 'Strikes Back' At WSJ And US on Jan 16, 2018
The fight is on. Hikvision and their owner, the Chinese government, 'strikes back' against the Wall Street Journal and US politicians raising...
The 2018 Surveillance Industry Guide on Jan 16, 2018
The 300 page, 2018 Video Surveillance Industry Guide, covering the key events and the future of the video surveillance market, is now available,...
Edward Snowden Haven App Tested on Jan 16, 2018
Global coverage followed the December 2017 announcement that Edward Snowden was leading a team developing Haven, an app "that leverages on-device...
This High Schooler Is Excited About His Future Security Career on Jan 15, 2018
A common lament is that smart, young people have little interest in surveillance systems. In fact, discussions like Should Talented Young People...

The world's leading video surveillance information source, IPVM provides the best reporting, testing and training for 10,000+ members globally. Dedicated to independent and objective information, we uniquely refuse any and all advertisements, sponsorship and consulting from manufacturers.

About | FAQ | Contact