Higher video resolutions are something that the surveillance sector is having to get used too. After many years where resolution was limited by PAL standards, the freedom of networked-based solutions has seen greater detail become the order of the day. In many applications HD video is considered to be the de facto standard, and demand for 4K UHD video is growing. This necessitates an increased focus on bandwidth management, and a number of manufacturers are now introducing dynamic video encoding to reduce network load. Benchmark considers the performance of two leading options: Zipstream from Axis Communications and Wisestream from Hanwha Techwin.
The growth in dynamic encoding technologies is a by-product of the leap forwards in processing capabilites which has also seen on-camera apps become a reality. Many of today’s higher resolution cameras have spare processing power, and some innovative manufacturers are using it to address the bandwidth issue by implementing dynamic encoding.
As with any processing functionality, manufacturers have different names for the process. Axis Communications calls its version Zipstream, Hanwha Techwin’s is WiseStream, and there are a host of other names alluding to smart and intelligent streaming.
Whilst there will be differences in each vendor’s implementation, the concept of dynamic encoding allows a forensic engine to examine captured video streams. The goal is to identify static parts of the scene and those with a high degree of change and/or detail. The process is not too dissimilar to that carried out by predictive compression engines.
Areas of the image with low detail and no changes are considered to be less important for compression coding. After all, nothing much is happening. As a result, higher levels of compression can be applied. What’s important is that data isn’t removed or compressed to the point of being unusable. The quality will still be acceptable and for proof of continuity it will be fine.
Where change is detected and high levels of detail exist, the algorithm sees the video data as important and therefore increases the available bitrate. This ensures that important parts of the video retain detail and sharpness.
As a result, video containing faces, vehicle number plates, identifying elements such as tattoos or logoed clothing, etc., will be preserved and streamed with a higher bitrate. However, video of an empty car park can be streamed with far less load on the bandwidth.
Because the decoding is dynamic, it reacts to the scene content. If someone walks into the car park, or if a vehicle enters or leaves, the bitrate is increased.
The dynamic encoding engine from Axis Communications has been upgraded recently, and it is included as a standard feature on a wide range of the manufacturer’s models. Whilst the core of the technology is dynamic compression encoding, the latest version (which can be added to many devices via a firmware upgrade) also includes dynamic frame rate, but that was not looked at for this assessment.
The manufacturer claims that Zipstream, which is an H.264 implementation, is capable of reducing bitrate requirements by around 50 per cent or more for many common surveillance use cases.
Zipstream is configured via the camera’s video stream menu, and the page for the dynamic encoding is straightforward and very installer friendly. Zipstream has four settings: Off, Low, Medium and High. Axis qualifies the impact on quality of the various settings. Low is designated as no visible effect in most scenes, Medium as less detail and less noise in some parts of the scene and High as less detail and less noise in many parts of the scene.
Zipstream also includes dynamic GOP. This allows the GOP to be varied from 30 to a defined rate of up to 300. This feature is flagged as Advanced, and the likely intention is to dissuade anyone from using it who doesn’t understand the possible impact of increasing GOP. Effectively, a GOP of 300 would see one full reference frame of video every 10 seconds. This might not have much impact in a static scene, and the dynamic nature of the function should reduce the GOP when activity occurs, but it’s a feature to use with consideration.
To assess the performance of Zipstream, the camera was configured to deliver a quality-based HD1080p stream in real time. Bitrate was variable, and compression was High Profile H.264. The compression was set to level 10 (out of 100) which is minimal, and GOP was set to 30. We used two viewed scenes: Scene 1 was static with occasional motion and Scene 2 included passing traffic.
With Zipstream disabled, image quality was very good with sharp detail throughout the coverage in both scenes. The average bitrate for Scene 1 was 27.5Mbps, and for Scene 2 it was 39Mbps.
Setting Zipstream to its Low setting didn’t produce any unwanted effects in either of the viewed scenes. Detail was still sharp, motion smooth and despite knowing what to look for, there didn’t appear to be any signs of the dynamic encoding. However, the impact on required bitrate was noticeable.
In Scene 1 the average bitrate fell to 18Mbps and in the busier Scene 2 it fell to 31Mbps. This represents a 34 per cent decrease for Scene 1 and a 25 per cent decrease for Scene 2.
Moving Zipstream to the Medium level did show some lowering of detail in tonally static parts of the scene, but you had to actively look for this. Bitrate did reduce further, but by a less significant amount. Scene 1 saw average bitrates of 14.5Mbps and requirements for Scene 2 fell to 28Mbps. This represented decreases of 47 per cent and 28 per cent respectively.
Interestingly, with Zipstream set to High it was obvious that the dynamic encoding was affecting static parts of the scene, but the reduction in bitrate over the Medium setting was minimal, around 1-2Mbps. Our opinion was that the detail trade off for such a small gain wasn’t worthwhile, and we’d expect only those with a very tight bandwidth budget would use the High setting. That said, our approach was always to try and deliver the highest quality HD1080p stream possible!
Using dynamic GOP did not have a significant effect on bitrate, unless there was absolutely no motion in the scene. However, it did give the image a slightly unreal look with depth of field suffering a little.
Low light did affect the performance, with the reductions being less dramatic in twilight, before the device switched to night mode. However, the real killer for the dynamic encoding was rain. This was not unexpected.
Scene 1 in good weather conditions generated a stream of 27.5Mbps without Zipstream, and this fell to 14.5Mbps (47 per cent) with Zipstream set at Medium. With rain, Scene 1’s bitrate without Zipstream increased to 38Mbps, and with Zipstream at a Medium setting reduced to 31Mbps (18 per cent). In Scene 2, the decrease in good weather was 28 per cent, but rain altered this to a mere 7 per cent. Of course, rain will cause a spike in bitrate for any camera using predictive compression.
The dynamic encoding engine from Hanwha Techwin is WiseStream, and it is included as a standard feature on the latest range of the company’s cameras. Hanwha Techwin claims that WiseStream can reduce bitrate requirements by up to 75 per cent.
WiseStream is compatible with both H.264 and H.265 compression, but has been optimised to provide greater efficiency using H.265, according to the manufacturer. Like Zipstream, WiseStream combines dynamic encoding with a dynamic GOV.
To assess WiseStream we used the PNV-9080RP camera. As with the Zipstream assessment, or focus was on a high quality image. The camera was configured to deliver a 3840 x 2160 pixel resolution image at the camera’s maximum rate for that resolution of 20 frames per second. Compression used was H.265.
The configurations for Dynamic GOV and WiseStream reside in two different areas within the menu. Dynamic GOV is a part of the video profile, which WiseStream has its own menu page. This consists of a single drop-down menu. There is also a feature to create a region of interest which can reduce the portion of the image which is of importance, but we did not use this for the assessment.
As with the previous part of the assessment, we viewed two scenes to give a distinction between a largely static scene and one with more occasional traffic.
With WiseStream and Dynamic GOV disabled, the bitrate requirement for Scene 1 was an average of 17Mbps, while for the busier Scene 2 it was 23Mbps. The image contained a good degree of detail, motion was as smoothy as it can be at 20fps, colour rendition was good and the quality was consistent across the viewed scene.
With WiseStream enabled at the Low setting, there was not visible difference in image quality. However, the bitrate requirement fell to 12Mbps; a reduction of 29 per cent. Switching to the Medium setting did increase the signs of the dynamic encoding, albeit in the tonally bland and static portions of the image. That said, you had to be purposefully looking for it.
Average bitrate fell to 8Mbps, representing a reduction of 52 per cent. At the High setting bit rate fell to an average of 6Mbps (a reduction of just under 65 per cent), although the impact on image quality remained very much as it was with the Medium setting.
Scene 2 did see smaller decreases, with the Low level delivering an average bitrate of 19Mbps, Medium giving 16Mbps and High delivering 13Mbps. These represented reductions of 17 per cent, 30 per cent and 43 per cent respectively.
The savings are increased by switching the dynamic GOV on. This can be adjusted from 20 (due to the frame rate) up to 320.
This means that if the scene is static, I-Frames will only be captured every 16 seconds. As is the case with Zipstream’s dynamic GOP (the difference between GOP and GOV is purely down to semantics – group of pictures and group of video), there can be some visible impact on the image if the GOV is extended, but it could be argued that the video affected doesn’t contain critical information.
Savings are in order of 5-15 per cent, as any discernable activity will switch the GOV back to its minimum setting.
As with Zipstream, the savings achieved by WiseStream fall off a cliff when it rains. Indeed, the impact on both the dynamic encoding engines is similar, with reductions falling to between 5 and 15 per cent. Often the reduced figures were still higher than those achieved with the dynamic encoding disabled in good weather conditions!
Dynamic encoding certainly can deliver significant bandwidth savings without having a dramatic impact on final image quality. That much is clear. However, installers and integrators must also realise that the technology is not a silver bullet where bandwidth budgets are tight.
The savings available from dynamic encoding are variable, and no doubt the figures quoted in this assessment might rise or fall depending upon the viewed scene. It is clear that the busier the scene, the less efficient the technology will be, simply because it works by reducing detail in areas with little or no visual interest.
Dynamic encoding will also be affected by various environmental conditions, in much the same way that predictive compression is affected by them. Rain, snow and low light will all push figures higher, and in some instances this will make the effectiveness negligible. However, the technology is included as standard, so it’s not like you’ll be making an investment to achieve the benefits.
Equally, as manufacturers enhance noise reduction and low light performance, so the cameras with dynamic encoding enabled will become more efficient.