Intelligent video functionality promised so much to the security sector, but there is still some debate over what it has achieved. At the higher end, there are a mix of success stories and cautionary tales. In the mainstream, intelligent functionality is increasingly included in cameras. Benchmark looked at some options to see what they deliver.
Click to view larger image
[dropcap]I[/dropcap]t is often referred to as IVA (intelligent video analytics), smart video, VCA (video content analysis) or intelligent VMD. However, putting the terminology aside, the process of analysing information gathered from video footage to initiate alerts, actions or system status changes is not new. Despite what some like to intimate, the concept pre-dates the introduction of digital video.
As computational resources have increased and processing costs have fallen, so the possibilities of smart analysis of one type or another have moved into the mainstream.
An increasing number of manufacturers offer some form of analytics as an integral feature of cameras. Obviously, these variants of the technology deal with a single video stream, typically with a limited number of rules or options. However, often the best use for analytics is dealing with simple and clearly defined situations, so this approach does deliver benefits!
It is unlikely that we will see a dramatic increase in the capabilities of on-board intelligence in the near future, because the cost of processing could be prohibitive in competitive markets such as the camera sector. Having said that, so long as your expectations aren’t based upon a fully-blown analytics engine, then there will be more than enough capability to deliver a wide range of value-added benefits for most applications.
This test predominantly focuses on the intelligent capabilities of the products, but also considers whether the overall video quality and camera performance is impacted upon by the additional processing load of the analytics. Video quality was tested, but is not reported upon in detail unless there is a level of performance inconsistent with the product’s specifications.
Click to view larger image
Axis Communications – Q1614
The Q1614 is a box-type HD720p camera with on-board intelligent video. It uses a 1/3 inch CMOS HD sensor to deliver HD720p streams at frame rates of up to 50ips, allowing cleaner images where fast motion exists. The camera also supports 1MP and a number of standard definition resolutions. The day/night camera is supplied with a 2.8-8mm infrared corrected megapixel lens.
The main stream algorithm is H.264, and Motion-JPEG is also supported. Multiple streaming is possible, and these can be separately configured.
Sensitivity is quoted as 0.2 lux for a 50ips stream, or 0.1 lux for a 25ips stream. Day/night operation is achieved using an infrared cut filter. Switching can be activated manually, by integral light sensing technology, or via a contact which is configured in the Event menu: more about that in a moment, as this menu is integral to the intelligent video functionality.
Other video-based features of the Q1614 include AWB, BLC, WDR, exposure control and privacy masking. Two-way audio, data communications and edge recording are supported. Power is PoE or 12V DC/24V AC.
At first glance, you might wonder where the ‘intelligent video’ menu is, but it is split between several menus. In truth, what you get is video motion detection, but with discriminations and the ability to add conditions to events. It is, therefore, quick to set up, but some of the detection tasks we wanted to achieve could not be realised without the addition of secondary devices such as detectors. Whilst this isn’t a negative, it does exclude the ability to use contact-based day/night switching in some circumstances, as you’re using the I/Os for other purposes.
There are three parts to the intelligent video: tamper protection, video motion detection and audio detection.
Tamper protection covers the standard video-based tamper alerts. These include defocusing, significant change in the viewed scene, masking, etc.. There is a facility to set a time window which the condition has to exist for in order to generate an alarm. There is also an option to generate an event if the image goes dark; this could be down to lighting issues, spraying or other causes. Finally, integral shock sensing can be enabled.
Motion detection can be set up in multiple alarm zones, and the camera allows discriminations to be set for object size, history and sensitivity. These configurations can differ for each zone. Setting the parameters is simplified as the menu page uses a clear visual histogram to show how motion is sensed with regard to the current alarm threshold.
Audio detection allows an event to be created by audio volume. Again, a histogram is used to help set the alarm threshold.
Whilst these functions might be more than some VMD-enabled devices offer, the use of the Events menu does also allow additional functionality to be added, with regard to the creation of more complex events.
For example, when creating an Event, it is possible to require separate activations to be grouped within a specified time window. For example, an event might only be considered an alarm if a motion detection incident in zone 1 is followed, within 30 seconds, by a contact-based alert, and then by another motion detection event in zone 2.
Motion detection, audio detection, tamper alerts, contact closures and other system events can all be used to create scenarios.
Once an event is established, then actions can be applied to it. These can range from enabled day/night switching or applying WDR, through to sending alerts, streaming video, generating alarms or recording footage.
Performance-wise, image clarity is good, with high detail, and there are no issues with regard to video quality.
The intelligent video configuration is quick and simple, but you don’t get the flexibility on offer from some other devices. The functions work well, but are more basic than IVA; you have to make sure that your expectations (and those of your customer) are more aligned towards flexible VMD that works!
Some our test violations could not be specifically detected, because they relied on behaviours rather than motion alone.
Bosch Security – NBN-733V-IP
The NBN-733V-IP is an HD720p box-type unit with on-board video analytics. It uses a 1/3 inch CMOS HD sensor to deliver HD720p streams at frame rates of up to 50ips. This higher frame rate delivers clearer images in scenes with fast moving objects. The camera also supports other standard definition resolutions with various aspect ratios.
The main stream processing algorithm is H.264. Motion-JPEG is also supported. The camera can handle multiple streams, which are independently configurable.
The unit carries a ‘Starlight’ designation, which is claimed to combine advanced sensor technology with noise reduction algorithms to deliver colour performance in minimal light levels. Sensitivity is quoted as 0.017 lux for a 30IRE image.
Day/night operation is achieved using an infrared cut filter. Switching can be activated manually, by integral light sensing technology, or via a contact.
Other video-based features of the NBN-733V-IP include AWB, intelligent BLC, contrast enhancement, dynamic range of 84dB and privacy masking, with up to four zones being programmable. Two-way audio, data communications and local recording are supported. Power is PoE or 12V DC/24V AC.
Regions of interest (RoIs) can be established as separate streams, and the IVA engine allows tracking within the RoIs.
The IVA can be configured in different ways. There is ‘Silent Motion+’ which only generates metadata. There is also an ‘Event Triggered’ option to deliver verified IVA performance, allowing sensor-based activations, for example, to trigger IVA functionality by initiating a Profile.
There are three options: Motion+, IVA and IVA Flow. Motion+ is advanced motion detection with a few additional configurations. These are detection sensitivity, minimum object size and debounce time. It also has a selectable aggregation time to ensure that multiple activations are not triggered by a single event.
IVA works on the basis of objects being detected within the scene, and the Profile applies discriminations to those objects. IVA Flow differs in that it looks for the ‘flow’ of objects within the detection zone.
IVA rules include: object detection with discrimination for speed, size, shape, direction and colour; line crossing; loitering; change in condition of an object within a given time window (using the same discriminations as object detection); route following to identify deviations; object left; object removed; entering a zone; leaving a zone; similarity to a previously defined object. A ‘head detection’ filter can also be used to identify events which include a human presence. Tamper protection is also included.
Installation is straightforward, and configurations are carried out via the Bosch configuration manager. Some configurations can be changed via a browser, but not the IVA functionality. Image clarity is very good, and detail is high.
The IVA configuration isn’t difficult, but in order to get the best level of performance it does take a little time. Each rule has a number of discriminations which can be applied, and the settings are simplified through the use of a Task Wizard.
Once set up, the IVA works well, and the degree of flexibility does ensure that most applications can be catered for. Some settings needed a few tweaks in the early stages of testing, but after that the camera was stable!
The camera detected all of the test violations which included line crossing, entering a secure area, directional detection of vehicles and object missing. Accuracy was high, even during periods when there was also a lot of permitted activity in the viewed scene.
The Head Detection feature works, but you have to be realistic about expectations in some scenes!
Panasonic – WV-SFV631L
The WV-SFV631L is an HD1080p static dome camera. As standard, the camera is only equipped with basic VMD, along with some of the other Panasonic smart features such as face detection to trigger dynamic adjustments, and variable image quality in selected areas. In order to access what the manufacturer refers to as i-VMD, an Extension Software package must also be purchased.
The camera uses a 1/3 inch MOS HD sensor to deliver HD1080p and HD720p streams at frame rates of up to 50ips. This higher frame rate does require certain functions such as Super Dynamic Range to be disabled. The rate can also be affected by the i-VMD functionality.
The image stream processing algorithm is H.264 or JPEG. Multiple streaming is supported, using the Uniphier platform.
The unit has a claimed sensitivity of 0.04 lux. The camera features integral infrared illuminators for night time viewing. Day/night operation uses an infrared cut filter.
Other video-based features include Super Dynamic Range, Adaptive Black Stretch, privacy masking, audio, data communications and edge recording. Power is PoE or 12V DC/24V AC.
The camera also has variable image quality in selected portions of the image. The VIQ feature is similar to regions of interest.
During testing we had an issue with the supplied software. In an attempt to cure this, a firmware upgrade was carried out, and this effectively removed the extension software!
The viewing element of the Panasonic software is blocked by the latest versions of Internet Explorer. Whilst many cameras using Active X drivers will function with IE10 and IE11, there are a number that will not, Panasonic’s included. Many manufacturers are aware of the issue, but few have updated manuals, added addendums or posted technical bulletins on websites! The cure for the problem is simple; the browser needs to have the camera’s IP address added to a list of sites which use Compatibility Mode.
Logging into the camera is achieved via a Simple IP utility; this worked well and allowed quick changes to be made to the network configuration. Once logged in, the various software elements load, and you’ll then need to specify the device to display in Compatibility Mode if setting up with Internet Explorer 10 or 11.
Access to the i-VMD menu is via the Alarms menu. There is a radio button to select either standard VMD or i-VMD. Below this is a link to the set-up pages. Don’t be surprised if you select i-VMD and then go the settings page to find it’s all a bit basic; you need to select the i-VMD radio button, then Set the changes, then use the link!
There are two main settings screens in the i-VMD menu. These are for detection mode/area, and depth. To our way of thinking, the depth configuration should be completed first. This allows the i-VMD to understand perspective. This is simple to achieve, by marking the size of an object when in the foreground, and again when in the background.
The i-VMD feature can have two ‘detection programs’, and each of these can include up to eight alarm zones. It is worth noting that the i-VMD functions do have a slight impact on the performance of the camera, so common sense needs to be applied. The different programs can then be scheduled.
Setting the detection area and mode is also relatively straightforward. There are a number of drawing tools – line, rectangular zone or mask and polygonal zone or mask – and these allow detection zones or virtual lines to be created. The type of alarm event is then selected from a drop-down menu, and relevant additional criteria can be adjusted to further define events. The Panasonic alarm protocol allows notifications for each individual detection zone if required.
Finally, there is a menu to allow enhanced tweaks of the general i-VMD parameters, but this cannot be accessed from the camera’s menus. Instead a separate URL needs to be used.
Once configured, the i-VMD functionality works well, and careful combinations of detection zones, masked areas and directional discriminations do allow relatively complex scenarios to be created which are free from nuisance activations.
The choices of event types meant that all of our test scenarios – line crossing, entering a secure area, directional detection of vehicles and object missing – could all be achieved with a high degree of reliability. Of these, only the object missing rule required a bit of tweaking to keep it stable.
Video quality was good, although if you are using a lot of i-VMD features you will see slight signs of the heavy processing workload.
Riva – RC3502HD-5311
The RC3502HD-5311 is an HD1080p static dome camera equipped as standard with core intelligent video analysis functionality. The IVA, powered by VCA Technology’s analytics engine, includes intelligent video motion detection and tamper protection as standard.
Further VCA filters are optionally available at an extra cost. These include intrusion detection, people counting and object detection packages, and allow application specific scenarios to be created. As these packages are additional purchases, and were not supplied, they are beyond the scope of this test.
The camera offers H.264 and Motion-JPEG compression, and features include dual streaming, edge recording, audio support and digital WDR.
The camera uses a 1/2.7 inch CMOS sensor to deliver HD1080p and HD720p streams, plus other standard definition resolutions, at frame rates of up to 25ips. It is important to note that if an HD1080p stream is used while the VCA functionality is enabled – which will be likely – then the maximum frame rate drops to 15ips. With a HD720p resolution, real-time streaming can be preserved. The image stream processing algorithm is H.264 or Motion JPEG, and maximum bit-rate is 6Mbps.
The unit has a claimed sensitivity of 0.5 lux. The camera features a 3.0-9.0mm motorised lens with one-touch autofocus.
Other video-based features include sens-up, noise reduction, composite output for set-up, BLC, privacy masking, motion detection, two-way audio, edge storage via MicroSD cards and alarm input and outputs. The camera is compatible with PoE, or traditional 12V DC low power can be used.
Network configurations can be carried out in two ways. The first is via an Admin utility, and the second is using a static address. This is calculated using the MAC address of the device and a look-up table.
Our test camera was supplied with a USB stick which included the utility, but it wouldn’t run. Instead we used the static address, which worked well, but the camera wouldn’t load its viewing element. We tried all the typical cures, but with no joy.
A visit to the Riva website found another copy of the utility, which did run, but that didn’t include the driver. Next we downloaded and installed a new version of the firmware, and this solved the problem.
Making the VCA configurations is relatively intuitive, which is good because the electronic manual isn’t the best. It covers many functions which are not included as standard, and the screenshots are either from another product or outdated.
Metadata needs to enabled to ensure correct operation. Setting up zones, lines, and adjusting for mounting height and perspective are all simple tasks, and are aided by easy visualisation.
Setting zones and lines is relatively straightforward. The interface uses a basic contextual menu, and once the desired element is selected it can be drawn onto the screen visualisation. It can then be given a dedicated colour, and various parameters can be adjusted to allow performance tweaks to be applied. These include detect/non-detect, enter/exit, appear/disappear, loiter and dwell times, etc..
To ensure performance is stable, a calibration screen guides you through setting the correct parameters for any given camera. The VCA element uses ‘mimics’ – effectively adjustable generated figures – to simplify the process of setting the perspective correctly. Also, certain parameters can be ‘classified’ which allows relevant additional criteria to be specified.
The VCA functionality works well, and use of detection and non-detection zones allows the elimination of most types of typical nuisance activations.
The choice of event types meant that all of our test scenarios were achieved with a very good degree of reliability. Video quality was good, and whilst there are some obvious performance restrictions when using VCA, there were no real issues.
Samsung – SNB6004P
The SNB6004P is an HD1080p box-type camera which features the manufacturer’s WiseNet III processing engine. This incorporates the Smart Codec, which combines IVA, video motion detection, video-based tamper detection, face recognition and audio detection.
The camera streams at rates of up to 50ips, using H.264 and Motion-JPEG processing.
The intelligent video analytics functionality includes elements such as line crossing and object appear/disappear. When a face is detected, the smart codec uses variable processing to ensure that the quality is high enough to deliver a positive identification.
The camera uses a 1/2.8 inch CMOS sensor to deliver HD1080p, HD720p and 1.3MP streams, plus other standard definition resolutions. Frame rates of up to 50ips at all resolutions are achievable when using H.264, although these fall to 15ips for the higher resolutions if M-JPEG is utilised.
The unit has a claimed sensitivity of 0.1 lux for a 50IRE image. Other video-based features include noise reduction, image stabilisation, defogging, privacy masking, focus control, two-way audio, edge storage via SD cards and alarm I/Os. There are also all the usual suspects such as AGC, BLC, etc.. Cropped streaming is also supported, effectively delivering regions of interest. Power is PoE or 12V DC/24V AC.
Network configurations are carried out via an IP Installer utility, and this can be used to set the address details. Once this is done, the viewer element is loaded. The manual warns that if this process doesn’t complete, Smartscreen Filter – an Internet Explorer tool – needs to be disabled. With our unit the viewer loaded, but wasn’t recognised as having done so, so it kept on reinstalling itself. The camera needs to be added to the IE Compatibility View list to solve this; this point is not covered in the connection guide.
Making the IVA configurations is intuitive, and whilst at first glance the settings seem almost too basic, additional settings do pop up on screen as you draw lines or boxes.
The options are somewhat limited when compared to some of the other devices in the test, but with careful planning and judicious use of detection and non-detection areas, relatively stable scenarios can be created, which deliver consistent results.
Setting zones and lines is relatively straightforward. The stages involve enabling IVA and selecting associated alarm actions (email, FTP, record to edge media, alarm output). You then set the sensitivity, and define minimum and maximum sizes, before marking out detection and non-detection areas. Finally, you add the analytics elements and rules. These are either ‘Passing, Enter/Exit’ rules, which can either be for a zone or a line, or ‘Appear/Disappear’ rules, for either the entire detection area or for a defined part of it.
When a zone or a line is drawn, a pop-up box gives you a few additional criteria which can be set. These are very basic, so we found that we needed to tweak the settings for the zones and detection areas to achieve the results we required.
All of the cameras tested offered something of a balance between flexible IVA performance and ease of set-up. The Samsung camera is biased towards ease of set-up, and as a result you do lose some of the flexibility available from the other units. That said, all of our test scenarios were achieved, albeit with a bit of tweaking.
Whilst the facial detection feature was outside of our test remit, the functionality does work. However, camera placement to optimise it isn’t always conducive to achieving best performance with the other analytics features.
Video quality was good, and there was little to no degradation caused by the processing load of the IVA.
Verdict
What constitutes intelligent video (or VCA, IVA, smart video or any of the other widely used terms) doesn’t have a specific industry definition. This is both a positive and negative. The installer or integrator has to establish exactly what each different manufacturer means by these descriptions. However, it equally allows different devices to be focused on specific needs, which does deliver a level of flexibility and an element of choice.
The Q1614 from Axis Communications is a very good camera, and has VMD that is both usable and useful! You can filter out innocuous activations with scenarios, which does allow you to achieve a higher level of security. However, it can’t separate behavioural actions from general motion. If we were looking solely at VMD performance, it would have rated much higher, but in this test the more basic set-up means it just scrapes in as a Recommended option.
The Bosch NBN-733V-IP has a good level of flexibility, and allows behavioural traits to be detected. This gives a good degree of flexibility, and will make the camera attractive to those seeking advanced video-based detection. You might not get all the functionality that is available from a fully blown IVA package, but you certainly get all you need for a mainstream application!
The Panasonic WV-SFV631L delivered a good degree of intelligent functionality, and putting the issues with the software down as an anomaly which has been addressed, there were no problems with the unit.
The RC3502HD-5311 from Riva gives a good degree of flexibility, and whilst it is intuitive, we’d like to see a better quality manual included. Our initial set-up issues were resolved with a firmware upgrade.
The Samsung SNB-6004P does deliver a decent level of IVA functionality, but lacks some of the flexibility of the other units. That said, it achieved what was required of it.