fbpx
Home Technology CCTV Test: Object-based Video Analytics (Part 2)

CCTV Test: Object-based Video Analytics (Part 2)

by Benchmark

One of the first video ‘analytics rules’ to emerge in the security sector, way back in the analogue days before terms such as VCA or IVA were invented, was the detection of object removed violations. It seemed very advanced at a time when video motion detection was considered to be the pinnacle of security technology. Today, ‘object removed’ and ‘object left’ analytic rules are typically only mentioned in the lists of additional rules for devices or software packages. Benchmark took a look at some of the object-based options to find out if these basic IVA functions offer anything of use to installers and integrators. The concluding part of this test looks at options from Panasonic, Samsung, Sony and Riva.

READ PART 1 OF THIS TEST

untitled

Click for larger image

When it comes to video analytics, the headline features tend to be those that allow the creation of the more ‘exciting’ applications. Demonstrations and screen grabs used to promote the technology tend to involve intruders scaling fences at critical infrastructure sites, vehicles being pursued through high risk environments and individuals being tracked as they travel across campus-type applications. The general impression being generated is one of smart and innovative solutions tackling crime at its most insidious.

Whilst it is very true that video analytics does deliver in all of these scenarios, the technology can also offer simpler solutions to more basic problems. Talk to those tasked with delivering IVA, and they will tell you – if they’re being honest – that for every high-end deployment of the technology, there are dozens of others that carry out more menial tasks.

While the technology can still shine in such applications, they often don’t get highlighted because they lack the ‘sexy’ nature of the more challenging roles that IVA fills.

Of course, for most installers or integrators, it is this ‘bread-and-butter’ business that forms the backbone of customer demands, and as a result the core functionalities of any IVA set-up are likely to be more important than the headline-grabbing features.

When you look down the list of rules incorporated in many IVA packages, ‘object left’ and ‘object removed’ are often present. These are very common scenarios, and are important to a wide range of customers. The reason they sometimes don’t attract the headlines is because, on the face of things, they appear to be very simple.

If the implementation of these rules is done well, they are not only simple, but also offer a solution for a wide range of applications. In this test, Benchmark sets out to see if this functionality is reliable and robust in a very basic scenario.

One thing that can be confusing is what manufacturers mean by ‘object left’ and ‘object removed’. The term ‘object’ is used by some to include people, vehicles, targets, etc..

Some use the terminology to describe an alert caused by the arrival of an ‘object’; for example, if a car drives into a region of interest. However, with ‘object left’ analysis, expected objects behaving in ordinary ways should not trigger any type of alarm. Equally, with ‘object removed’, temporary obscurations or other environmental changes should not be confused with relevant events. Often a field trial is the only way to understand what type of object analysis is really being offered.

For the core test, Benchmark used an office which included a storage area. The goal was to detect left objects or removal of stored items. The object left and object removed tests were carried out separately; where simultaneous detection was possible, this is also reported.

Panasonic: WV-SPN631

The WV-SPN631 is an HD1080p camera with a frame rate of up to 50fps. As standard it supports a range of features including VMD, but with an optional software expansion it can also support intelligent functionality.
This includes object left or object removed, face detection, intruder detection, loitering, directional motion detection, line crossing and scene change. The object-based analytics can detect up to eight objects in the viewed scene. Detection requires a time window of 30 seconds.

The camera makes use of a 1/3 inch MOS image sensor and has a quoted sensitivity of 0.04 lux. Other features include Enhanced Super Dynamic, Adaptive Black Stretch, noise reduction, VIQS which allows variable quality based upon regions of interest, privacy masking, alarm handling, edge recording via SD cards and fog compensation. The camera supports PoE; 12V DC can be used as an alternative power source.

When it comes to installation the Panasonic camera should be supplied with an IP utility for address setting. Our unit didn’t have this included. That leaves installers and integrators with two choices: they can either find a download on the Panasonic website, or use a DHCP-enabled server and look up the allocated address at the switch. The utility only operates for the first twenty minutes that the camera is on-line.

Once the set-up is complete, the general menus are straightforward, and camera configuration gives access to a wide variety of options. Image control is very good, and a high quality stream can be obtained in most conditions with use of the various processing elements.

In the Alarms menu there is a choice between VMD and iVMD: this only appears if the extension software has been licensed. Selecting iVMD takes you into the menus for configuration of the analytics, of which there are two. The first sets ‘Detection Mode/Area’ which includes active detection zones and the types of rules deployed. The second is ‘Depth’ and applies additional discriminations to enhance the accuracy of the analytics.

The first time the iVMD window is launched you are prompted to load another viewer. This was straightforward, but afterwards attempts to view the configuration screen showed a warning that someone else was currently configuring the iVMD. As there were no other users this was obviously a bug. Logging out and reauthenticating as a user didn’t cure the problem, and eventually a camera reboot was the only resolution.

This small anomaly aside, the actual set-up is relatively straightforward. The first task is to create the detection area and name it. There are numerous zone creation tools allowing a rectangular zone or mask, a polygonal zone or mask and a crossing line. Up to eight zones or masks can be created, and each has a colour to simplify set-up. Once drawn, an iVMD rule is selected. These are Object, Intruder, Direction or Loitering.

The next stage of the configuration is to set the Depth. This consists of creating two boxes to allow the analytics engine to compensate for perspective. Care should be taken when setting this up. A bit of trial and error did achieve our goals.

There are some advanced settings which can be made, such as altering the alarm time. To access the menu a URL is required as there isn’t a direct link in the GUI. Advanced settings can be tweaked for the various rules, so sensitivity can be different for object recognition and line crossing, for example! The advanced settings are for sensitivity, alarm time and detection size.

Once configuration is complete, accuracy levels are consistent, and all attempts at removing or leaving objects in the protected area are detected. Effectively the analytics uses a single rule for both tasks, but we didn’t see any real difference in performance aside from object removed alerts taking a few seconds longer than object left alarms.

Significant scene changes such as switching of lighting did not generate false alarms. There is an option during alarm zone configuration to disable global scene change detection, and this enhances stability.

Riva: RC3502HD-5211

The RC3502HD-5211 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 available at an extra cost. These include intrusion detection, people counting and object detection packages.

The camera offers H.264 and Motion-JPEG compression, and features include dual streaming, edge recording, audio support and digital WDR. It 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 then the maximum frame rate drops to 15ips. With a HD720p resolution, real-time streaming can be preserved. Maximum bit-rate is 6Mbps.

The first thing you notice when you install the RC3502HD-5211 is that if you use the rear cable entry point there isn’t much room. A Cat 6 cable with standard RJ45 boot necessitates forcing the plug into the socket because it is located so close to the housing. While plugging the connection in is a pain, unplugging it is actually more difficult.

On initial power-up the camera’s fan was pretty noisy, constantly emitting an audible whirring sound. This gives the impression that it is working much harder than it maybe should be! Over the course of test the fan fell silent a few times but never for very long. More often than not the noise was present. If the camera is installed in an external location, ambient noise will drown this out, but in a quiet internal location it could become irritating for some.

The camera is provided with a simple IP utility and this works well. It finds the camera almost immediately and allows the network configurations to be made. With this completed, the viewing elements load and final configurations can be carried out. The menus are straightforward and intuitive.

The camera is supplied with a USB stick which includes the IP utility and two PDF manuals. One is for the camera and the other covers the video analytics.

Riva’s cameras offer a range of analytics rules, and all but the most basic require a license. When setting the camera up, the menu structure is easy to follow once you have a handle on how the analytics work.

There are a number of object-based rules, but it is important to consider the term ‘objects’ as covering people and vehicles. In object classifications, these two are supplemented by ‘clutter’, which is basically a catch-all term for busy scenes. There is an option to define ‘unclassified’ objects. This allows the setting of minimum and maximum sizes for custom classifications.

The set-up of the actual rules is straightforward, and double-clicking on the Zone listing reveals additional parameters. These include the exact type of analytic to be deployed. Multiple choices can be selected, and if you are seeking to deploy traditional object detection you do need to find ways to optimise the performance. This is because the available rules do not specifically include object appear and object disappear; well, they do, but not in the typical sense.

The appear and disappear rules are for objects that do not travel into the protected area. They look for objects appearing, such as coming out of a doorway or exit in the protected area. The entry and exit rules look for objects that travel into or out of the protected zone. This is optimised for vehicles driving into a protected area, for example.

As such, whilst the analytics are flexible and offer accurate detection, if you’re seeking to detect objects – such as assets or items – being left or removed, you might have to experiment with various options to find the solution that is right for any given application.

In many cases by linking either the appear/disappear rules or the entry/exit rules with object filtering and time discriminations, you could deliver a credible and accurate solution. However, in some applications it might be better to select software or a device with a dedicated option left/removed rule.

If you opt for the Riva camera, a site trial would be essential to ascertain whether the VCA approach to object detection would be suitable.

Samsung SNB-6004P

Samsung’s SNB-6004P is a networked-enabled day/night HD box camera. The unit utilises the WiseNet III processing engine which includes a number of video analytics rules as standard. These include object appear, object disappear, entry to a zone, exit from a zone and passing.

The camera uses a 1/2.8 inch sensor to deliver streams at HD1080p and HD720p resolutions, along with megapixel and standard definition resolutions. Multiple streaming is supported, and compression is via H.264 or M-JPEG.

Frame rate is 50fps at all resolutions when using H.264 streams; M-JPEG is 15fps at HD and megapixel resolutions, or 25fps for standard footage.

Other features include focus control, wide dynamic range, backlight compensation, contrast adjustment, noise reduction, image stabilisation, AGC, motion detection, privacy masking and de-fogging. Edge recording is supported via an SD card slot. Sensitivity is quoted as 0.1 lux for a 50 IRE image. Power is PoE or 12V DC/24V AC.

The camera is supplied with an IP utility which is used for initial network configurations. This works well. It found the camera within a few seconds and allowed the network settings to be initialised. Following connection, the viewing software loads and the set-up of the camera and its integral analytics can commence.

The WiseNet III platform has been around for some time now, and it does deliver a high degree of flexibility. Its support for open platform applications allows the use of third party apps, including those that deliver intelligent video analytics. However, the camera also boasts its own integral IVA functionality.

There are three types of analytics supported. These are passing (line-crossing), entry/exit of a detection zone and appear/disappear. The latter covers the object detection tasks as required for this test.

Video analytics and motion detection parameters are configured from the Events menu. This is well laid out and is intuitive, thus allowing a relatively simple set-up. With regard to configurable discriminations to filter out nuisance activations, the options are limited, but do allow a fairly high degree of accuracy to be configured.

Once the video analytics feature has been enabled (a simple but important button selection), the screen view is shown, along with an option to select motion detection, analytics or both functions.

With analytics selected you can then access four tabbed menus. These are Sensitivity, Size, Area and Analytics
Sensitivity settings are applied via a drop-down menu with five levels. The next task is to set minimum and maximum target sizes. This can be done by simply drawing boxes for each size. Alternatively, pixel-based sizes can be entered.

The Area menu allows the creation of detection zones. These can be specified as detection and non-detection (masked) regions. The shapes are node-based, thus allowing a good degree of flexibility.

The final menu allows the selection of the analytic rule to be deployed. The options are for Passing (line-crossing), Entry/Exit and Appear/Disappear. The latter was the rule tested. When this is selected an additional configuration option appears, allowing the rule to look for objects in the defined detection area or the entire screen image.

The detection and rule activity can be overlaid on the live image; two checkboxes allow this to be set up. Finally, the alarm action can be specified (FTP, Email, local recording, output triggering and ePTZ preset are the options) and the rule itself can be scheduled.

In general operation, the object appear/disappear works well, and once settled it detects genuine events quickly and consistently. Apart from the size discrimination and general sensitivity changes, you can’t really tweak the performance. This isn’t an issue in decent conditions, but as light levels fall noise and shadows do contribute to occasional nuisance activations. In truth any analytics rule aimed at applications with harsh lighting will need additional discriminations to eliminate false alerts, and care is required when using the integral analytics in such environments.

Where lighting is less challenging, the analytics function works well and will be suitable for basic detection.

Sony: SNC-VB635

The SNC-VB635 is a networked day/night HD1080p camera with a frame rate of up to 50fps. It includes a number of video analytics rules as standard. These include object left, object removed, intrusion and passing. The SNC-VB635 also includes face detection, tamper protection and audio detection.

The camera utilises H.264 and M-JPEG compression and supports multiple streams. It features a 1/2 inch Exmor sensor powered by the Ipela Engine EX to deliver improved low light performance; sensitivity is quoted as 0.04 lux (50 IRE). Along with the usual functions such as AGC and BLC the camera also includes exposure compensation and tone correction, along with Sony’s proprietary View DR wide dynamic range.

Other features include electronic image stabilisation, privacy masking, two-way audio, voice alerts and edge recording via SD cards (SD and SD-HC formats; SD-XC is not supported). The camera can be used in conjunction with smartphones or tablet devices. Power is PoE or 12V DC/24V AC.

The camera is shipped with a quick start guide; an included DVD contains a manual and an IP utility. This finds the camera quickly and effectively allows network settings to be finalised. One note is that the manual only covers the utility. There is no documentation for the camera. If you need it you’ll have to hunt it down on-line!

The menu structure is relatively straightforward, and most of the screens are intuitive. Certainly, with regards to image set-up there isn’t anything that will confuse.

Video analytics parameters are set via the Action Input menu. This screen has six tabs: Event Condition, Sensor Input, Camera Tamper, Motion Detection, Face Detection and Network Disconnection. We initially wondered whether the analytics needed to be licensed as there was no specific IVA menu, but Sony bills the object appear and disappear functions as a VMF function within VMD.

The first task is to set a motion detection area. Non-motion zones can also be established. Sensitivity and motion response time can be configured, as can elimination of alerts caused by shadows. The final task is to set minimum and maximum target sizes, achieved by either drawing boxes or entering numerical pixel counts.

Once done, VMF filters can be added. Up to three filters can be supported. Choices are object removed, object left, intrusion or passing. When selecting object left/removed, there is an option to set time until alarm.

Each filter needs an active zone to be created; the process is the same as setting a VMD region. The menu screen for each filter allows visual checking of correct operation. It pays to let the system stabilise before testing in order to see realistic results.

In operation the VMF filters are accurate in stable scenes with good lighting. High levels of background activity do result in missed triggers, and image noise also causes some degree of irregular activity. As such, care should be taken with its deployment in certain applications.

The simplicty of set-up highlights that the functionality isn’t intended to offer high level analytics-based detection.

Verdict

Object left and object removed analytic rules have a wide number of potential uses, and while they don’t attract the headlines like some other IVA rules, they still offer a degree of opportunity for many installers and integrators.

The WV-SPN631 from Panasonic is a good quality camera, and the delivery of iVMD options certainly enhances what it offers. It does require a license to free up this functionality, so that needs to be factored in when looking at its price. Performance is accurate and consistent, and object detection works well, even in relatively busy scenes.

The RC3502HD-5211 from Riva does deliver high quality IVA, but for some looking for specific object left and object removed detection, it’s approach might not be right. This doesn’t mean it’s not a credible IVA tool, because it is very good. However, its real strength lies in the detection and tracking of people and vehicles rather than static objects. In tests involving tracking, it would certainly rate much higher.

Samsung’s SNB-6004P delivers a good standard of object detection, and is also easy to set up. However, the ease with which it is configured means that it lacks the discriminations to filter out nuisance alerts in harsh lighting and hostile environments. In fairly stable applications, it does offer a simple and effective choice.

Sony’s SNC-VB635 adds object recognition to its motion detection function, and that indicates that it doesn’t enjoy the enhanced functionality inherent in high-level analytics. It works in stable environments, but can’t compete with other devices on the test.

READ PART 1 OF THIS TEST

Related Articles

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More

Privacy & Cookies Policy