Description

Specification

 Autonomous device of complex security

 videoblazer version “Forzet” series 3. 

Developed and manufactured at SpecLab 

APPLICATION 

The multifunctional climate device “videoblazer” with neural network object recognition can be used, depending on the program, as a video surveillance complex for any type of object, outdoor alarm with a high degree of protection against natural interference, a workstation for video surveillance operators, a tool for parking automation, an anti-terrorism security system with weapon detection, adaptive “smart traffic lights,” a component of the “Safe City” video surveillance point, a database for automatic collection of biometric facial data, a database for automatic collection of license plates on roadways, a situational center structure for displaying video streams on video walls, and a system for controlling technological processes and other purposes where there are video cameras, sensors, control devices, as well as communication and notification means.

Software developers can use the integration API to write their own software for their tasks.

TACTICAL AND TECHNICAL CCHARACTERISTICS

Streaming video:

Each IP camera only uses one stream (the best one) for viewing and recording. This has no impact on the device’s performance, does not overload its processors— they are powerful enough to handle a large number of HD channels. It also reduces network congestion (no need to stream a second channel) and lightens the camera load—its second stream can be utilized by other devices.

Displaying real-time video and events:

1.  The monitor is connected to the device. The monitor can be directly connected to the device with a resolution of up to 8K via an HDMI port. It can display both multi-channel video alongside event panels and each IP camera channel individually—enlarged to full-screen view. 

As part of the event presentation, it’s possible to display not the entire frame, but an enlarged area within it where the detection of the specified object occurred. For example, displaying a close-up of a face instead of the entire person with the full background. 

You can also view the Event log with cropped sections from frames where the actions of recognized targets occur. This reduces the amount of information that the user needs to cover.

2.  Viewing video over the network. Similarly, streams are displayed over network in any browser any device. The Web interface receives video streams with overlay of neural network data. Attached API allows the of metadata in other and applications.

Live video in both cases is streamed from the camera devices to the WEB in full resolution! At all times, whether one camera is being viewed on full screen or all cameras are in a reduced view. This is necessary for processing these streams by neural networks in good quality, as well as for other tasks and client connections. The video browser has sufficient capabilities for a large number of high-resolution channels (the most common question – no need to worry!), it has hardware co-processors, so the main processor is typically loaded only at 30-40%. All additional videoblazer functions related to scenario logic are not affected and operate independently.

Performance against multitasking: 

Personal computers on Windows or video recorders on Linux are fundamentally structured in such a way that it is impossible to control the load of processors or memory. Large operating systems themselves can unexpectedly start swapping memory, indexing files, or performing a whole range of their service tasks. You have probably seen how the processor load of your computer suddenly soared to 100%. Video streams can freeze for seconds to minutes, skip data, or deny access to clients. Multitasking is a problem when it comes to high-load video surveillance. The video browser is designed in such a way that a corresponding hardware part is allocated to each task. Compression - decompression of codecs is handled by a separate chip, neural networks by another, and so on. And the main processor is almost completely unloaded for general tasks. In its standard mode, it only utilizes about 30-40% of its capacity (this information can be verified from the operating system). Therefore, in situations where complex logic is involved (evaluating a large number of factors, sending alerts, distributing alarm data, monitoring its own parameters, etc.), the videoblazer has significant reserves to perform all operations without delays.

The frame rate for viewing in the browser does depend on the power of the computer or device on which the browser is running - unfortunately, the H.264/265 codec requires resources for decompression. If there are special chips for video unpacking in the video browser, then the Central Processor of the PC will be loaded.

Deinterlacing scaling.

To display images on a monitor directly connected to the videoblazer via HDMI, 3D image processing filters are used to eliminate negative effects.The hardware resources of the video browser enable achieving the highest picture quality, and this high-load processing does not affect the main processor in any way. When using the browser, the device’s graphics card is utilized. Without proper hardware support, the image may exhibit artifacts, poor interpolation, and other negative effects primarily caused by scaling. 

A great option for video walls. 

Videoblazers can serve as receiving equipment for situation centers, displaying video streams on large screens with resolutions up to 8K. Instead of noisy, cooling-demanding, expensive maintenance, and space-consuming PCs with large video cards, videoblazers address all these issues by delivering compressed video to monitors for viewing.

Video client.

Any device, even a weak computer, for example, a cheap laptop, smartphone, or smart TV, can act as a video client. Powerful video cards for neural networks are not required, as the client receives pre-processed meta-information from neural networks in the form of lightweight textual data.

The Videoblazer delivers live IP camera streams as if they were being sent directly by the cameras, without compression. A JSON protocol (with various options available) is added on top, and visual information about detected objects is overlaid on the video in the form of geometric shapes with labels on them.

The video server can be connected to any browser or more advanced software - the client part of the Windows software - GOALcity (provided free of charge for up to 4 cameras). The "Alert!" analytical panel can receive event video clips.

Separate partially free software is provided for tablet intercoms when setting up event video surveillance in cottages. This software is available for both Windows and Android tablets.

The open API protocol allows developers to connect their devices and programs to any operating system.

Metadata. The metadata about all detected objects with their coordinates is received along with the streaming video. Live channels are not needed for building situational logic, as analyzing the metadata is sufficient. It is hundreds of times smaller in volume compared to live video. And fully conveys the formalized content of what is happening - receiving programs do not need the video itself to conduct analysis.

Video analytics 

Event-based video analytics is represented by two main modes:

1.   Everything is recorded if useful objects specified by the user are detected, with segmentation into N-second clips (for convenient distribution). Full recording of the neural network archive with cyclic self-cleaning. People and other moving targets are always recorded, while vehicles are recorded only when in motion. Repeated frames (empty background) and noise are not recorded.

2.   Initial moments of events identified by video semantics. N-seconds at each scene change caused by new objects or new behavior of existing ones. (A person appeared, another person or several appeared, a car started moving, a weapon was detected, more items resembling weapons were detected, a vehicle plate was detected, another plate or several were detected, a QR code was scanned, etc.) There are no triggers for already detected targets unless they change their properties or movement characteristics. For example, the plate of the same vehicle is not identified twice unless it goes out of the visibility zone for a specified time – it will be identified again after reappearing.

The use of event logic. Each of the video events (appearance of a video clip) in both modes can serve as a trigger for activating the Logic algorithm programmed by the user. It can also be a composite condition of this logic.

This is a rigid, non-adjustable mode. In specialized versions (for specific orders), a custom reaction and recording mode may be set. It simplifies the setup of the logic but does not allow deviation from this logic. For example, instant closing of doors at the first frame with a weapon. 

In the basic version, it may take several seconds, or even tens of seconds (depending on the settings), before a reaction occurs. This needs to be taken into account.

Immutable logic is also used in specialized products, for example, automatic parking systems.

The user is provided with only the minimum number of settings for connecting devices, as well as tools for maintaining a database of license plates.

The interface of traffic light control systems or tools for recording traffic violations looks like a separate product.

In any case, the device's purpose can be changed by requesting a different firmware. For example, a home security system can be transformed into an automatic parking system and vice versa.

Partial visibility: Neural network objects can overlap with a high degree of overlap (in some cases up to 80%), while maintaining high recognition accuracy even in perspective removal.

A separate folder is allocated for storing FACES for counting, long-term storage, simultaneous display, and searching.

Neural network size: The basic configuration uses a neural network size category "S", which is sufficient in most cases. For complex street objects with a particularly high level of interference, high-resolution neural networks of the "extrasize" category are supplied. For example, for the protection of perimeters in a forest with heavy rainfall, birds, insects crawling on the cameras, snow whirls, vegetation, wind loads, etc. and all of this over long distances!

Such heightened sensitivity is often used for guarding state facilities, such as bridges, to fully engage neural networks for drone recognition to detect distant targets.

Setting up neural networks. Specialab neural networks are developed based on artificial intelligence, therefore they do not require special settings. However, out of habit, users are provided with an interface to adjust recognition accuracy, object sizes, stable detection time, and masking of specific areas.

The most in-demand feature is masking because interior statuettes may enter the field of view.

However, stationary objects may not be taken into account by the Videoblazer algorithms even without masking.

But it becomes more challenging with dynamic images from the television.

It is better to completely cover its screen with a mask.

Video recordings can go in the following directions:

• Messenger Telegram,

• In any regular BROWSER

•  "Attention!" panel of the GOALcity program

• In the cloud portal Farwit (Distant Witness)

• To the owner's smartphone (currently through the cloud or "Attention!" panel),

• Long-term event archive of the ACS

• In the client's personal or corporate space via an integrated protocol. 

Comprehensive Logic

The videoblaser can analyze both meta-data from neural networks from video cameras and data coming from various types of sensors – in a general set of conditions for decision-making.

The user simply selects the possible options for each element, described in simple human language.

In the basic version, 4 sensors are available in a "dry contact" connection with a closed-open line rating. Specialized versions integrate protocols of third-party devices connected via IP or RS-485 with assessment of states of any level and type. In this way, you can connect any system from turnstiles with intercoms to traffic lights with anti-drones.

Devices are managed in the same way. In the basic version, 4 sufficiently powerful 220 V relays are installed on board the video player. Third-party devices are integrated over IP or RS-485.

Logic programming is done in human spoken language.

So, if a person told you what he wants to do. 

Safety. 

In addition to standard security protocols, the videoblazer uses floating IP change logic to protect against physical connection to the network wire of the data transmission network. Even if an attacker has penetrated the object itself into the internal network, it is not enough to try to select a username and password on the device, you also need a username and password from the router, because the binding to the device goes only through the MAC. And for this, aerobatics is already needed, if the admin is not asleep. You can access the videoblazer only through the router. Even if a dynamically changing IP address is somehow caught, after some unknown time, it will change. Thus, the internal enemy will not be able to install a permanent device or program that tracks the videoblazer.