Operates locally — protects data and prevents leaks. AI analyzes events in real time, reducing staff workload: monitoring is possible even with a single operator
Every recorded event is accompanied by an automatic incident clip, and notifications are instantly delivered to the app
Optimized neural networks running on CPU processors deliver high performance without the need for expensive GPU servers
We deploy the system on your devices (Linux and iOS) — fast, secure, and private
We deploy the system in the cloud — convenient and scalable for your needs
Training the system for one custom event included as a gift (with annual payment)
We adapt the system's neural network to recognize unique events, objects, and actions specific to your environment — whether it's an industrial site, retail space, logistics center, or another scenario. You get a system tailored to your individual use cases and needs
We adapt the system's neural network to recognize unique events, objects, and actions specific to your environment — whether it's an industrial site, retail space, logistics center, or another scenario. You get a system tailored to your individual use cases and needs
The system detects an event within no more than 3 seconds from the moment it occurs and sends a notification to the application.
Video stream analysis is performed at a configurable frequency — from 5 to 20 seconds, depending on site conditions and the expected frequency of events.
This makes it possible to balance response speed, computational load, and recognition accuracy.
The system supports a set of pre-trained scenarios that can be launched at the site immediately after installation.
If recognition of unique or highly specific events is required, additional neural network training is carried out for the customer’s specific conditions and objectives.
With local deployment, the system is fully autonomous and can operate without an internet connection, using only the site’s local network.
If a cloud-based deployment is selected, a stable internet connection is required for system operation and video stream transmission.
The system supports a flexible deployment architecture:
The choice depends on security requirements, scalability, and available resources.
In adverse weather conditions (rain, snow, fog), recognition accuracy may decrease; however, the system continues to operate and record events.
For night-time operation:
The system is compatible with all types of cameras that support a minimum resolution of 480 pixels, including IR cameras.
The system can operate on both CPUs and GPU accelerators.
In practice, it has demonstrated high performance:
The choice of configuration depends on the number of cameras, scenario complexity, and processing speed requirements.
Video recordings are stored on the deployment site — on local servers or the customer’s infrastructure.
In the case of cloud-based processing, only the video stream required for analysis is transmitted to the cloud, without long-term archive storage, which increases data security and control.
All incident notifications are delivered to the notification center of the AVA desktop application, available for:
Operators receive real-time alerts with the ability to quickly jump to the relevant video fragment.
After launch, the system requires several days to adapt to real site conditions (lighting, camera angles, behavioral patterns).
When deployed at a new site, AVA specialists perform:
Training the system to recognize new, previously unsupported events takes from 1 to 2 months.
The timeframe depends on:
Yes, the system can be extended with modules for facial recognition, license plate recognition, and other analytical technologies.
Such features are enabled upon request and in compliance with applicable personal data protection laws.
