Skip to content
UA Discuss your project
  1. Home
  2. Resources
  3. How to choose a video analytics system for a business in Ukraine in 2026: 10 questions for the vendor

How to choose a video analytics system for a business in Ukraine in 2026: 10 questions for the vendor

9 min read · Updated 2026-11-12

Author: DuoVision team

For: owners, plant, warehouse and security managers, IT directors and buyers comparing video analytics in Ukraine

Summary

  • To choose video analytics, define the task in one sentence first, for example "boxes on the floor in the workshop" or "counting visitors at the entrance". The task decides the type of solution: a module in a video management system, a cloud service, a narrow product or an integrator working to your task. Then check the vendor with the ten questions below.
  • Ask a vendor about ten things: cameras, where frames are processed, who reviews false alerts, shadow mode, accuracy with sample size, personal data, model licences, integrations, support, and the pilot format.
  • Model licences are the item most often missed in contracts. YOLO-family models from Ultralytics are distributed under AGPL-3.0, and a commercial product needs either published code or a separate commercial licence.
  • The most reliable way to compare vendors is a pilot on your own cameras, with success criteria written down before the start and the result judged by your own person.

For: owners, plant, warehouse and security managers, IT directors and buyers comparing video analytics in Ukraine

Where to start: the task and the type of solution

Video analytics is software that analyzes camera frames and points out where an event happened: a person without a hair cap, a truck at the ramp, a queue at the checkout, a licence plate at the gate. Buyers purchase an answer to a specific task, so first write it in one sentence and add where, on which cameras, and who will look at the result. If the terms are new, see what video analytics is.

The Ukrainian market roughly splits into four types of offer. The table helps decide which one to start the conversation with.

Solution typeWhat it suitsWhat to check
Analytics modules in a video management system (VMS)Standard tasks: line, zone, counting, plates. Cameras and software already exist and a function must be added. Example: TRASSIR, whose site lists video management, access control and an analytics sectionWhich modules actually exist and on which cameras they were tested. Whether the software is on the prohibited software list: Macroscop and Axxon were added on 28.01.2026; there is no direct ban for private business, but questions about licences and support appeared
Cloud servicesSmall sites, quick start, little own IT infrastructureWhat stream leaves the site, where the servers are, what happens when the internet drops, who owns the frames
Narrow products (licence plates, visitor counting)One known task that a ready product already solvesWhether it integrates with your accounting system and whether a second task can be added without a second vendor
Integrator working to your task on existing camerasNon-standard rules, industry specifics, zones and thresholds built for your siteWho supports the system after launch and whether there is a shadow mode before going live

We collected an overview of Ukrainian companies with links to their sites separately: video analytics companies in Ukraine. DuoVision is one of the companies there, not the only one. The questions below work for any type.

Questions 1-3: cameras, where processing happens, false alerts

1. Does the system work on your existing cameras, and which ones exactly?

A good answer contains requirements: protocol (RTSP or ONVIF), resolution, field of view, lighting, mounting height for your task. The vendor should look at frames from your cameras before making an offer and check a few of them before the contract. Be wary if it works with only one make of equipment. Some tasks do not suit existing cameras; licence plates at speed need a separate camera at the right angle. You will be told honestly if the vendor looks at a frame. More on choosing cameras in cameras for AI video surveillance.

2. Where are frames processed: on site or in the cloud?

Processing on a server at your site means the video stream does not leave the enterprise network, and only event cards may go out. Cloud processing is easier to launch, but video or frames go to the vendor's servers, so you need to know where they are, how long they are kept and who can access them. Add the question of what happens to the system if the internet goes down.

3. Who reviews false alerts, and how?

Every system has false alerts. What matters is what happens to them next. It is good when every alert is a card with a frame that a person reviews, and the verdict ("violation" or "false") is stored. Then you can see how many cards were false and improve the rule by changing zones and thresholds. Ask how many cards per shift to expect and who will review them: the time spent on cards is part of your cost. It is bad when the system fines people, blocks passage or sends accusations by itself without a human check, or when the vendor cannot estimate the number of cards. A card should hold a frame, the rule, the camera and the time, and false alerts come, for example, from a shadow, a similar object or a person hidden by another person. In our experience, zones drawn by the food-safety manager on the frame removed more false cards than changing model thresholds did. Details in the article about launching in a meat workshop.

Questions 4-6: shadow mode, accuracy, personal data

4. Is there a shadow mode before launch?

Shadow mode means the rule works and collects cards, but fines nobody and sends no alarms. After some time you see how many alerts were real, and only then is the rule moved to live mode. Without this step, the first weeks turn into alarms that staff quickly learn to ignore. Ask to see what shadow mode looks like in the system and who decides on the move to live.

5. How is accuracy counted, and on what sample?

An accuracy figure without sample size, period, cameras and the person who judged means nothing. An example from our practice, the glove rule: on 45 frame fragments the share of false verdicts was 47%, while on all 219 fragments (131 events) reviewed blind it was 23%. The conclusion is that a small sample cannot be carried over to the whole system. Ask the vendor:

  • how many events and how many cameras were in the sample;
  • whether it included empty frames and hard cases from other days;
  • whether neighbouring frames of one event were counted as one event;
  • who made the verdict and whether they knew the model's answer.

Best of all is when the vendor runs such a measurement on your cameras during the pilot.

6. What about personal data and retention periods?

Recording people in a workshop or warehouse is personal data processing. The Law of Ukraine "On Personal Data Protection" (No. 2297-VI) defines personal data as information about a person who can be identified (Article 2), requires a purpose set in advance, no excessive data and a retention period no longer than needed (Article 6). Processing needs a legal ground (Article 11), and the controller and processor must protect data from loss and unlawful access (Article 24). You remain the data controller, so people must be notified about the recording (Article 12), and if frames are sent abroad, Article 29 applies. In practice ask the vendor: how long frames and cards are kept, who has access, whether data can be deleted on your request, and what the processor agreement says. A good vendor does not switch on face recognition without need and the people's written consent. Face recognition falls into a separate category of biometric data and is covered in video analytics and personal data.

Questions 7-10: licences, integrations, support, pilot

7. What licences do the models, code and training data have?

Few buyers ask this, and the consequences for a business are real. Some open models and libraries are distributed under AGPL-3.0. Section 13 of the licence requires that users interacting with the program over a network be offered the source code of the modified version. The licensing page of Ultralytics, the author of the YOLO model family, states directly that AGPL-3.0 is free but requires open-sourcing the whole project, while a commercial product, a SaaS, embedded camera solutions and fine-tuned models in commercial use need a separate paid licence. There are also models and datasets limited to "non-commercial use only".

For you this is a risk if the vendor built the system on such a model without the licence it needs: the vendor would have to change the model or the terms, and that would affect your project. Ask in writing which models and libraries sit underneath, what licence each one has, whether there is a commercial licence where one is needed, and where the training data came from. We check the licences of code, weights and dataset before using them.

Ask separately about your frames and annotations. Objects, containers or tools specific to your production usually need training on your frames. Find out who annotates the frames, where the annotations are stored, whether the vendor uses your frames for other clients, and whether this can be prohibited in the contract.

8. What does the system integrate with: 1C or BAS, access control, scales?

A card on a screen is useful, but the most value comes where the event enters your processes: a licence plate is matched against a booking in the accounting program, a gate event goes to access control. That depends both on the analytics vendor and on whether your program provides bookings (as a file, a shared database or an API). Find out before purchase which integrations already work for the vendor at real sites and which are only "possible", and who is responsible if your accounting system is updated and the exchange breaks.

9. What support is included, what does it cost, and how do you leave the contract?

A camera may be moved, the lighting may change, or the workshop layout may differ, and a rule stops working well. Ask who reacts to such changes, how fast, and whether notices about a camera that went dark or shifted, retraining and zone tuning are part of support.

On price, ask what it consists of: survey, equipment, setup of each rule, training for your task, integrations, support. Find out what happens if training does not give the needed result. Discuss the exit in advance: in what format you will receive the event log, the confirmed cards and the evidence frames, what happens to annotations made on your frames, and when the vendor will delete your data from its systems. These terms must be in the contract before work starts.

10. What does a pilot look like, and who judges the result?

A pilot beats a presentation because it shows the system under your lighting, with your people and your cameras. Agree in advance on:

  • one or two specific tasks and the list of cameras;
  • duration and mode (shadow first);
  • success criteria written before the start: what share of cards must be real, how many missed events are acceptable;
  • who judges: your specialist who sees the cards, not only the vendor's report;
  • what happens after the pilot and who owns the collected cards and labels.

If a vendor is not ready for a pilot on your cameras, that already answers the question about product maturity.

What should raise concern: a table for the meeting

QuestionWhat should raise concern
1. CamerasA demand to replace all cameras without a survey of the site
2. Where processing happensUnknown where the frames are stored, or "works everywhere" with no requirements for the link
3. False alertsAutomatic fines or blocking without a person checking, no estimate of the number of cards
4. Shadow modeA live launch at once, with no check on your events
5. AccuracyA percentage without sample size, period and who judged; a demonstration only on the vendor's own clips; a promise of "almost no errors"
6. Personal data"This is not personal data, because we only analyse"; an offer to recognise employees' faces "for convenience"
7. Licences and your dataUnknown model licences; rights to your frames and annotations not written down, or the vendor reserves the right to use them without limits
8. Integrations"With anything" without questions about what data your system provides
9. Support, cost, exit"We set it up once and it works for ever"; one sum without a breakdown; support with no description of what it includes; a log available only in the vendor's interface with no export
10. PilotNo pilot on your cameras; payment for training in advance without checking the result

Before and after the meeting:

  1. Write down two or three tasks you want to cover with cameras, and the cameras that already see these zones.
  2. Write each vendor's answers into the same rows of the table so that you can compare them.
  3. Ask to see the shadow mode result on your events, with the number of events and the period.
  4. Work out how many cards per shift to expect and which of your people will review them.
  5. Before signing, check the contract for rights to frames and annotations, the format of the log export and the data deletion period.

Where DuoVision fits

DuoVision is us, and we are only one of the options. We work to your task on existing cameras. Our deepest experience is in food production and HACCP: at a meat processing plant we checked the hair cap or net, mask, beard cover, gloves and a box on the floor, and riding a pallet truck we checked on recordings. A person's decisions went into a log for the audit. For ramps, warehouses and counting we tune rules to your site, in shadow mode first. We recognize licence plates.

We start with a demonstration on your cameras, and only then talk about launch. If your task is standard and is covered by a ready product or a module in your video management software, that may well be the better choice. Use the ten questions above with every vendor, us included. Tasks by industry are described on the industry pages of the site, for example production processes and warehouses and logistics. What goes into the price is covered in how much video analytics costs.

Frequently asked questions

Do I need to replace cameras to connect video analytics?

Most often no. Cameras that deliver video over RTSP or ONVIF connect to analytics. New cameras are needed where existing ones do not see the required zone or do not give sufficient quality, for example for licence plates.

Which is better: cloud video analytics or on-site?

It depends on the task. On-site processing keeps video inside the enterprise network and works without the internet. The cloud launches faster and needs less own equipment, but frames go to the vendor's servers. The key question is where the frames are stored and who can access them.

How long does a pilot take?

The duration depends on the task and the number of events on the camera: you need enough real cases to judge accuracy. The exact period and mode (shadow first) are agreed before the start together with the success criteria.

Can a free open model be used in a commercial product?

It depends on the licence. For models under AGPL-3.0, commercial use without publishing the code requires a separate commercial licence from the rights holder. Check the licence of every model and dataset before use.

Who provides video analytics for enterprises in Ukraine?

There are developers of video management software with analytics modules, narrow products for plates and counting, cloud services, and companies that tune a solution to a task. We collected an overview with links to their sites in video analytics companies in Ukraine. DuoVision among them specializes in food production and HACCP, and also in ramps and warehouses.

How do I compare vendors who quote different accuracy?

Ask for the sample size, the period, the cameras and who judged the result. A figure without them means nothing. A good sample includes empty frames and hard cases from other days and counts neighbouring frames of one event as one event. Question 5 above has the full list.

Is it legal to record employees for video analytics?

Recording people in a workshop or warehouse is personal data processing under the Law "On Personal Data Protection" (No. 2297-VI). It needs a purpose set in advance, no excessive data, a retention period no longer than needed and a legal ground (Articles 6 and 11). Details are in video analytics and personal data.

Where to apply this

Sources

  1. Law of Ukraine "On Personal Data Protection" of 01.06.2010 No. 2297-VI, Articles 2, 6, 11, 12, 24, 29
  2. GNU Affero General Public License, version 3, section 13 "Remote Network Interaction"
  3. Ultralytics, licensing: AGPL-3.0 and Enterprise commercial licence
  4. TRASSIR, official site (products: VMS, access control, analytics)
  5. Law of Ukraine "On the Basic Principles of Ensuring Cyber Security of Ukraine", Art. 4 pt 4 (prohibited software list)
  6. dev.ua. Another 13 programs added to the prohibited software list (Macroscop, Axxon), 09.02.2026

Let’s discuss your site

Request a consultation or a project estimate. We will tell you what your cameras can monitor, where it is best to process the video and how much it will cost.