Perne
Computer vision: visitor counting, queues, product recognition: illustration 1
Data and AI analytics

Computer vision: visitor counting, queues, product recognition

We work as a team and take a recording or a live stream from your camera. First we clarify what exactly we count or recognise and where the camera sits, then we pick a ready detection model, for example based on YOLO, and fine-tune it on your frames. For counting we use OpenCV and tracking of objects between frames, and we output results to a table or a simple dashboard. We do not recognise people by their faces, do not determine biometrics and do not store images of people: we count anonymously only. Accuracy depends a lot on lighting, camera angle and image quality, so we will show the result on a test recording before handover.

from 200,000 ₸ (≈ $400) one-off

timeline: 14–30 working days

  • Analysis of the task and choosing camera positions from your recording
  • Choosing and fine-tuning a detection model on your frames
  • Counting visitors or people in a queue, taking entries and exits into account

Order

Message on WhatsApp
TelegramCall

The service name is added to the message for you.

This is a starting price; we give the final amount before work starts. USD is approximate.

What is included

  • Analysis of the task and choosing camera positions from your recording
  • Choosing and fine-tuning a detection model on your frames
  • Counting visitors or people in a queue, taking entries and exits into account
  • Recognition of a preset set of products on a shelf, up to 20 items
  • Export of results to a table or a simple dashboard with hourly charts
  • A test run on your recording with an accuracy report and setup instructions

Who it is for

For shops, food outlets, banks, clinics and car parks that need to understand footfall, queue length or whether a product is on the shelf. It is not a fit if you need to identify specific people, search by face, or recognise staff.

Not included

To avoid surprises, here is what the base price does not cover and what is discussed separately:

  • Face recognition, identification of people and biometrics: we do not do this
  • Installing and configuring the cameras and network equipment: a separate job
  • A powerful server or GPU rental for real-time processing: paid separately

FAQ

After payment the code, the settings and the fine-tuned model are handed over to you, and you use them in your work. The base model is distributed under its own open licence, and we follow it. Recordings from your camera remain yours, and we do not use them for other projects without your consent.

The price includes two rounds of revisions after the test run: changing counting zones, the trigger threshold, or the product set within 20 items. Adding new cameras or a substantially different task is paid separately, and we agree on it in advance.

Usually 14 to 30 days from the moment we receive the recordings and agree on the task. A large part of the time goes into labelling your frames and tuning the model, so the timing depends on the number of cameras and the quality of the footage. We will name an exact date after we have looked at your video.

Recordings from the camera covering several hours at different times of day, a description of what we count or recognise, and the place where the camera is mounted. It helps to say which lighting conditions apply and whether you need live mode. Access to the camera stream is needed only for real-time operation, and we will set it up together with your technical specialist.

Yes, as the owner of the premises this is usually your responsibility: post information about video monitoring and that counting is done without identifying people. We can help draft the sign on request. Personal data laws vary by task, so if in doubt, check with a lawyer.

Accuracy depends on lighting, the height and angle of the camera, how crowded the space is and the picture quality. In the test run we show figures on your own recordings, so you see the real result. We cannot guarantee accuracy in every condition, but we can honestly say what would improve it, such as changing the camera angle or lighting.

Standard IP cameras that stream over RTSP, and recordings in MP4 or AVI files, will work. Cameras with very low resolution or heavy compression give worse results. If you are unsure about your camera, send us its photo and model name, and we will tell you whether it is suitable.

You may also need

CCTV installation on site

CCTV installation on site

On-site visit: Almaty

We mount the cameras, run the cables, set up recording and check the picture on site. On-site work is currently available in Almaty only.

from 29,000 ₸ (≈ $60) one-off

Timeline: timeline: 1–3 working days

A custom ML model for your task

A custom ML model for your task

We train a machine learning model on your data so it solves a specific task: a forecast, a classification or anomaly detection.

from 150,000 ₸ (≈ $300) one-off

Timeline: timeline: 14–30 working days

Shall we discuss your project?

The product name and a link to this page are filled into the WhatsApp message for you.

Message on WhatsApp
TelegramCall

Or send a request

We will reply in the channel you choose: WhatsApp, Telegram, phone or email. Your contact is used only to reply.

Service
How to reply

from 200,000 ₸ (≈ $400) one-off

Message on WhatsAppTelegramCall