Внедрение Computer Vision в ресторане: раннее обнаружение гостей у входа + конверсия вход-посадка

Connection to Cameras
The system connects to already installed IP cameras via video stream (RTSP). The cameras are not reconfigured or replaced. Video is analyzed in real-time, and recording and storage of video are not required. Operates 24/7 with automatic reconnection in case of failures.

Guest Detection at Entrance
A virtual line or zone is set up at the entrance. When a person crosses this zone, the system records the event "guest entered." Immediately after this, a notification is sent to the manager in Telegram or in the work chat.

Attention Wait Control
After entering, the system tracks how long the guest remains in the entrance zone. If the guest waits for attention longer than the specified time (for example, 10-15 seconds), a repeated or enhanced notification is sent. This allows staff to respond promptly and not miss guests.

… Seating Determination
In the hall, a seating area (common table area) is set up. If a guest enters the hall and stays in this area longer than the specified time, the event "guest seated" is recorded. This is a simplified but effective logic for counting metrics.

Left Without Seating
If a person entered, was not seated, and exited back within the specified time, the event "left without seating" is recorded.

Metrics Counting
The system automatically counts the number of guests who entered, the number of guests seated, the number of guests who left without seating, and the conversion rate (seated / entered) by days and shifts. All events are saved with timestamps and camera information: entry, waiting, seating, leaving.

Reporting and Data Access
Data is available in the form of tables (CSV or Google Sheets) or through a simple web page with metrics and graphs. It is possible to analyze dynamics by days, shifts, and hours, identifying problematic periods.

Notifications
All notifications are sent via Telegram. A notification is sent when a new guest appears and when the allowable waiting time is exceeded. The delay is a few seconds, almost in real-time.

Additionally
The system does not recognize faces and does not collect personal data. The solution is scalable and allows for the connection of additional cameras. This is an MVP solution focused on quick launch and practical benefits.

Show original

Translate

Good day! Interesting project. I have good experience working with CV and AI. Write to me in private.

Show original

Translate

Hello!

This is a very clear and practical task, and I have worked with real-time computer vision systems where the goal was not just detection, but timely actions from the staff. I understand how important the first seconds after a guest enters are, and the solution needs to be quiet, reliable, and fast enough to truly change the behavior of the staff in the hall.

I would approach this as a lightweight, always-on service that connects to existing RTSP streams and monitors only specific entry and seating areas to keep performance stable around the clock. Entry is triggered by crossing the zone, and alerts are sent to Telegram within a few seconds with simple context, such as time and camera. One idea that works very well in restaurants is a short timer that adapts to the time of day, so during peak hours the escalation threshold is narrower, while outside of peak hours it remains calm and avoids overwhelming alerts.

Detection of locations should remain deliberately simple and reliable, using the duration of presence within the hall zone rather than trying to recognize tables or poses. This makes the MVP reliable and prevents false positives. All events are logged and linked together, so the conversion from entry to seating can be easily measured over a shift or day, and exporting to CSV or Google Sheets remains simple for managers.

I can provide a working MVP with real-time alerts, understandable metrics, and a small web interface or export capability for reports. The system will be designed in such a way that it can later be expanded for analyzing staff performance or comparing camera metrics without further development.
…
Thank you!

Show original

Translate

I am among the top 5 developers in the category of "Artificial Intelligence and Machine Learning" among ~2100 specialists on the platform. I guarantee:
- Fast and high-quality task execution
- Strict adherence to deadlines
- Regular communication throughout the entire process
I would be happy to discuss the details of your project in private messages.

Show original

Translate

Hello!
I am a Python developer with experience in Computer Vision. I am ready to implement a video analytics system for your establishment based on the combination of YOLO (detection) and ByteTrack (tracking). This solution will allow the system not just to "see" people, but to keep a continuous record of each guest from the door to the table.

How the technical solution will work:

- Thanks to ByteTrack, each guest who enters is assigned a unique identifier. This eliminates duplicate notifications if a person is just standing at the door and allows for a precise connection between the event "Entered" and the event "Sat at the table."

- Two virtual zones are marked in the frame:
1. Entrance Zone: fixation of entry and starting the waiting timer.
… 2. Seating Zone: the area of tables. If the guest's ID moves here and lingers — we record the seating.

- Notification logic (Telegram):
0 seconds: Guest crossed the line — the manager receives a notification "New guest."
15 seconds: If the guest's status has not changed to "Served/Entered the hall" — escalation in the chat (Warning).

- The system automatically matches the number of unique IDs at the entrance and in the seating area for the day/shift.

Technology stack:
Detector: YOLO (optimized for real-time).
Tracker: ByteTrack (high accuracy in preserving ID during partial object overlap).
Programming Language: Python 3.12 + OpenCV.
Database: SQLite/PostgreSQL for storing event history and generating reports (Web/Google Sheets).
Notifications: Aiogram (Telegram Bot API).

The code will be wrapped in Docker and set up for automatic restart in case of RTSP stream failures or computer reboots.

If desired, a web page can be implemented for viewing and exporting statistics for any period based on data from the database.

Show original

Translate

Hello! I am ready to complete this very interesting project.

Show original

Translate

Hello!
I am ready to develop a system that determines guest entry via RTSP cameras, sends notifications to Telegram, counts seated/walkaway, and generates conversion.

Show original

Translate

Hello, I have been developing in Python for over 5 years and there is a great library called OpenCV that I have been working with for a long time. It allows the use of models for face recognition, people, their poses, and the classification of this data. I also work with Telegram bots using the Aiogram library in Python. The task is complex, so it requires enough time to set up the infrastructure, but there is understanding in everything and there will be no problems in execution. I would be happy to discuss the details!

Show original

Translate

Hello, I developed a similar system but with slightly different tasks, I have a couple of questions for clarification of information. Is it a network of restaurants or just one? The price is conditional) I am waiting for you in personal messages)

Show original

Translate