AI video monitoring: cameras that report by themselves
Most CCTV only records, and nobody watches the footage until after an incident. MirasWatch sits on the same cameras and the same recorder, analyses the picture live and sends a message for the events you have chosen.
- Since 2014
- Four NVIDIA A100 systems in Tehran
- No VPN needed
Smart monitoring on the cameras you already own
People and vehicle detection
The system separates people, vehicles and animals in the live picture; alerts are based on that detection, not on any movement in the frame.
Crossing direction, entry and exit
A virtual line is drawn on each door; the system says that someone came in or went out, not just that something moved.
Several cameras, one journey
Sightings on neighbouring cameras are joined by camera adjacency and time into one journey: from the yard gate to the lobby and the floor.
Alerts with a picture on your phone
An alert with one picture and the place name arrives in the Bale messenger; the sightings of one journey are grouped into one message.
Questions about the picture
An open vision-language model on the same on-site server answers simple questions about a frame: a person or a dog? A man or a woman? Carrying a parcel? The answer is a signal, not a verdict.
Processed on site
All analysis runs on a GPU server inside your own building; camera footage does not leave the building; only the alert picture is sent to the messenger.
Miras infrastructure →How it works
- Your cameras and recorder (NVR) stay; the system takes the video from the local network.
- A GPU server is installed on site and analyses the camera video.
- Zones and crossing lines are defined per camera: entrance door, yard, car park.
- Events are joined into journeys, and you get a message only for the journeys you have chosen.
- Your answer to each message (“known person”, “courier”) is recorded, and the rules are measured and corrected against those answers.
Where it helps
Today it runs in Miras’s own building. The same functions fit these places:
- Residential and office buildings: entry and exit at doors, yard and car park.
- Construction sites: people and vehicles passing the site gates. AI in construction
- Factories and warehouses: entrance and loading doors. Industrial AI
What it does and what it does not do
What runs every day in Miras’s own building
People, vehicle and animal detection; crossing direction at a door; cameras linked into one journey; alerts with a picture; questions about a frame. We built this system for our own building first and work with it every day.
What we do not promise
We do not currently offer reading of Persian number plates. Face recognition is possible only where a camera sits at face height, not from a ceiling camera, and it is always one signal among several. Automatic parcel-delivery detection is not yet accurate enough for us to promise. The system does not replace a guard; it helps the guard.
A builder that is its own first customer
MirasWatch rules are measured and corrected against the building owner’s real answers to the messages. Because we work with the system every day ourselves, we are also plain about what it cannot do yet.
- Tested on Hikvision cameras and recorders; other brands are not tested yet and are tested before we start.
- Footage stays in your building; the only thing that leaves is the alert message and its picture.
- Open models on your own hardware; no foreign subscription fee.
Machine vision across Miras
Miras also works on construction site monitoring and industrial machine vision; the vision-language model is described on the Persian LLM and GPU infrastructure page.
Frequently asked questions
What is AI video monitoring, and how is it different from ordinary CCTV?
Ordinary CCTV records and notifies on any movement. AI video monitoring analyses the picture: it separates people from vehicles and animals, knows the crossing direction and messages you for the events you have chosen. In MirasWatch that intelligence sits in a server beside your existing cameras, not inside the camera.
Do I have to replace my cameras?
If your cameras and recorder are networked Hikvision units, usually not; the system has been tested on that equipment. Other brands are not tested yet and are tested before we start. Analogue cameras and cameras with no network stream are not supported today.
Where is the footage processed and kept?
Processing runs on a server in your own building, and recordings stay on your own recorder. The analysis needs no internet connection and footage does not leave the building; the internet is needed only for the alert message and its picture to reach your phone.
Does it read number plates?
No. We do not currently offer Persian number-plate reading. The system detects a vehicle and records its entry and exit, but does not read the plate number.
Does it recognise faces?
Only where a camera is mounted at face height and close range, such as beside an entrance door, and even there it is one signal, never the only basis for a decision. Ceiling cameras looking down do not give a reliable face, and we do not promise one.
Where do we start?
With a few cameras and one clear need, such as the entrance door or the car park. After you see the result on your own video, you decide about the rest.
Bring intelligence to your plant or project
We start with a data audit: where your data is, where cost leaks, and which model pays back first. Tehran · Isfahan · Noshahr · Dubai.
- Your call or messageOne line about your industry and the problem is enough.
- Data-audit sessionWe look at the data you have and which model pays back first.
- Pilot proposalOne measurable indicator, a fixed scope, a fixed time.
Farmanieh, Paramis Tower, 7th floor, unit 702, Tehran
or use the contact form (Persian-first page) →