A gold-shop security system detecting weapons with YOLO, violence with an RNN and theft with MediaPipe, plus PIR and flame sensors on NodeMCU and Telegram alerts.
An academic project to develop an intelligent security system for protecting gold shops that combines computer vision, deep learning, physical sensing, and IoT, carried out by a team of students with technical assistance from Techno Enjaz. The system analyzes the surveillance camera feed to detect three kinds of threat: weapons with the YOLO algorithm, violent behavior with a recurrent neural network (RNN), and theft attempts such as reaching into drawers, using body-pose estimation with the MediaPipe library.
Alongside the camera, a PIR motion sensor and a flame sensor monitor abnormal movement and fire indicators through a NodeMCU ESP8266 controller. When any threat is detected, the system sounds an alarm inside the shop, saves a video clip of the event, and sends an alert to the shop owner through a Telegram bot describing the incident. The project documents these functions working with test screenshots, without publishing numerical accuracy metrics.
| Item | Details |
|---|---|
| Project type | Intelligent surveillance system combining AI video analysis with physical sensors |
| Field | Computer vision, deep learning, IoT, security systems |
| Project status | Working prototype with practical tests documented by screenshots |
| Techno Enjaz role | Technical support and assistance to the students during project development |
| Core technologies | Python, YOLO (pre-trained Ultralytics models), RNN, MediaPipe, OpenCV, NodeMCU ESP8266, HC-SR501 PIR sensor, flame sensor, Telegram Bot |
| Outputs | Weapon, violence, and theft detection, motion and flame detection, local alarm, saved event video clips, Telegram alerts |
Gold shops are among the commercial premises most exposed to theft, because they hold high-value goods that are easy to liquidate. Yet many rely on traditional means: cameras that record without analyzing, or alarms that trigger on a mechanical fault without interpreting the nature of the threat. In many cases the owner learns of an incident only after it has happened.
The project started from the need for a low-cost, easy-to-install system that moves protection from after-the-fact documentation to early detection and instant alerting, by analyzing the scene in real time and combining image, motion, and flame data in one system.
The project was carried out academically by the students, with technical assistance and support from Techno Enjaz during development. It is presented among the office's work as an example of its support for projects that integrate computer vision models with IoT controllers and sensors in practical security applications.
According to the project's component integration diagram:
| Module | Technology | What it detects |
|---|---|---|
| Weapon detection | YOLO via pre-trained Ultralytics models (such as YOLOv5 and YOLOv8) | A weapon in the frame, marked with a box and a label such as "knife" |
| Violence detection | Recurrent neural network (RNN) | Violent patterns in motion sequences within the video |
| Theft detection | MediaPipe pose estimation (33 body landmarks, 21 hand landmarks) | A hand entering a predefined restricted region (ROI) such as drawers or display cases |
The project used ready-made YOLO models from Ultralytics rather than building a model from scratch, customizing them for the project's needs.
| Component | Documented specifications | Role in the system |
|---|---|---|
| NodeMCU ESP8266 | 32-bit Tensilica L106 processor, 2.4 GHz Wi-Fi (802.11 b/g/n), 11 usable GPIO pins | Reads the sensors, coordinates, and sends alerts over the network |
| HC-SR501 PIR motion sensor | Detection range up to about 7 meters, adjustable sensitivity, delay from 1 second to 3 minutes | Detects any entry or abnormal movement in or around the shop |
| Infrared flame sensor | Detects wavelengths of 760–1100 nm over about 60 degrees, with a potentiometer for sensitivity | Early detection of fire |
| Surveillance camera | Video feed to the Python unit | Frame source for the detection algorithms |
The project presented a set of tests, documented with screenshots, to show the components working together:
These tests show the functions working in the scenarios presented, but the project publishes no figures for weapon or violence detection accuracy, false-alarm rates, or response time, and does not state the size of the test data. Percentages mentioned in the project's theory chapters, such as 97.4% accuracy for an RNN model on the UCSD Ped2 dataset, are results of earlier studies, not of this system.
According to the directions set out in the project:
The system analyzes video of people inside the shop and stores clips of it, so any real deployment requires informing customers and staff that monitoring is in place, controls over clip retention and access rights, and compliance with local video-surveillance regulations.
If you are working on an AI-based security surveillance system and need technical help developing the idea or building a prototype, you can contact the Techno Enjaz office to discuss your project's requirements and the appropriate scope of support.
An academic system built by a team of students with technical assistance from Techno Enjaz to protect gold shops. It analyzes the camera feed with AI to detect weapons, violence, and theft attempts, adds motion and flame sensors via a NodeMCU ESP8266 controller, then raises an alarm and alerts the shop owner through Telegram.
It uses pre-trained Ultralytics YOLO models to mark weapons such as a knife with a box and label, and a recurrent neural network (RNN) to analyze motion sequences for violence. Theft is detected through MediaPipe body and hand pose estimation, triggering an alert when a hand enters a predefined restricted region such as drawers.
The HC-SR501 PIR sensor detects any entry or abnormal movement within a range of up to about 7 meters, and the infrared flame sensor detects early signs of fire. The NodeMCU ESP8266 reads these sensors, coordinates them with the software analysis outputs, and sends alerts over Wi-Fi.
The system sounds an alarm inside the shop and saves a video clip of the event to a dedicated folder. A Telegram bot then sends a message such as "Weapon detected!" or "Violence detected!" with the detection time and a video link, so the owner can respond even when away.
The project documents the functions working with test screenshots but publishes no figures for detection accuracy, false-alarm rates, or response time. It notes the need to improve performance in difficult lighting, varied angles, and crowded spaces, and the percentages in its theory chapters belong to earlier studies.
Reach out to Techno Enjaz engineering team to discuss technical feasibility, prototype hardware, and development scope.
Discover more engineering prototypes and academic implementations
A children's learning system that uses a camera to recognize animals with YOLO11, colors with HSV, and numbers with OCR, then shows and speaks the result in Arabic.
A computer vision system that detects knives and pistols in real time from a laptop or phone camera using two trained YOLOv8 models, Python, and OpenCV.
An interactive learning system for children that recognizes objects via the camera with YOLOv8, speaks their names, detects colors with HSV, and labels uploaded images.
