⚡ electrical engineering

A Convex Envelope based IoT System for the Prevention of Baby Theft in Hospitals

This paper presents a low-cost, high-reliability IoT system for preventing newborn theft in hospitals, utilizing a multi-criteria architecture that combines RFID and GPS with convex hull modeling and an A* algorithm to achieve 100% detection of critical threats and zero false positives across 18 simulation scenarios.

Original authors: Meli Kevine, Franklin Tchakounté, Fréderic Raoul KIENTEGA, Audace Amen Vioutou DOSSOU-OLORY

Published 2026-07-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Meli Kevine, Franklin Tchakounté, Fréderic Raoul KIENTEGA, Audace Amen Vioutou DOSSOU-OLORY

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a hospital maternity ward as a giant, bustling maze. Usually, keeping a newborn safe in this maze relies on cameras and human eyes, which can get tired or miss a sneaky thief slipping through a crowd. But what if the hospital itself could "think," map out every safe path, and sound an alarm the second a baby takes a wrong turn?

That's exactly what Meli Kevine and their team at universities in Cameroon, Burkina Faso, and Benin have designed. They built a smart system called MoovMonitor to detect and prevent baby theft. Think of it as a high-tech "guardian angel" for newborns, but instead of wings, it uses a mix of digital maps, laser beams, and radio tags.

The "Smart Maze" Map

First, the team turned the hospital into a giant digital map, which they call a weighted graph. Imagine the hospital corridors are lines on a map, and every intersection or door is a dot. The system knows exactly how long it should take to walk from the nursery to the waiting room.

To make sure no one wanders off into the "danger zones" (like the basement or the exit doors), they used a mathematical trick called a Convex Hull. Picture a rubber band stretched tight around all the safe areas where a baby is allowed to go. As long as the baby stays inside the rubber band, they are safe. If the rubber band snaps because the baby is outside the allowed zone, the system knows something is wrong immediately.

The "Super-Brain" Algorithm

The brain of this operation is a special version of a pathfinding algorithm called A*. You might know it as the GPS in your phone that finds the fastest route. But this GPS is stricter. It doesn't just look for the fastest way; it looks for the safest way.

If a baby's carrier tries to take a shortcut through a restricted area, the system adds a massive "penalty" to that path, making it look like a dead end. The system calculates the perfect route and then watches the baby's movement in real-time. It checks four things constantly:

Who is carrying the baby? (Is it the mom or an authorized nurse?)
Are they on the right path? (Did they turn down the wrong hallway?)
Are they inside the safe rubber band? (Did they step outside the allowed zone?)
Are they moving at the right speed? (Did they stop for too long or run too fast?)

The "Laser Gate" Checkpoint

Here is where the system gets really cool. At every entrance and exit of the hospital services (like the maternity ward or the operating room), they installed laser barriers. Think of these as invisible tripwires.

If someone tries to walk through a door without scanning their ID card, the laser beam breaks. This triggers an immediate "Zero-Trust" check. The system demands an ID scan from the person carrying the baby. If the ID doesn't match the list of people allowed to move the baby at that moment, the system triggers an immediate alarm, sends notifications, and activates cameras to record the event. It's like having a bouncer at every single door who knows exactly who belongs where and instantly alerts security if rules are broken.

Did It Work? (The Simulation Results)

The team didn't just dream this up; they built a working prototype using a Raspberry Pi 5 (a tiny, powerful computer) and tested it with 18 different scenarios. These scenarios were like practice runs, simulating everything from a mom taking her baby for a walk to a thief trying to sneak a baby out.

The results were impressive, but remember, these were simulations:

100% Detection: The system caught every single simulated theft attempt. If the baby was in danger in the test, the alarm went off.
0% False Alarms: This is the big one. The system never cried wolf. It didn't panic when a mom was just walking slowly or when a nurse took a slightly longer route. It knew the difference between a mistake and a crime.
94.4% Accuracy: Overall, the system got it right almost every time in these tests.

One interesting moment happened in the tests: the system sometimes chose a "safer" path over the "fastest" path. In one case, the math said the fastest route went through a slightly less secure area, so the system suggested a different route. The authors say this isn't a bug; it's a feature. It shows the system prioritizes safety over speed, which is exactly what you want when protecting a newborn.

What It's Not

It's important to know what this system doesn't do. It doesn't use fancy biometrics like scanning a baby's palm print or ear shape (which can be tricky because babies move so much). It also doesn't rely on just one thing, like a simple radio tag that might get lost or a GPS that struggles indoors. Instead, it mixes RFID (radio tags), GPS, lasers, and cameras together so that if one part has a glitch, the others still have the baby covered.

The Bottom Line

This isn't a magic wand that has solved baby theft forever in every hospital on Earth. The authors are clear that this was a prototype tested in simulations. However, the results suggest that this "Convex Envelope" approach—combining a smart map, a strict rubber-band safety zone, and laser gates—could be a game-changer. It offers a way to protect newborns that is both high-tech and affordable, specifically designed for hospitals that might not have unlimited budgets.

In the world of hospital security, this system acts like a vigilant, math-loving guardian that never sleeps, ensuring that the only thing a baby ever does in the hospital is grow, not get stolen.

Technical Summary: A Convex Envelope based IoT System for the Prevention of Baby Theft in Hospitals

Problem Statement
Infant abduction from healthcare facilities represents a critical threat to patient safety and public confidence, with statistics indicating that a significant portion of child abductions occur directly within hospitals. Existing security solutions face notable limitations: Radio Frequency Identification (RFID) systems often lack precise real-time trajectory verification; Global Positioning System (GPS) solutions are frequently cost-prohibitive or susceptible to indoor interference; and biometric approaches (e.g., palmprints or ear shape) struggle with image quality and pose variations in real-world conditions. Furthermore, most current systems rely on simple zone alerts without integrating advanced behavioral analysis to assess movement compliance or carrier authorization. There is a distinct gap in multi-criteria security architectures capable of dynamically analyzing newborn movements against secure hospital mappings while remaining adaptable to the infrastructural and budgetary constraints of hospitals, particularly in Sub-Saharan Africa.

Methodology and System Architecture
The proposed solution is a four-tier Internet of Things (IoT) architecture designed to transform static surveillance into real-time monitoring. The system integrates hardware-based identification, spatial modeling, and algorithmic supervision.

System Architecture:
- Acquisition Layer: Utilizes a heterogeneous network including RFID tags for identity, GPS modules for localization, laser-based sensors for physical access control at service entrances, and surveillance cameras for visual context.
- Processing Layer: Hosted on a Raspberry Pi 5, this layer acts as the decision-making center, employing anomaly detection algorithms to correlate data streams and execute business logic.
- Data Layer: A central repository for historical logs, configuration parameters, and system states.
- Application Layer: Provides multi-modal responses, including audible alerts, digital notifications, and a supervision dashboard (MoovMonitor).
Hospital Environment Modeling:
- Weighted Graph Model: The hospital is modeled as a directed graph $G = (V, E)$ , where nodes represent Points of Interest (POIs) and edges represent authorized paths. Edge weights are calculated based on the minimum theoretical travel time ($Distance / Average Speed$).
- Convex Hull Delimitation: To strictly define safety zones, the system employs the Convex Hull concept. This geometric model ensures the newborn's position remains within the authorized polygon during inter-service transfers, detecting any deviation outside the legitimate path.
- Optimal Pathfinding (A Algorithm):* A modified A* algorithm is used to determine the optimal path. The evaluation function $f(n) = g(n) + h(n) + P(n)$ includes a Security Penalty Factor ( $P(n)$ ). If a node lies in a restricted area or violates security protocols, $P(n)$ approaches infinity, effectively flagging the trajectory as a high-risk anomaly.
Anomaly Detection Logic:
The system employs a multi-criteria decision logic based on four binary parameters:
- P1: Authorized Carrier (RFID verification).
- P2: Optimal Path Respected (Spatial deviation analysis).
- P3: Within Secure Zone (Convex Hull check).
- P4: Travel Time Correct (Temporal threshold check).
These parameters are processed through a Karnaugh map to derive a simplified Boolean function $S(t)$ . An alert is triggered if the function evaluates to a state indicating an anomaly (e.g., unauthorized carrier, zone exit, or time overrun). The system also incorporates a "Zero-Trust" mechanism where laser beam interruptions at service boundaries require immediate RFID re-verification.

Key Contributions

Multi-Criteria IoT Architecture: The synergistic integration of RFID for identification and GPS for localization, enhanced by laser-based physical access control, provides redundancy and increased accuracy.
Innovative Spatial Modeling: The application of weighted graphs combined with the Convex Hull concept to rigorously define and monitor authorized safety zones and paths.
Advanced Behavioral Analysis: The implementation of a modified A* algorithm that incorporates penalty mechanisms and time control to detect deviations, zone exits, and time overruns in real-time.
Contextual Intelligence: The system prioritizes security over pure mathematical optimization; for instance, it may reject a theoretically faster path if it traverses unsecured areas, favoring a safer route.

Results and Performance
The system was validated through 18 simulation scenarios covering standard operations and unauthorized intrusion attempts.

Detection Rate: The system achieved a 100% detection rate for critical threats (theft attempts).
False Positives: The system recorded 0% false positives, ensuring that normal hospital workflows were not disrupted by erroneous alerts.
Overall Accuracy: The system demonstrated an overall concordance of 94.4%.
Metrics: Precision was 100%, Recall (Sensitivity) was 91.67%, and the F1-Score was 95.65%.
Latency: Implementation on the Raspberry Pi 5 allows for trajectory updates in under 10ms, ensuring real-time responsiveness without processing lag.

Significance and Claims
The authors claim that this work offers a reliable, low-cost technological solution specifically adapted to the infrastructure constraints of African hospitals. By moving beyond simple zone alerts to a framework of advanced behavioral analysis, the system significantly improves newborn protection. The study emphasizes that the absence of false positives is a critical feature for clinical deployment, preventing "alarm fatigue." The proposed architecture serves as a proactive tool for preventing infant abduction, balancing high-performance detection with the practical realities of hospital environments. Future work suggested by the authors includes pilot testing in real hospital settings and the integration of machine learning for predictive analytics.