Fall Detection System (Graduation Project)
An end-to-end smart healthcare platform that combines wearable IoT hardware, AI-based fall prediction, a FastAPI backend, a React Native mobile app, and an admin dashboard to support detection, monitoring, and faster emergency response.
A stronger hero that frames this project like a real case study instead of a plain information page.
Impact
Overview
This graduation project was designed as a full ecosystem rather than a single model or a simple app. The goal was to build a connected healthcare product that links wearable sensing, AI-powered prediction, backend decision logic, mobile monitoring, and caregiver support into one coherent system.
The platform starts with an ESP32-based wearable bracelet, uses AI sequence models for fall detection and early fall-risk prediction, stores and coordinates data through a FastAPI backend, and exposes the experience through a mobile app and an admin dashboard. That combination made the project much closer to a real product than an isolated academic prototype.
Problem
Falls are a major health risk for elderly users and people with mobility or neurological challenges, and delayed response can lead to serious complications.
Many existing systems are expensive, fragmented, or focus on only one part of the problem, such as detection without caregiving context or monitoring without reliable device integration.
We wanted to solve not only fall detection, but also early risk prediction, remote monitoring, emergency communication, BLE-based device setup, and practical caregiver workflows in one usable system.
Key Features
- ESP32 wearable bracelet with BLE provisioning and Wi-Fi connectivity
- Dual AI outputs for immediate fall detection and near-future fall risk prediction
- FastAPI backend for authentication, device management, alerts, predictions, and care relationships
- React Native mobile app for monitoring, emergency contacts, device connection, and caregiver workflows
- Admin dashboard for users, devices, alerts, and system-level visibility
- Role-based control rules so caregivers can monitor without taking over bracelet ownership
Timeline
Defined the healthcare problem, system boundaries, and the end-to-end product flow across hardware, AI, backend, and mobile.
Built the ESP32 wearable path and the motion-based prediction pipeline for immediate detection and early risk forecasting.
Connected device ownership, alerts, emergency contacts, care links, and prediction handling through backend APIs and business rules.
Refined BLE provisioning, owner-only bracelet control, caregiver visibility, localization, and safer alert handling.
Extended the platform with an admin dashboard for system visibility, user/device monitoring, and operational oversight.
User Workflow
- The bracelet is provisioned from the mobile app using Bluetooth Low Energy, which sends Wi-Fi credentials, device identifiers, and communication settings to the firmware.
- Once connected, the device streams structured telemetry that can be monitored, stored, and used for prediction workflows.
- The backend processes user, device, and care-management logic, stores alerts and health-related data, and coordinates prediction outcomes and system rules.
- The mobile app allows users to log in, manage their profile, connect the bracelet, review alerts, set emergency contacts, and participate in caregiver-linked monitoring flows.
- When risk or alert conditions are detected, the platform supports emergency-oriented follow-up through contacts, care links, and monitoring visibility.
Architecture
- The wearable hardware acts as the sensing and device endpoint layer, responsible for provisioning, connectivity, and structured reporting.
- The AI layer handles time-series motion analysis using models such as LSTM, BiGRU, and hybrid sequential architectures.
- The backend acts as the operational brain of the platform, enforcing business rules around authentication, ownership, care relationships, predictions, alerts, and system coordination.
- The mobile app is the primary user-facing interface, translating complex system behavior into practical monitoring and emergency workflows.
- The admin dashboard adds operational oversight and makes the platform feel like a complete product instead of a one-user demo.
Technical Focus
- Improved BLE provisioning behavior so the app and firmware agree on payload format, status reporting, and failure handling.
- Worked on device ownership protection so the same bracelet cannot be silently claimed by multiple users.
- Refined owner-only bracelet actions, separating device administration from caregiver monitoring permissions.
- Improved network status feedback, device removal and re-linking, repeated alert handling, and overall mobile reliability in real-world conditions.
- Strengthened project documentation and architecture explanation so the system is easier to understand as a full product.
Challenges Solved
- Aligning firmware, BLE payloads, mobile state, backend ownership rules, and AI workflows without fragile integration points.
- Handling repeated alert loops, misleading connection failures, and network-related edge cases in ways that do not overwhelm the user.
- Designing permission boundaries so caregivers have visibility without being allowed to interfere with the owner's hardware controls.
- Building for failure scenarios such as Wi-Fi issues, provisioning retries, device resets, translation inconsistencies, and reconnect flows.
Deployment
- The platform included backend APIs, mobile monitoring flows, hardware communication, and dashboard visibility as parts of one integrated product architecture.
- The project was approached as a real operational system, with attention to structured status reporting, system health behavior, and administrator visibility.
Outcome
The final result is a strong portfolio project that demonstrates embedded systems integration, AI inference workflows, backend logic, mobile UX, and caregiver-oriented product thinking in one system.
It also shows my ability to work across layers, improve real-world reliability, enforce safe business rules, and turn a technically complex prototype into a more coherent end-to-end platform.