in this where ai is used in this project 👵 The p...

in this where ai is used in this project

👵 The problem

Elderly people may eventually live alone because their children move away for education, careers, jobs, or other responsibilities.

Living independently is valuable, but an emergency can become dangerous when nobody is physically nearby.

A fall, abnormal heart rate, low SpO₂, loss of consciousness, or another sudden event may go unnoticed.

Existing smartwatches provide individual health features, but our focus is elderly safety + continuous monitoring + intelligent emergency escalation.

⌚ Our solution

We propose a smartwatch-based elderly safety companion that continuously monitors important health and movement signals.

The watch connects the elderly user with predefined emergency contacts.

The system detects potentially dangerous events and communicates relevant information to the family.

The watch is designed around simplicity, because the target user should not have to navigate complicated technology during an emergency.

🧠 The core innovation

The system does not immediately assume that every sudden movement means the person has fallen.

First, it determines whether the watch is actually being worn.

Physiological and contact-related signals establish wear confidence.

Only when the system has sufficient confidence that a person is wearing the watch should motion be interpreted as a possible human event.

Therefore, a watch falling onto the floor should not automatically be treated as the elderly person falling.

The central question of our system is: “Did the watch fall, or did the person wearing the watch fall?”

❤️ Health monitoring

The watch continuously/periodically monitors heart rate.

The optical sensor can also estimate blood oxygen saturation, SpO₂.

The system checks signal quality before treating measurements as meaningful.

Abnormal measurements can trigger a verification process rather than immediately sending a panic alert.

The system can eventually establish a user's normal baseline and identify unusual deviations.

🧍 Fall detection

Accelerometer data identifies sudden changes in linear acceleration.

Gyroscope data identifies sudden changes in orientation and rotational movement.

The combination is much more informative than using an accelerometer alone.

A possible fall is detected from a movement pattern, not simply one sudden acceleration.

The system then checks whether the watch is actually being worn.

It can also look for post-event inactivity before escalating the event.

🚨 Emergency verification

After a suspected fall, the watch should first ask the user something like “Are you okay?”

The watch can vibrate, display a large confirmation screen, and optionally produce an audible/voice prompt.

The user gets a short window to cancel the emergency alert.

If the user confirms that they are okay, the alert is cancelled.

If the user does not respond, the event is escalated.

This extra verification layer helps reduce false emergency notifications.

👨‍👩‍👧 Emergency contacts

The user or caregiver can configure trusted emergency contacts during setup.

Multiple contacts can be stored rather than relying on one person.

An emergency notification can include the type of event detected.

The notification can include the user's latest health readings.

The notification can include the user's location if location functionality is implemented.

The goal is to give the family context, not merely a generic “Emergency!” notification.

📍 Location

The system can attach the user's location to a confirmed emergency.

Location can be obtained through the watch itself if it has GNSS, or through a paired smartphone depending on the prototype architecture.

Location should preferably be transmitted only when necessary to reduce unnecessary battery and privacy costs.

📱 Family/caregiver interface

A companion mobile application can display the elderly user's current status.

The dashboard can show heart rate, SpO₂, activity, battery status, and recent alerts.

Emergency events should be visually obvious and require minimal interpretation.

Caregivers can receive alerts without needing to constantly monitor the dashboard.

💊 Additional elderly-focused features

Medication reminders can notify the user at scheduled times.

Hydration reminders can encourage regular fluid intake.

Activity monitoring can identify prolonged inactivity.

A low-battery warning can prevent the safety system from silently becoming unavailable.

A connectivity warning can tell the caregiver when the watch loses communication.

A simple SOS button can allow the elderly user to manually request help.

Optional voice interaction can make the system easier for users with limited dexterity or vision.

🤖 AI component

AI should not be added merely to claim that the project uses AI.

The AI component can analyse combinations of sensor signals to classify events.

Instead of looking at one number, it can consider motion + orientation + physiological signal + activity + wear confidence.

The model can eventually learn the user's normal activity/physiological patterns.

An unusual combination of signals can receive a higher risk score.

Example: sudden acceleration + orientation change + inactivity + valid wearer signal = high-confidence possible fall.

AI can therefore function as the decision layer, while sensors provide the raw evidence.

🔋 Smart power management

Health sensors do not necessarily need to operate at maximum sampling intensity continuously.

The system can use low-power monitoring and increase sensing when an event is suspected.

The BMI270, for example, is specifically designed for low-power wearable applications and includes motion-triggered functions. �

Bosch Sensortec +1

This can help extend battery life.

🔐 Privacy and safety

Health information should be treated as sensitive data.

Data transmission should be encrypted in a real deployment.

Emergency contacts should be explicitly configured by the user/caregiver.

The prototype should clearly state that it is a safety/wellness system, not a medical diagnostic device.

Thresholds and alerts should be validated rather than presenting arbitrary values as medically definitive.