MoodSip is an intelligent hydration assistant designed to encourage optimal water intake through emotion-aware interactions. It leverages the Arduino Nicla Vision board to combine computer vision, environmental sensing, motion detection, and ambient temperature monitoring. Using AI-based emotion recognition, it provides proactive hydration reminders based on the user’s psychological and physiological state.
- Monitor the user’s presence and emotional state in real time.
- Suggest optimal hydration moments based on environmental and emotional data.
- Detect when the user lifts or tilts the bottle to confirm a drinking action.
- Adjust reminder frequency based on mood and stress indicators.
The user places the bottle on a desk during work. The system initializes an internal hydration timer, which accounts for:
- ambient temperature;
- duration since the last drinking action.
When the timer expires, the blue onboard LED on Nicla Vision flashes to prompt hydration. If MoodSip detects negative emotions (e.g., stress, sadness, fatigue), the timer interval is shortened dynamically to encourage earlier water intake.
The onboard accelerometer detects lifting and tilting motions, confirming that the user has drunk from the bottle. Detection resets the hydration timer.
The architecture integrates both edge and cloud AI components for balanced efficiency:
- Onboard Processing – Conducted locally on the Arduino Nicla Vision using optimized, quantized vision models.
- Cloud Processing (Companion App + Remote AI) – High-performance emotional analysis handled remotely through BLE connectivity.
To be defined – include a high-level block diagram illustrating the connections among camera, sensors, BLE module, and cloud API.
| Stage | Description | Technologies |
|---|---|---|
| 1. Local Monitoring | The Nicla Vision camera continuously observes the user and analyzes the surrounding environment. The onboard classification model detects if the user is present and positioned near the bottle. | Focoos AI SDK, Z-Ant |
| 2. Emotional Analysis (Cloud) | When paired with the Companion App, a cloud-based AI model (Gemini 2.5 Flash) performs deeper analysis of facial expressions to assess emotional states and determine if an early hydration reminder is required. | Gemini 2.5 Flash, BLE |
- MIT Indoor Scene Recognition Dataset (via Roboflow)
- Celebrity Face Image Dataset (via Kaggle)
Additional dataset augmentation and normalization details TBD.
BLE is used to exchange:
- presence and stress detection results;
- hydration timer state;
- user acknowledgment of prompts.
- Trained using Focoos AI SDK.
- Quantized to 8-bit precision for efficient embedded inference.
- Deployed on Nicla Vision via Z-Ant Deployment Toolkit.
- Model inference time: TBD (approx. 200–400 ms/frame).
| Layer | Technology | Primary Function |
|---|---|---|
| Firmware | Arduino IDE | Sensor logic, LED control, BLE communication |
| Computer Vision | Focoos AI SDK | User and scene detection |
| Vision Deployment | Z-Ant | Optimized deployment of quantized models |
| Cloud AI | Gemini 2.5 Flash | Emotion recognition and stress inference |
| ID | Requirement | Priority | Verification Method |
|---|---|---|---|
| FR-01 | Detect user presence within 2 seconds of camera activation. | High | Real-time testing |
| FR-02 | LED must flash for 5 seconds when hydration reminder is triggered. | Medium | Visual inspection |
| FR-03 | BLE connection must transmit data packets every 10 seconds. | High | BLE log analysis |
| FR-04 | Timer must decrease by 30% upon stress detection. | High | Model simulation |
| FR-05 | System must reset timer upon drinking detection. | High | Accelerometer event logs |
| FR-06 | User emotion classification accuracy ≥ 85%. | High | Dataset benchmark |
| FR-07 | Ambient temperature sampling every 30 seconds. | Medium | Sensor output log |
| Category | Requirement |
|---|---|
| Performance | Emotion inference latency < 500 ms/frame. |
| Energy Efficiency | Average consumption < 500 mW (TBD). |
| Reliability | False positive rate in face detection < 10%. |
| Maintainability | OTA or USB firmware updates supported. |
| Scalability | Cloud AI must support at least 1000 concurrent users (TBD). |
| Security | BLE connection encrypted using AES-128. |
| Privacy | No image storage; only anonymized feature vectors sent to the cloud. |
| Usability | LED notifications must be visible under standard office lighting. |
- Arduino Nicla Vision Board
- Integrated camera module
- 6-axis accelerometer and gyroscope
- Temperature sensor
- RGB LED
- Power source (battery or USB-C, TBD)
- Optional: environmental humidity sensor
- Arduino IDE (Firmware)
- Focoos AI SDK (Model training)
- Z-Ant (Model deployment)
- Gemini 2.5 Flash (Cloud inference)
- BLE API (Device communication)
- Companion App framework: TBD (Flutter / React Native / Kotlin)
| Element | Description |
|---|---|
| Service UUID | TBD – Hydration Service (custom) |
| Characteristics | Hydration Timer, Emotion State, Temperature, Device Status |
| Data Format | JSON or binary (TBD) |
| Update Frequency | 10 seconds |
| Encryption | AES-128, BLE 5.0 compliant |
- Endpoint: TBD (Gemini API or custom microservice)
- Data exchanged: anonymized emotion features and temperature context.
- Response: stress probability and hydration adjustment recommendation.
- Power on and BLE connection established.
- Camera detects user presence.
- Timer starts and tracks hydration interval.
- Emotion analysis performed (onboard + cloud).
- Timer dynamically updated based on stress level.
- LED prompts hydration reminder.
- Accelerometer confirms drinking event → timer reset.
Detailed flowchart TBD.