TeaTime: Local AIoT Tea Service Detection System

What's my project about?

TeaTime is a context-aware AIoT system designed to solve a simple yet persistent office problem: missing the tea or coffee service while deeply focused or wearing noise-canceling headphones. It serves as a personal "lookout," using computer vision to detect the specific uniform of service staff and instantly notifying you on a dedicated desktop display.

This project bridges the gap between high-performance Edge AI and practical, daily utility. It operates entirely on a local network (no cloud dependencies), ensuring privacy while delivering real-time alerts.

Why did I decide to make it?

In many office environments, the "tea break" is a crucial ritual for productivity and social connection. However, when you are "in the zone" -coding, writing, or in deep work mode, it's easy to tune out the background noise of the service cart passing by.

 I built TeaTime because:

1.  Focus vs. FOMO: I wanted to maintain deep focus without the "Fear Of Missing Out" on coffee or tea.
2.  Privacy-First AI: I wanted to demonstrate that useful AI doesn't need to send video feeds to the cloud. All processing happens locally on the edge device.
3.   Hardware Integration: It was a perfect opportunity to combine powerful Edge AI processors (like the RDK X5) with lightweight IoT microcontrollers (ESP32) in a practical application.  

How does it work?

The system consists of two main hardware nodes communicating over a local WiFi network:

 1. The Edge AI Node (The "Brain")

 This node uses an RDK X5 (Horizon Robotics) developer board equipped with a camera. It runs a ROS2 (Robot Operating System) pipeline that:

  > Detects People: Uses an FCOS (Fully Convolutional One-Stage) object detection model to identify humans in the corridor.

  > Verifies Uniforms: Extracts the upper body of detected persons and analyzes the color spectrum (HSV) to confirm they are wearing the specific purple uniform of the tea service staff.  

  > Checks Context: Validates detections against specific time windows (e.g., 10:00-12:00 and 14:30-16:00) to filter out false positives.

 Code Snippet (Python - Edge Node):

  1 # Logic to check uniform color and trigger alert

  2 if is_uniform and self.is_time_in_window():

  3   self.get_logger().info(f"[DETECT] Tea Staff detected! Purple %: {percent:.2f}%")

  4   self.send_alert(confidence)

 2. The IoT Alert Node (The "Notifier")

 This is a TTGO T-Display (ESP32) sitting on the user's desk. It connects to WiFi and runs a lightweight HTTP server.

•   Idle Mode: Displays a digital clock and system status.
•   Alert Mode: When it receives a POST request from the Edge Node, it flashes the screen purple and displays "TEA ARRIVED" along with the timestamp.

 Code Snippet (C++ - IoT Node):

  1 // Handling the alert on ESP32

  2 if (event && strcmp(event, "tea_service_detected") == 0) {

  3   drawAlertScreen(timestamp); // Flashes the screen

  4   server.send(200, "text/plain", "Alert Received");

  5 }

 System Flow

  1. Camera captures the hallway.

  2. Edge Node detects a person + purple uniform.

  3. Edge Node sends POST /alert to the ESP32 IP address.

  4. ESP32 flashes the notification on your desk.

 Hardware Used:

  * Compute: Horizon Robotics RDK X5 (Edge AI)

  * Microcontroller: TTGO T-Display ESP32

  * Camera: MIPI / USB Webcams

  * Software: ROS2 , OpenCV, PlatformIO