Recognize household waste, route it into one of four compartments, and display each compartment’s fill level. A team capstone combining YOLOv8s, Python, and a finite state machine on Raspberry Pi 4 Model B.
Project recognition
Best Capstone Design Project
Batch 1 · July 2023
The challenge
The project addressed household waste sorting with four compartments: organic, inorganic, plastic, and mixed waste. The challenge was to coordinate camera capture, classification, motor positioning, lid control, and capacity measurement in one repeatable workflow.
My contribution focused on integrating the software and hardware and implementing the system flow in Python on the Raspberry Pi.
The prototype used a controlled capture chamber with LED lighting and 640 × 480 camera images. Its object labels were mapped to the project’s four waste categories.
Interactive 3D
Powered by Raspberry Pi 4B, with camera-based recognition and motor-driven sorting. Inspect the hardware or follow a waste category through the system.
Preparing the 3D illustration…
Choose a waste category
Fruit and vegetable waste, such as apples, bananas, and spinach.
An illustrated journey from camera to compartment.
Python coordinates the Pi Camera, YOLOv8s recognition, NEMA 17 motor, MG996R servo, and capacity feedback.
Schematic hardware illustrations based on the project paper, with simplified shapes and placement. Colors identify the four waste categories; the photos below show the physical build.
System overview
The Raspberry Pi connects the recognition model with the sorting mechanism and the display. A finite state machine coordinates when each part runs.
YOLOv8s · Pi Camera
An illuminated chamber captures 640 × 480 images. Recognized object labels are mapped to organic, inorganic, plastic, or mixed waste.
Python FSM · NEMA 17 · MG996R
The state machine coordinates the rotating chamber, servo-driven lids, limit-switch calibration, and supporting door solenoid to route waste.
HC-SR04 · ST7735 display
Calibrated ultrasonic sensing estimates the fill level. The 1.8-inch display shows the active process, classification result, and compartment capacities.
Control workflow
Select a stage to see how the prototype moves from loading waste to updating the capacity display.
Stage 01
The system checks components, closes the lids, checks the camera, loads the recognition model, calibrates the distribution motor, and measures the initial compartment capacities.
Workflow summarized from the project paper.
Inside the prototype
Mechanical and camera integration
The paper shows the enclosure, top lid with LED lighting, rotating mechanism, and camera chamber above the bottom lid.
Project paper · Figure 6 · Select the image for full size
Object detection examples
Detection examples for cardboard, a can, a bottle, and tissue. The bounding-box confidence labels belong to these individual detections.
Project paper · Figure 11 · Select the image for full size
My contribution
The result
The team delivered a working system that recognizes waste, distributes it into the selected compartment, and updates the capacity display. The paper reports 87% classification accuracy in the prototype tests, with the detailed test scope below.
Recognized as Best Capstone Design Project, Batch 1, July 2023.
Measured results
Tables 2 and 3 separate single-object and multiple-object classification. Together, they record 78 correct classifications in 90 trials, approximately 86.7%; the paper rounds the accuracy to 87%.
| Test setup | Correct / trials | Reported accuracy | Average classification time |
|---|---|---|---|
| Single object · Table 2 | 52 / 60 | 87% | 7.344 seconds |
| Multiple objects · Table 3 | 26 / 30 | 87% | 7.292 seconds |
Capacity monitoring · Table 6
At fill heights from 0 to 35 cm, the paper reports an average height error of 0.894 cm and a maximum error of 1.53 cm. The calibrated HC-SR04 measurement is converted to a fill percentage for the display.
Test scope
The tests used the paper’s selected object labels and a chamber with controlled lighting. Some cardboard, paper, and tissue inputs were misclassified, and some plastic inputs were undetected. Multi-object inputs could also map to the wrong waste category.
Project demonstration
Watch the team’s project demonstration on YouTube.
Research basis
Implementation of Smart Trash Can for Household Scale Waste Management Using Raspberry Pi 4 B Model Using Finite State Machine (FSM) Method
Adhi Muhammad Fahmi, Zidan Surya Gibran, Husneni Mukhtar, and Willy Anugrah Cahyadi · Telkom University · July 2023 manuscript.
Figures 6 and 11 show the prototype and recognition examples. Tables 2–3 provide classification results; Table 6 provides capacity measurements.
Let's Connect with me on LinkedIn
or email me about your project