main detection loop for EcoKiosk plastic classification process
This is the final product prototype being used to scan a plastic items. The machine uses comptuer vision to seperate recyclable items apart.

EcoKiosk: AI-Powered Recycling Machine for Schools

Problem Statement

What real-world problem are you solving?

Students in schools across Northern Dallas are not recycling. Every lunch period, plastic bottles, food containers, and packaging end up in the trash. The issue is not a lack of knowledge. Most students understand what recycling is. The problem is indifference. Traditional recycling bins do nothing to engage students. There is no feedback, no recognition, and no incentive. Without a reason to act differently, most students choose the easiest option and throw everything away.

The data reflects what is seen in school cafeterias every day. Texas disposed of more than 41.3 million tons of municipal solid waste in fiscal year 2024 across 206 active landfills. The average Texan produces 7.24 pounds of trash per day. The state recycling rate is about 22.7 percent across all materials, but plastic recycling is much lower. Nationwide, only about 5 percent of plastic waste is actually recycled, down from 9 percent in 2018. The infrastructure exists, but behavior has not caught up. [7][9][10][11]

Which UN Sustainable Development Goal does it address?

SDG 12: Responsible Consumption and Production

Target 12.5 — By 2030, substantially reduce waste generation through prevention, reduction, recycling, and reuse.
Target 12.8 — By 2030, ensure that people everywhere have the relevant information and awareness for sustainable development and lifestyles in harmony with nature.

EcoKiosk directly supports SDG 12.5 by turning recycling into an active and rewarding experience instead of a passive option. It also supports SDG 12.8 by helping students learn what is recyclable through real-time interaction instead of posters or lectures. [5][6][14]

Your Solution

What is your idea or tool?

EcoKiosk is an AI-powered recycling kiosk that rewards students for recycling. A student holds a plastic bottle or container in front of the camera. The system checks whether it is a real recyclable item. The student then enters a name on the touchscreen and immediately earns reward points. The full interaction takes about 10 seconds. The hardware cost is under $150, which makes the system realistic for schools to adopt.

How does it use AI?

EcoKiosk uses a YOLOv8 computer vision model trained on recyclable plastics. The model runs locally on a Raspberry Pi 5 using INT8 quantization for fast edge inference. It identifies three common categories of recyclable plastics:

Resin Code	Name	Examples
PET (1)	Polyethylene Terephthalate	Water bottles, soda bottles
HDPE (2)	High-Density Polyethylene	Milk jugs, detergent bottles
PP (5)	Polypropylene	Yogurt containers, bottle caps

The AI system includes several reliability features:

Dynamic brightness normalization to handle different cafeteria lighting conditions
Center-region detection filtering to reduce false positives
Temporal persistence checking so the item must remain visible before points are awarded
Confidence thresholding at c >= 0.75 to ensure accuracy

Everything runs locally on the Raspberry Pi 5. No cloud connection is required. [1][2][4][12]

Impact

How will your solution help people or the planet?

Increases recycling by giving students an immediate reason to participate
Reduces plastic waste sent to landfills in the Dallas Fort Worth area
Teaches students what materials are recyclable through direct interaction
Keeps costs low enough for public schools with limited budgets
Works fully offline, which makes it reliable and easier to maintain
Can be replicated by other schools as a low-cost model

Why is it meaningful by 2030?

SDG Target 12.5 focuses on reducing waste generation through recycling and reuse by 2030. Texas alone disposed of more than 41.3 million tons of waste in 2024, and plastic recycling rates remain extremely low. One of the biggest barriers is not infrastructure alone. It is behavior, especially among younger generations. [6][7][9][10]

EcoKiosk focuses directly on that behavior gap. If deployed across schools in the Dallas Fort Worth metroplex and later across Texas, it could help build recycling habits early. Students who form these habits in school are more likely to carry them into adulthood. Over time, that creates a compounding effect. Each recycled item may seem small, but at scale it leads to meaningful environmental impact.

Technical Report

Title Page

Field	Details
Project Name	EcoKiosk: AI-Powered Recycling Station for Schools
Team Member	Sofia Bouazizi
UN SDG	SDG 12: Responsible Consumption and Production
Project ID	Env1498
Category	Environmental Science

Abstract

EcoKiosk is a low-cost, AI-powered recycling kiosk designed to increase student participation in recycling in schools across Northern Dallas, Texas. Texas disposed of more than 41.3 million tons of municipal solid waste in 2024, while only about 5 percent of plastic is recycled nationwide. EcoKiosk addresses this gap by using a YOLOv8 computer vision model running on a Raspberry Pi 5 to identify common recyclable plastics in real time. When a valid recyclable is detected with high confidence, the student enters a name on a touchscreen and instantly receives reward points.

Key technical features include INT8 model quantization for fast edge inference, brightness normalization for variable lighting, and a simple touchscreen flow designed for real-world use. The complete system costs under $150, making it practical and scalable for schools. EcoKiosk supports UN SDG Target 12.5 by reducing waste through recycling and Target 12.8 by increasing awareness through everyday interaction. [1][4][5][6][7][10]

Introduction

What is the problem?

Plastic waste is one of the clearest and most persistent environmental challenges in the United States. In Texas, the issue is especially significant because of the state’s size and continued population growth. Texas disposed of over 41.3 million tons of municipal solid waste across 206 active landfills in fiscal year 2024. The average Texan produces 7.24 pounds of trash per day. While the state’s total recycling rate is around 22.7 percent, plastic recycling is much lower. Nationally, only about 5 percent of plastic waste is actually recycled, which is down from 9 percent in 2018. [7][9][10][11]

The problem is especially visible in schools. Every lunch period, plastic bottles and food containers go into the trash. Students are not necessarily against recycling. In many cases, they are simply not motivated to do it. Traditional recycling bins provide no feedback, no recognition, and no reason to care. Without some kind of incentive, most students default to convenience.

Why is it important?

Texas is growing rapidly, and waste volumes continue to increase with population growth. Although landfill capacity still exists, that capacity is not evenly distributed and will not last forever. Landfills also contribute to greenhouse gas emissions, including methane. Without changing behavior, better infrastructure alone will not close the recycling gap. [7][9]

What UN SDG target are you trying to achieve?

EcoKiosk supports SDG 12: Responsible Consumption and Production. Specifically, it addresses Target 12.5, which focuses on reducing waste through recycling and reuse, and Target 12.8, which focuses on improving awareness and education around sustainable lifestyles. By engaging students daily through an interactive AI system, EcoKiosk helps build recycling habits and material awareness at a young age. [5][6][14]

Background Research

What research did you do?

Research for EcoKiosk focused on three main areas: waste management statistics in Texas, behavioral research on recycling incentives, and edge AI deployment on low-cost hardware.

Waste statistics: Data from the Texas Commission on Environmental Quality, the U.S. Environmental Protection Agency, and the CleanHub Plastic Perspectives Report showed that Texas disposed of 41.3 million tons of waste in FY2024, the average Texan generates 7.24 pounds of trash each day, and only about 5 percent of plastic is recycled nationwide. [7][9][10][11]

Behavioral research: Studies on gamification and incentive-based behavior change show that immediate rewards are much more effective than delayed or abstract encouragement, especially for students. Point systems and visible rewards are more likely to influence daily behavior than passive reminders.

Edge AI research: Technical research focused on deploying YOLOv8 on ARM-based hardware, using INT8 quantization to improve speed and efficiency, and building an OpenCV-based preprocessing pipeline that could perform well under inconsistent lighting conditions. [1][2][4][12]

What existing tools or data did you explore?

YOLOv8 by Ultralytics for real-time object detection
OpenCV for camera input and preprocessing
PyQt5 for the touchscreen interface
Raspberry Pi OS for deployment on low-cost hardware
A custom image dataset of approximately 2,000 labeled images of PET, HDPE, and PP plastics
TCEQ waste data for official Texas disposal and recycling statistics
EPA plastics data for national recycling rates by material type

Solution Description

Explain your idea clearly

EcoKiosk is a standalone system that verifies recyclable items and rewards students instantly. A student shows a plastic item to the camera. The YOLOv8 model classifies the item in real time. If the item is identified as PET, HDPE, or PP plastic with confidence of 0.75 or higher, and the detection remains stable briefly, the system switches to a name-entry screen. The student enters a name, and points are immediately stored in a local SQLite database. The full interaction takes about 10 seconds.

What technologies and algorithms did you use?

Technology	Role	Why This Choice
YOLOv8n	Object detection	Strong accuracy to speed ratio for edge deployment
INT8 Quantization	Model optimization	Smaller model and faster inference on ARM CPU
OpenCV	Camera input and preprocessing	Standard computer vision library with Raspberry Pi support
PyQt5	Touchscreen UI	Responsive and reliable Linux GUI framework
SQLite3	Points storage	Embedded local database with no server needed
NumPy	Image normalization	Fast array operations for preprocessing

How does AI help solve the problem?

AI is what makes EcoKiosk practical. Without AI, every item would need to be checked manually, which would make the system slow and inconvenient. The model allows the kiosk to automatically identify recyclable plastics in real time without supervision. It also helps reduce false positives by combining image preprocessing, confidence thresholds, and persistence checks. Because all inference runs locally on the Raspberry Pi, the system remains fast, private, and reliable even without internet access. [1][2][4][12]

Design and Implementation

Architecture and System Design

EcoKiosk uses a modular three-layer design for simplicity, reliability, and low cost.

Layer	Components	Technology
Presentation	Touchscreen UI, camera preview, success screen	PyQt5
Application	State machine, frame loop, points manager	Python 3.11
AI and ML	Object detection, preprocessing, confidence filtering	YOLOv8, OpenCV, NumPy
Hardware	Camera capture, display output, enclosure	Raspberry Pi 5, Pi Camera 3

Application Flow

The camera continuously captures frames
Every third frame is sent to the YOLOv8 model after preprocessing
If a recyclable item is detected with confidence of at least 0.75, a short timer begins
If the detection remains stable, the name-entry screen appears
The student types a name and submits it
Points are awarded and saved in SQLite
The system resets to the scanning screen

Hardware Bill of Materials

Component	Purpose	Cost
Raspberry Pi 5	Main computer running the kiosk	~$60
Pi Camera Module 3	Captures images for AI classification	~$25
ELECROW 5 inch touchscreen	Displays interface and accepts input	~$40
3D-printed enclosure	Holds the hardware in a portable shell	~$15
Total		Under $150

Software Dependencies

Package	Version	Purpose
Python	3.11	Core runtime
PyQt5	5.15+	Touchscreen GUI
OpenCV	4.8+	Camera input and image processing
Ultralytics	8.0+	YOLOv8 model inference
NumPy	1.24+	Image array operations
SQLite3	Built in	Local points database

AI Model

The AI model is a fine-tuned YOLOv8n variant trained on a custom dataset of recyclable plastics. It detects three classes:

Resin Code	Name	Examples
PET (1)	Polyethylene Terephthalate	Water bottles, soda bottles
HDPE (2)	High-Density Polyethylene	Milk jugs, detergent bottles
PP (5)	Polypropylene	Yogurt containers, bottle caps

The model uses INT8 quantization for better performance on the Raspberry Pi 5. Input images are resized to 320 x 320 and normalized before inference to handle inconsistent lighting conditions.

Technical Challenges and Solutions

Inference Speed

Optimization	Before	After	Improvement
Input Resolution	640 x 640	320 x 320	4x fewer pixels
Weight Format	FP32	INT8	Smaller and faster model
Frame Sampling	Every frame	Every 3rd frame	Lower compute load
Overall Inference	Over 2000 ms	About 400 ms	About 5x faster

Lighting Variability

The model performed well in controlled testing but struggled under cafeteria lighting. This was improved by augmenting training data, adjusting camera exposure, and adding brightness normalization.

False Positives

False detections were reduced by increasing the confidence threshold, requiring persistence, and limiting detection to the center of the frame.

User Experience

The interface was simplified to remove unnecessary steps. Detection leads directly to name entry and point assignment.

Conclusion

EcoKiosk shows that low-cost edge AI can be used effectively in a real school environment to promote recycling through incentives and interaction. The system reliably identifies common recyclable plastics after optimization for speed, lighting, and accuracy.

Texas disposed of more than 41.3 million tons of waste in 2024, and the average Texan generates 7.24 pounds of trash per day. Nationwide, only about 5 percent of plastic is recycled. EcoKiosk addresses this gap at the behavioral level by replacing indifference with incentive and directly supports SDG Target 12.5. [6][7][9][10]

Key lessons learned: