99.4%
Detection Accuracy
AI & GreenTechReverse Vending Machine AI
RVM AI Object Detection & Sorting Engine
Developed a lightweight computer vision and deep learning object detection pipeline for automated Reverse Vending Machines (RVMs). Running custom YOLOv8 and TensorFlow Lite models on edge hardware, the engine analyzes camera feeds to instantly classify PET plastic bottles, aluminum cans, and glass containers, cross-referencing barcode scans and weight sensor inputs to prevent deposit fraud.
PythonYOLOv8OpenCVTensorFlow LiteFastAPIEdge IoTDockerMQTT
Measurable Outcomes & Key Metrics
< 65 ms
Inference Speed
1.2M+
Monthly Items Scanned
99.8%
Fraud Prevention
Key Capabilities & Features
System Architecture & Core Functionality
Sub-100ms camera stream item classification (PET Plastic, Aluminum, Glass, Non-recyclable)
Multi-sensor verification combining optical AI vision, weight sensors, and barcode scanning
Anti-fraud engine preventing duplicate item insertion, foreign objects, and barcode spoofing
Over-the-air (OTA) model deployment pipeline for machine fleet neural network updates
Edge telemetry stream transmitting container metrics over MQTT to cloud analytics
Engineering Impact
What We Achieved & Delivered
- Trained custom YOLO classification model achieving 99.4% precision on container streams under varying light.
- Optimized model inference to execute in under 65ms on ARM-based embedded IoT hardware inside RVM units.
- Reduced deposit fraud and invalid item acceptances by 99.8% across operational machine deployments.
- Shipped OTA update mechanism enabling zero-downtime model enhancements across 500+ RVM units.
Next Case Study
RVM EcoRewards & User Loyalty Platform
Gamified mobile rewards application and cloud ecosystem connected with RVM machines, allowing users to claim green points, track carbon offsets, and redeem retail partner vouchers.