Unmanned Vending Machine System

The problem

Integrate computer vision, tracking, and decision logic into an unmanned vending service that can be maintained and updated.

My contribution

• Oversaw end-to-end service work, team progress, and scheduling.
• Applied YOLOv8, TensorRT, and MobileNetV3 to detection and classification.
• Implemented ByteTrack/DeepSORT tracking and frame-zone decision logic.
• Maintained Docker/Kubernetes deployment workflows and integrated ZeroMQ messaging with optional MongoDB transaction storage.

Technical approach

Object detection and classification provide inputs to tracking and frame-zone decision logic. Messaging connects system components, while container-based deployment supports updates and service maintenance.

1. Detection and classification

   Apply YOLOv8 and MobileNetV3, with TensorRT optimization as part of the inference work.

2. Tracking and decisions

   Use ByteTrack/DeepSORT with frame-zone logic to connect visual observations to service decisions.

3. Service integration

   Connect components through ZeroMQ; maintain Docker/Kubernetes deployment workflows and optional MongoDB transaction storage.

Engineering considerations

The engineering concerns span model speed, tracking consistency, and dependable service behavior. TensorRT optimization is part of the model work; messaging and deployment are separate operational responsibilities.

Outcome

Integrated AI/software work across inference, tracking, application logic, messaging, and service deployment, alongside team coordination.

Project resources

This page summarizes the project. A public demo or source repository is not available here.