Keywords
AI, mobile, computer-vision, quality-assurance

Technologies
React Native, PyTorch, Directus, YOLO, CNN

This project focuses primarily on running optical object detection and classification models on a Raspberry Pi Model 4 for use in quality assurance in manufacturing.

For object detection, the project explored several different model frameworks which could be used such as Ultralytics YOLO and ResNet within the PyTorch ecosystem. The model that was chosen in the end was the Ultralytics YOLOv8s. This model had a good balance between speed and accuracy, maintaining over 90% confidence on the required labels as well as being able to complete analysis in less than 3 seconds. The YOLOv8n and YOLO26 models were also examined but were unsatisfactory when compared to the YOLOv8s Model.

The classification model that was chosen is a custom-built Convolutional Neural Network developed within the PyTorch ecosystem. It was able to complete inference within 0.2 seconds while maintaining good accuracy.

This project also includes a mobile app written in React Native which can connect via BlueTooth to the device for real-time updates, As well as a server allowing the app to view data from the device and allows users to make changes to the device wirelessly from the mobile app.