Embedded Computer Vision Pipeline for Edge Device

Overview

Developed an end-to-end Computer Vision pipeline to detect light bulb activity on industrial machines and generate operational statistics (working vs idle time). The system enabled automated reporting and provided actionable insights to improve machine efficiency and business decision-making.

Achievements

Successfully delivered a production-ready pipeline that enabled reliable on-device inference. Reduced manual intervention by structuring the full lifecycle — from training to deployment — into a repeatable and scalable workflow.

Responsibilities

• Designed and implemented the full ML pipeline: data preprocessing, augmentation, model training, and validation
• Experimented with and evaluated multiple object detection models to achieve stable performance in real-world conditions
• Optimise models for edge deployment (latency, memory footprint, inference constraints)
• Built and deploy inference workflows on embedded camera devices
• Integrated monitoring and validation steps to ensure consistent performance after deployment