Problem
Single-shot visual inspection fails under real conditions: dim light, glare and defocus hide defects, so bad parts escape.
Solution
ActiveInspect is an agentic OpenCV 5 inspector. It first measures image quality (exposure, specular glare, sharpness, noise), then acts to fix what it can't see: adjust exposure, switch polariser, refocus, average frames, or zoom. Only then does it decide PASS / REJECT, and it ESCALATEs to a human when it still can't see well enough.
AWS
Runs on AWS Lambda (arm64, Graviton), with images in Amazon S3, decisions and audit trail in Amazon DynamoDB, and optional Amazon Bedrock for operator-facing explanations. FastAPI front end.
Evaluation
On 630 parts: defect recall improved from 9.8% (single-shot) to 90.7%, escaped defects dropped from 379 to 8, with 0 false rejects.
Built With
- amazon-dynamodb
- amazon-web-services
- aws-lambda
- fastapi
- opencv
- python
Log in or sign up for Devpost to join the conversation.