Pharma Blister Pack, Cap-Seal & Liquid Fill Inspection
Industrial inspection R&D for blister-pack cavities, cap presence and seating, and liquid fill levels using lightweight geometry and image-processing methods with a queued inference runtime.
Internal R&D / Personal Engineering Project at Bondstein Technologies Ltd.
Module of Bondstein Vision Suite; presented as a separate case study.
Workflow at a glance
Acquire line imagery
Read frames from an inspection camera.
Inspect product regions
Analyze blister, cap, or fill-level regions.
Apply task rules
Evaluate configured visual inspection cues.
Review candidates
Present candidate defects for assessment.
Project facts & reported results
Implementation fact
Implementation fact
Implementation fact
Project Overview
Problem Statement
Packaging inspection must balance defect sensitivity with predictable latency and maintainability on production hardware.
Approach & Methodology
Built lightweight, classical CV modules powered by a shared inference runtime. Implemented grid cavity brightness/fill signature analysis for blister packs, minAreaRect geometry for cap tilt/height, and Sobel row-energy line regression for bottle liquid levels.
Outcome & Scope
Created configurable inspection modules and a micro-batched runtime suitable for controlled line trials and further calibration.
System Architecture
Lock-free microservice inspection suite for industrial packaging lines
Key Components:
Threaded RTSP/webcam reader with auto-reconnect and per-camera FPS throttling
Lock-free micro-batched queue managing processing requests without global locks
Per-cavity brightness and fill signature validation against reference template
HSV mask + minAreaRect geometry for detecting cocked or elevated caps
Sobel row-energy gradient search to measure exact liquid fill level percentage
Key Features & Capabilities
Blister pack cavity brightness & fill signature comparison against auto-learned reference
Cap presence, tilt (cocked cap), and seated height (high cap) geometry analysis
Liquid fill line regression inside bottle ROI using Sobel row-energy
Lock-free micro-batched InferenceServer queue architecture
Config-driven operation requiring zero global model locks