The thesis presents a compact, low-cost, vision-based inspection and rejection cell for mixed SKU household chemical bottling lines operating at realistic conveyor speeds. A single fixed-view USB camera with controlled ring light illumination and a matte background feeds a modular perception pipeline that performs SKU aware localisation, cap height profiling, label and color verification, expiry date OCR and silhouette based deformation analysis. A lightweight YOLOv8 detector provides robust bottle level region proposals, while physics based geometric measurements and rule based checks are used for the individual quality tasks instead of an end to end black box network. Per bottle decisions are converted into a binary reject signal for a PLC controlled electropneumatic pusher, with timing budgets derived analytically from conveyor flight distance, image processing latency and actuator dynamics to guarantee correct removal of defective units. Experiments on representative bottle types show perfect cap sealing classification and competitive performance on the other tasks, including 97.0% accuracy for deformation detection, 90.9% for label conformity and 87.9% for expiry code recognition using only commodity imaging hardware and open source software. The work demonstrates a practically validated, retrofittable AOI solution that combines modern deep learning with interpretable geometric features and structured data logging, and is therefore well suited to Quality 4.0 upgrades in small and medium sized bottling plants.