Speaker
Description
Maintaining consistent quality in pharmaceutical manufacturing is especially challenging in the manual production of gene therapy medicinal products, where operator interactions within isolator environments play an important role. Continuous monitoring can provide additional information about the manufacturing process and support its documentation, evaluation, and traceability. This work focuses on developing a computer vision system to support real-time monitoring and analysis of manufacturing processes in an isolator environment.
The core of the work is the development and training of deep learning models for detecting laboratory objects and DataMatrix codes. Dedicated image datasets were created for both tasks, containing images acquired under varying conditions representative of the isolator environment. YOLO-based models are used to detect selected laboratory objects and DataMatrix codes in camera images. DataMatrix code detection is combined with code reading to retrieve the encoded information. The three-dimensional position of detected laboratory objects can be determined using the camera system. The obtained 3D coordinates provide a basis for subsequent tracking of individual objects and monitoring their movement within the isolator.
Training and optimizing detection models require processing large image datasets and repeated evaluation of different model configurations. High-performance computing resources are therefore used to accelerate computationally demanding training and experimentation, allowing multiple model configurations and training parameters to be efficiently evaluated. The trained models can subsequently be integrated into the real-time monitoring system for continuous processing of image data acquired within the isolator environment.
By combining laboratory object detection, DataMatrix code detection and reading, 3D position estimation, and real-time image processing, the proposed system provides information about the objects present in the monitored area, their identification, and their spatial position. The obtained information can serve as a basis for subsequent object tracking and movement monitoring, contributing to automated documentation and traceability of the manufacturing process. Further development of the system will focus on computer vision-based monitoring of the process for gene therapy medicinal products, including the recognition and analysis of operator activities.