Last year, Kaunas University of Technology (KTU) researchers in cooperation with “UAB Freda” implemented a project based on the task presented by the company. At the invitation of the KTU Artificial Intelligence Centre, the team was developing a system for the detection of visual defects in the surfaces of furniture parts.
The researchers aimed to create an intelligent system that would identify the defects of furniture’s panels’ surfaces (scratches, deformations, angular breaks, or colour discrepancies) moving on the conveyor from visual information in real-time.
The solving problem had to include methods to perform automatic segmentation of individual parts, as the conveyor can transport up to three parts in one row at a time, which are not aligned, can be tilted (without leaving any gaps) and/or in several lines; as also classification of the defective and suitable part, where the defective part is assigned one of the predefined possible defect classes.
According to the scientists who carried out the project, in the system created, the image of furniture panels is obtained from two 6Mpx parallel linear cameras, which are analysed by distinguishing individual objects. Large surface objects are analysed by dividing them into small analysis windows. The quality of the surface of the object is evaluated using computer vision methods, in this case, a convolutional neural network.
The task presented by UAB Freda was implemented by the team of KTU Faculty of Electrical and Electronics Engineering: associate professor Arūnas Lipnickas, student Lukas Zabuilis, and former doctoral student of the faculty Rytis Augustauskas.
The value of the project “Development of a system for the detection of visual defects in the surfaces of furniture parts” (project no. PP2022/67) was EUR 50,000.
