Application of Model Chaining in Computer Vision for industrial environments with limited data diversity.

Details

Supervisors

Vande Kerckhove, Corentin ; Satinet, Chloé

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

In recent years, the integration of computer vision in industrial applications has significantly advanced, driven by the need for enhanced automation and efficiency. This thesis, completed in partnership with a Belgian company, aims to develop a computer vision solution for automating the detection and classification of truck loads in an industrial context characterized by limited data diversity. An initial approach was established to directly detect the transported materials. However, this approach suffered from the lack of diversity in the available data, as the training images were predominantly from a single 360-degree camera. To address this issue, the thesis brings a model chaining approach, combining a truck bed detection model with a material classification model. This method leverages the strengths of separate models for detection and classification, each trained on different data to enhance system flexibility and performance.