Title: Improving Generalization of Computer Vision Models in Warehouses: A Study on Anomaly Detection in Vertical Material Handling

Abstract: Traditionally, the implementation of computer vision models within warehouse environments demands substantial resources. The workflow typically involves camera installation, data acquisition, annotation, model training, and deployment—a cycle that often must be repeated for every new site due to varying environmental conditions and mounting limitations. This study introduces a novel strategy to optimize this workflow by executing the entire standard procedure exclusively in a controlled laboratory setting. Specifically, the research targets vertical material handling systems, focusing on detecting anomalies in their forks.

Our extensive experiments demonstrate that generalizing from lab conditions to diverse warehouse settings is achievable through a combination of four key factors: strategic camera placement, intelligent image triggering, prudent model selection, and model ensemble techniques. This approach has the potential to revolutionize warehouse automation by drastically simplifying deployment. Instead of the labor-intensive processes of annotation and retraining, facilities would only need to handle camera mounting, image collection, and final model deployment. This shift promises to conserve significant time and resources. It is important to note that this work represents an experimental research investigation rather than a live production deployment.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC