Automatic Annotation of Training Data using Visual Markers for Object Detection in Automated Factories

Keita Tomochika (1651077)


To accurately recognize various products in factories, using a deep learning- based vision systems are widely employed. Such systems require a significant amount of training data with annotations. Unfortunately, annotation is very time-consuming.

In this paper, we propose an automatic annotation method for fast training of the vision system. The method uses visual markers to obtain the annotation, such as types and pose of objects. This automatic annotation reduces the total amount of time needed to train the vision system. The experiments verified the effectiveness of the proposed method by comparison with manual annotation in both the time to generate training data and the accuracy of the vision system.