Object detection of unsorted parts for industrial applications

Author: Nilai Sallent

Supervisors: Josep Ramon Casas

Presentation time: 10:30

Virtual Room: 4.3 | Live presentation URL 

Abstract:

Automated bin-picking is a core problem in robotics. The
project presents an alternative method for those cases where no rigid
3D shape of the object can be matched with the point cloud to extract
correlations for grasping. Thus, the method utilised is an object
detection model that employs the information of both grayscale images
and the scene’s depth to extract the matches. The results of these
experiments have been compared with the standard object detection on
RGB images for contrast. In summary, the information that adds the
scene’s depth improves the mAP metric by 1.5% while maintaining the
IoU precision of the bounding boxes.

Leave a Reply

Your email address will not be published.

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>