Options
2007
Conference Paper
Titel
Swarm-based 3-D pose recognition in depth data for 3-D bin picking
Abstract
This paper explores a swarm-based approach to recognize 3-D-object poses in depth data for 3-D bin picking. This approach uses a set of features that are easy to calculate and robust to errors in the scan data. By combining these features, robust identifiers are obtained. Recognition is performed using Particle Swarm Optimization (PSO) in combination with histogram matching. The system is tested on different depth data with several complex objects. For each object a database of some thousands views is used to get an accuracy of 10° around each axis. The system achieves recognition accuracy above 93% on real scanned scenes with several objects that are typical for industrial robot handling applications.