Objekt detection for robot manipulation

Junckers Industries, who manufactures wooden floors, wanted to automate a physically hard job. The job was to depalletize woodbars and put them into a press-machine. This project concerns development and test of a detection system for theese woodbars making it possible for an industrial robot to do the job in a reliable way.

The detectionsystem is based on computer vision algorithms and pictures of the woodbars taken from above the stack. There were developed two different algorithms for detection of different error-types in the stack. The algorithms were implemented and tested with real pictures from a stack at Junckers.

The projects consists of a large amount of advanced computer vision theory and methods. The implementation were done with help from the open source computer vision library OpenCV. During the development of the algorithms a new and improved version of the wellknown Hough-algorithm were also developed.

Even though the the detection-algorithms almost
worked without problems (99 %) on the given test-images, they were not reliable enough for real operation on site. In this case Computer Vision is too sensitive against the textures of the woodbars.
Therefor Junckers were advised to use other detection technologies, such as laser triangulation.
This project served as a pre-project before the final automation of the job at Junckers.

The report for the project can be found here (in danish):
Objekt detektering for robot manipulation.pdf

For more information about this project, or a quote for a similar project, please fell free to contact Allan Krogh Jensen.