Toshikazu Onda

A handling system for randomly placed casting parts using plane fitting technique

We have developed a vision-based robotic handling system, especially for casting parts. To locate an object, we have focused on identifying its reference plane. The system uses a two-dimensional matching technique to recognize an object and stereo ranging to measure its heights, and then applies a plane fitting technique to decide the three-dimensional orientations. The system can recognize a randomly placed casting part resting on a supported plane such as a conveyer belt even when it is tilted or piled on other parts. This paper describes the system and presents some experimental results.

Three-dimensional circle detection and location of pipe joints for bin-picking tasks

This paper describes a practical application of visual sensing for detecting and locating pipe joint parts jumbled in a bin. We have focused to detecting three-dimensional circles because each section of the part is visible as a circle in the scene. In addition, we have tried to search another section of the part using stereo ranging to determine the rotation of the part. We first introduce the characteristics of the pipe joint parts and our approach, including three-dimensional circle detection and location of the pipe joint parts. We then describe some experiments in order to evaluate the accuracy of our circle detection and parts location procedures. Finally, we present a discussion of our approach, including implementation in a handling system and application to other sensing systems.

Virtual property lines drawing on the monitor for observation of unmanned dam construction site

We have developed ARLINER an observation system to register virtual property lines on a TV monitor at an unmanned dam construction site. The system tracks markers set at regular intervals on the baseline of the site in the input image, calculates the camera parameters from the positions of the detected markers, and registers the property lines in association with the camera motion. We describe the conditions of application at the construction, marker tracking, the registration algorithm for property lines, and practical operation using the system at a dam construction site.

Vision-based handling system using model-based vision and stereo ranging

We have many applications of a vision-based robotic handling system for casting parts in our works. We have focused on a three-dimensional model-based matching technique to decide the position and orientation of an object, because most of the target objects are rigid and have some visible patterns or holes as line segments and closed-curves on their surfaces. By combining model-based vision and stereo ranging, the system can recognize randomly stacked casting parts in a bin with no special lighting facilities. The paper describes the recognition methodology, system configuration, and successful experiments to recognize and pick up an actual object in our works.