Machine vision projects at VTT Automation, Machine Automation. Paper roll position measurement and preprocessing for label OCR Paper roll position and shape is required in automatic loading at sea ports and paper mill storage halls. They are measured by using a line projector laser, together with machine vision as shown in the figure. Another task in the roll handling is automatic optical character recognition (OCR) of the text found in the label. At VTT, software was built for performing the preprocessing of the camera image required by OCR. Typically, such preprocessing includes straightening of the curved surface of the cylinder, shading correction, improved segmentation and finding of the different regions of the labels. After this preprocessing, most commercial OCR packages are able to do the actual character recognition. Contact persons: Jani Uusitalo and Juha Korpinen.
FIGURE: a video image of a paper roll. The stripe made by a laser line projector located above the camera is clearly visible and allows efficient determination of the roll position with respect to a truck.
The camera and laser placement inside the paper roll clamp. Machine vision for small parts assembly In a joint project with VTT and Tampere University of Technology, machine vision was developed for use of small parts assembly. Some of the results are listed here, e.g. a technique was presented on how to improve the accuracy of vision systems when large areas need to be analyzed with optics using small focal lengths. A new calibration method was tested and its suitability to assembly cells improved, so that sub-pixel accuracy can be maintained over large areas. The figure shows a small target that can be placed in the assembly cell. It allows calibration from a single image: the software calculates both lens parameters (focal length, radial and tangential distortion) and camera position and orientation with respect to the cell base. Other results included a 3D vision system capable of measuring bright, mirror-like surfaces, and OCR preprocessing for correction of the distortion of the image geometry. Contact persons: Juha Korpinen, Matti Kutila, Jani Uusitalo, Jouko Viitanen (VTT), Reijo Tuokko, Matti Saarinen, Longjuan Niu (TUT).
FIGURE: A photograph of the target suitable for on-line calibration of cameras in an assembly cell Automatic measurement of long jump results Track and field athletics games have tight schedules. Therefore the results from e.g. long jump competitions are required very fast. At VTT, a system was developed using computer vision that measures the length in two seconds. It has facilities for either fully automatic operation with flash lighting or, where flash is not allowed, the operator can determine the length by using a cursor on the perspective corrected image on a PC screen. Measurement is done using cameras on the side of the sand pit, therefore the view of e.g. TV viewers is not obstructed. Automatic field calibration of the system can be performed, therefore the placement of the cameras is quite flexible on different stadiums. The inaccuracy of the system is typically +-2mm, depending on the camera placement, and it can cover an area of 2.7m by 3.6m. The system is commercialised by a company Jat-Tuote Ky. Contact person: Jouko Viitanen. Fast distance ranging with line scan camera and laser Autonomous mobile robots and vehicles typically use time-of-flight (TOF) laser ranging sensors for obstacle detection. However, in fast scanning, their inaccuracy is too high (few centimeters) for operations requiring high precision docking or target handling. At VTT, a different type of ranging camera was developed, where the short distance inaccuracy is only a few millimeters. Also, its construction is inherently lower cost than that of TOF sensors, because a standard 1mW visible CW laser can be used, as well as a 4$ line scan camera element. The principle of measurement is based on fusion of two methods, triangulation and depth from defocus. The speed of measurement is 1,5ms/point with the A/D converters on the card, or <0,5ms/point with faster external A/D. Contact person: Jouko Viitanen.
FIGURE: The laser range finder
and its embedded DSP processor card with serial connector to a host computer.
Micromanipulator on-line control with computer vision The position of a micromanipulator
tip is measured using a high speed DSP processor card . The coordinate
update rate is about 15 Hz. The position signal is used as a feedback for
the controlling system. The accuracy of the system is +/- 1 pixel in the
X-Y - plane. Relative z-coordinate
FIGURE: The image of the tip of the micromanipulator. Its diameter is about 10 micrometers.
FIGURE: Measured micromanipulator trajectory in two different planes in 3D; x-y plane units are micrometers, z in relative arbitrary units. Contact information VTT Automation
|
