Abstract
Machine vision is playing an increasingly important role in industrial applications, and the automated design of image recognition systems has been a subject of intense research. This study has proposed a system for automatically designing the field-of-view (FOV) of a camera, the illumination strength and the parameters in a recognition algorithm. We formulated the design problem as an optimisation problem and used an experiment based on a hierarchical algorithm to solve it. The evaluation experiments using translucent plastics objects showed that the use of the proposed system resulted in an effective solution with a wide FOV, recognition of all objects and 0.32 mm and 0.4° maximal positional and angular errors when all the RGB (red, green and blue) for illumination and R channel image for recognition were used. Though all the RGB illumination and grey scale images also provided recognition of all the objects, only a narrow FOV was selected. Moreover, full recognition was not achieved by using only G illumination and a grey-scale image. The results showed that the proposed method can automatically design the FOV, illumination and parameters in the recognition algorithm and that tuning all the RGB illumination is desirable even when single-channel or grey-scale images are used for recognition.
Highlights
Machine vision technologies have been widely applied in the industrial field for automated visual inspection, process control, parts identification, and robotic guidance [1,2]
A vision system is usually composed of a camera and an illumination and recognition algorithm [3], which are known as the main design factors of a vision system
This set up resulted in an Fmeasure four shoots and contour models generated by a contrast value of 4. This set up resulted in an Fmeasure value of 1, a maximum positional error of about 0.6 mm, and a 3.1° maximum angular error
Summary
Machine vision technologies have been widely applied in the industrial field for automated visual inspection, process control, parts identification, and robotic guidance [1,2]. In the object recognition system of a pick-and-place robot, for example, the camera position needs to be set to obtain a suitable Field-of-View (hereinafter referred to as FOV), the illumination requires to be changed to strengthen features in targets, and the image recognition process needs to be optimised through parameter tuning. As this creates a number of conflicting variables, the design process must be reiterated until acceptable results are obtained. This is a time-consuming task even when carried out by experts, and even a simple pick-and-place vision system usually takes several days to design
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
