Can you check the size and the ripeness of fruit with vision systems? Using Common Vision Blox makes is possible. The Software Tools CVB Edge and Color are successfully solving this task.
In the system, two cameras are used in two different positions on the line. Angled mirrors are placed in the field of both cameras to image three sides of each fruit with both cameras. This means that 80% of the fruit can be seen by both cameras.
The first camera is an RGB colour model, that uses an external trigger to capture the image. This camera is used to determine the colour of each piece of fruit with a high degree of accuracy. The resolution of the camera allows 4 oranges to be captured on each image. This means that there are 12 regions of interest to inspect per image.
The second camera is the monochrome model that acquires images using another external trigger. This camera also acquires 4 oranges per image and this is used to calibrate the fruit.
This arrangement means that each camera inspects 4 pieces of fruit. The processing time will define the number of inspections per second.
The following sections describe the components required to solve the application. It also explains the reasons behind the selection of these specific components.
The JAI M77 camera had been chosen for a number of different reasons. On one hand, the application required a progressive scan camera because the target is a moving object and using a progressive scan camera removed the interlaced effect. On the other hand, the camera needed to be RGB 3 x chip colour model because a high quality colour image is required for fruit classification. The images are aqcuired in frame reset mode to reduce the region of interest, which reduced the processing time. The camera is fitted with a circular polarizer to reduce light reflections on the surface of the fruit. Using this camera it is possible to acquire a field of view of 350 mm, which gives a resolution of 190 pixels per millimeter.
The other inspection performed on the oranges was a calculation of their volume using a second camera. Unfortunately, the JAI M77 could not be used as this introduced unwanted colour information and, even if only one channel was used (ie red or green), the resulting measurement would not be accurate because of the high degree of colour variation between the oranges. For these reasons, a monochrome JAI A11 was used in combination with software binarisation to produce images suitable for accurate volumetric measurements to be made.
The frame grabber selected was the PC2-Vision from DALSA Coreco. This frame grabber has two multiplexed RGB video inputs and two separate trigger inputs. This allows the application to run using only one board with the two cameras. The first video input was used with the RGB colour camera, and only one channel of the second video input was used for the monochrome camera.Therefore the acquisition from the two cameras is achieved sequentially with an aquisition cycle of 80 ms.
The optical system was not crucial in this application, but was defined by the geometrical placement of the vision system. It is strongly recommended that a lens larger than 6 mm is used to avoid optical aberrations.
The illumination has to meet two specific requirements. It should be white in order to reduce any effect on the colour identification, and it had to be a high frequency light to avoid mains 'strobing' effects. High Frequency fluorescent lighting was chosen in combination with a diffuser as the field of view was over 500 mm per camera.
The software used in this application was Common Vision Blox. This software uses a process library that makes it very easy to develop vision applications. Two Common Vision Blox tools were used: CVB Color for the colour detection and CVB Edge for the fruit calibration.
In the conclusion we like to point out the processing time. If the instructions mentioned above are followed, a color processing time of 200 ms can be achieved with a calibration processing time of 30 ms. A total processing time of 230 ms is thereby achieved. This gives, 4 frames per second, 16 oranges per second.
One of the problems this application had to address was being able to cope with over sized oranges. It was recognised that the system might, under certain circumstances, recognize two oranges as a single fruit. This issue was solved using some filters, but this increases the processing time of the application.
At the present there are many people studying defects in fruit. This can be very complicated because there is often a large variety in the shape and colour of both the fruit and the defects. Another tool that could prove very useful in cases like this, is CVB Manto, which is a neural net based pattern matching tool, which has a proven track record in the recognition of organic forms.
This application story aims to give you an overview of this particular application, and how it was solved. All the technical data featured in this article have been taken from 'real-world' applications, and only enhanced with theoretical notes where neccessary.
STEMMER IMAGING has been leading the machine vision market since 1987. It is Europe's largest technology provider in this field. In 1997 STEMMER IMAGING presented Common Vision Blox (CVB), a powerful programming library for fast and reliable development and implementation of vision solutions, which has been deployed successfully throughout the world in more than 40,000 imaging applications in various industries.