Image processing is the state-of-the-art technology for quality assurance in industrial production. In fact, "Machine Vision" developed from this field in the last couple of decades. The high standards of quality which dominate the automotive industry, e.g., ensured the development from a high-end technology to standard products which are used in various ranges of application today. One field only recently developed, is the sorting of agricultural products.

The enormous cost pressure in the automotive industry lead to the fact that out of engineering solutions products were generated that are small, compact and inexpensive.
The so-called intelligent cameras or the lower level vision sensors are a result of this development. These systems are in fact complete image processing systems which did need the space of a control cabinet less than five years ago. Today the electronics and the software, featuring the same or even a better perfor-mance, are enclosed in a camera body measuring only 50 x 50 x 65 mm. The integration of the complete processing unit, including the sensor and the lighting in one body permits the operation of such systems also in harsh environmental conditions like in the food industry.

Powerful Hardware - Compact Size
In surroundings where food is produced it is normal procedure to frequently spring-clean all machines. Therefore the vision systems employed here have to be resistant to water, cleaning agents and high temperatures as well. Due to the compact integration of all components within a very small housing it is possible to install the complete unit in a hermetically sealed cover body. The modern processors installed in these systems have a very small power loss which can be discharged via the equally very small surfaces of these enclosures. Thus the vision sensors are not only applicable everywhere without a control cabinet but they are also extremely robust towards environmental influences.
Despite the compact design all components found in a PC based system are also available with the vision sensor. The systems support a real time operation system with an Ethernet stack for the communication with the outside world, digital I/Os and a serial interface. For image acquisition the same high-quality sensors (e.g. from Sony) are used that are employed by "non-smart" cameras. The processing power of the vision sensor ranges up to 8,000 MIPS which equates a Pentium CPU with a frequency of 2 GHz.

From Smart Camera to Vision Sensor
There are no hardware limitations for porting the powerful software packages which have been developed on the PC-based architectures to the compact systems. Thus powerful and user-friendly vision software is available. In the comfortable way of programming these systems do not fall short to the discrete build systems. The image analysis software can be generated simply by drag and drop so that the users themselves can parameterize and/or program these systems and by thus integrate them into their machines.
With the goal to even more simplify the usage of the vision sensors, EVT modified the graphical user interface which had been originally developed for smart cameras. The function range has been systematically reduced to specific applications and the user interface has been customized. In this way the EyeSpector smart camera family turned into the EyeSensor family. The systematic simplification of the software turns a smart camera into a vision sensor. The hardware stays the same, only the range of available functions and the operation are simplified until the vision sensor can be applied as easily as a light barrier.

Olive Sorting in the Twinkling of an Eye
One example for a highly specific sensor from the EyeSensor product family is the sensor for the sorting of olives. This sensor is based on a smart camera with a color image sensor, the image analysis is realized with the color vision tools from the EyeSpector software. A special user interface has been developed so that the user does not need to deal with any image processing specifics. The user only has to teach-in the color of the olives and - optionally in an advanced extension - the spots that can occur on the olive surface. The teach-in procedure is especially adapted to the olive sorting machines.
The olives fall on a sorting tray with n x m rows and columns. At each intersection there is a recess for one olive respectively. As soon as the tray is filled with olives the camera acquires an image of the complete tray. All fields including olives are evaluated and the olives will be either sorted out or passed on for further processing depending on the match between the inspected olive and the training parameters.
The user specifies the number of rows and the columns of the sorting tray. Afterwards the user has to position the first inspection field on the tray and based on this the left and the bottom inspection field. All other inspection fields are automatically set in equidistance based on this user input. With these steps the sensor is installed and the inspection can be started. Next the color characteristic has to be taught-in. To achieve this the user puts an olive in the desired color in the left upper corner of the tray. This color is then adopted by the software as the target for sorting. Thus all fields are automatically parameterized and can now be inspected for the required olive color. This already describes the whole teach-in procedure the user of the EyeSensor vision sensor has to execute for the color control. Now the system is ready for operation. A change-over to another color is equally fast.

First and Foremost: the Sensor Concept
In addition to sorting for colors the user has also the option to check the olive surface for dark spots. This feature is as well realized on the basis of a sample olive placed in the inspection field. In this case the user has the possibility to set a percentage of the surface which is covered. The result of the sample olive is also a percentage which is automatically determined by the software. The result is displayed so that the user has the possibility to intervene. Again the goal for the user interaction has been to develop a system with the sensor concept as a guideline.
The most complex step of adjusting this sensor is to activate the section flap gate situated under the tray. However, this complete unit is integrated by the machine builder. Therefore an easy protocol based on the serial interface of the sensor could be implemented. A string of zeros and ones is transferred for each row. The gate is opened by the controller at the position one. The user of the machine does not get to see this communication because it is part of the complete unit. The advantage of this data transfer is that it constitutes a simple and robust system and can be checked by a service technician anytime. Again the sensor concept was followed through. In addition it was taken into consideration that the machine is used in areas where know how of complex technical systems is not necessarily a given.

High Comfort, Low Cost
The EyeSensor for the sorting of olives is one example for the new product category vision sensor which constantly opens up new fields where so far discrete image processing systems have been used. One reason for this success is the fact that the processing power can be compared with those of a PC-based vision system but the cost is significantly lower.
The software product is established on the hardware platform of the vision sensor, the know-how to implement successful applications is available and the price advantage will most certainly add to the success of EyeSensors.
Especially the sensor for the sorting of olives is a great example for this kind of machine vision: in the Turkish olive harvests where the vision sensors are used, it would not even be possible to install a PC-based vision system. The then required control cabinet with the climate control unit alone would have a higher price than the vision sensor.