Quality right down the line
In scenarios where moving, continuous material needs to be inspected for faults, line-scan cameras will generally provide a better solution than traditional area scan cameras. Despite this fact, people regularly harbour reservations about deploying line-scan cameras; reservations that generally arise from insufficient experience with this method of inspection. This often leads to a situation where image processing users often continue to deploy familiar area scan technology even when it would make more sense to use line- scan cameras for a given application. But the fact is that line-scan cameras offer an inexpensive method of generating high-resolution images and making them available to the software for evaluation on standard PC platforms whose performance is increasing all the time.
How line-scan cameras work
The traditional fields of application for line-scan cameras include situations where continuous materials need to be analysed (see figure 1).
The reason why line-scan cameras are more suitable than area scan cameras for such continuous processes is explained by the fundamentally different design of the two technologies.
Depending on the camera type, area scan cameras deliver a fixed (synchronous) or variable (asynchronous) sequence of images of a moving object. In practice, uninterrupted capture of continuous materials is achieved by capturing overlapping images. When this has been done, software is required to painstakingly crop the individual images, eliminate distortion and assemble the images in the correct sequence.
Line scan cameras, on the other hand, only have a single row of light- sensitive pixels, which constantly scan moving objects at a high line frequency. Typically, the resolution of line cameras varies between 512 and 12,888 pixels. Sensors with pixel edge lengths of 7µm, 10µm and 14µm are commonly available on the market. In the case of very high resolutions, smaller pixels are used in the design of the sensor in order to take the optics into consideration, since a sensor with 8000 pixels and an edge length of 10µm, for instance, could only be “exposed" without distortion by a lens with an image diameter of at least 8cm.
In operation, the charges of the individual pixels are read into a horizontal register arranged in parallel and converted pixel by pixel to digital values. These can then be stored and processed by the PC. To illustrate this point, if a line-scan camera were to observe a motionless object and was operated at a constant line rate of 1kHz, 1,000 lines of a PC monitor would be filled with lines of identical greyscale information in 1 second.
It's all a matter of the right settings
This example makes it clear that a 2D image of a flat object can only be generated with a line-scan camera if the object moves under the camera at a sensible speed. Of course, it is also possible to move the camera across the object, and this is indeed a possible for some applications.
The line rate of the camera must be synchronised to the speed of a moving object if the same resolution in the direction of travel (Y) is to be achieved, as the resolution across the object width. If this is not the case, and the line rate is fixed while the object speed varies, the object image on the monitor or in image memory is elongated or compressed. However, the speed of conveyors or the positioning equipment is often subject to load changes and acceleration or braking, which means that they rarely move at exactly the same speed. This in turn means that it is not generally possible to work with a fixed line rate.
The hardware must provide a method of adjusting the line rate to match the current speed of the material under inspection, as only then can a precise, meaningful 2D analysis of the image be made, using software algorithms.
In practice, this is usually implemented using an incremental encoder coupled to the drive unit. Of course, this feedback should be generated at a position where the minimum of slippage is expected with respect to the object. This issue needs to be considered on a case-by-case basis. The pulses generated by the encoder are then passed to the frame grabber and conditioned for the required resolution in the direction of travel using adjustable frequency scalers, before being passed to the line-scan camera in order to trigger the illumination.
This overcomes the first obstacle to maintaining a constant resolution at varying speeds. But, unfortunately, that's only half the story: The line rate is now “hardwired to the speed, but the monitor would display images with fluctuating brightness depending on the line speed. The reason for this lies in the varying exposure times of the line-scan camera at different line rates. In this “basic operating mode the exposure time is roughly the inverse of the line rate or the period between two trigger pulses.
The use of appropriate timing signals allows the integration time of the sensor to be kept constant. The setting is made on the frame grabber and is geared to the maximum expected line frequency. For instance, if the maximum line rate is 10,000Hz, the integration time must be set to a value shorter than 1/10,000s. This achieves an ideal situation where the image is exposed at a constant level and the resolution is maintained even if the line frequencies change. The control signal for the line-scan camera now contains the information about the current line rate and the exposure time and, in modern strategies, is a mixture of the pulse sequence and the pulse length.
Fast and sensitive
There are huge differences in the technical characteristics between different line-scan cameras. For instance, models are available which have been specially optimised for high speed and sensitivity operation. One such example is the Spyder3 line-scan camera from the Canadian manufacturer Teledyne DALSA, which is available in 1k, 2k and 4k (monochrome version) resp. 2k and 4k (color version) models with data rates of 68 MHz (monochrome) resp. 108 MHz (color). With a pixel size of 14 x 14µm² and a second line that can be activated, this model can also be used in low-illumination applications (see figure 2).
As far as sensitivity is concerned, the most effective strategy has is known by the name TDI (Time Delay and Integration) and has been perfected by Teledyne DALSA. This sophisticated sensor technology involves image data in one sensor line being accumulated with several other lines, synchronous with the movement of the object and then exposed again using the same image information. This method offers 100x more sensitivity than standard methods.
Today's sensor and A/D converter technology means that 8 to 12bit signals with line frequencies of up to 110 kHz are no longer an issue. Due to physical limits in workload the sensor is read synchronously over multiple readout channels. In high-performance line-scan cameras, it is possible to divide the sensor into a number of areas which can then be read in parallel at maximum speed. Thus, for instance, the sensor of Teledyne DALSA’s high-performance, 8142 pixel Piranha3 camera is read over 8 channels at 40MHz per channel. This method allows a line rate of up to 33.7kHz to be achieved! (see figure 3).
In applications where the use of a high-resolution camera is still not sufficient, two or more cameras can be installed in parallel, as several cameras with half the resolution can be read very much faster.
Manufacturers are constantly finding new, intelligent solutions to improve the precision of line-scan cameras still further. Thus Teledyne DALSA uses a specially developed procedure to ensure that the sensors in their cameras are ideally mounted in exactly the same position in the camera. In doing this, the manufacturer makes subsequent replacement or adjustment of the cameras a more attractive option. Not only that, but the maintenance costs are also reduced.
The wide variety of potential line-scan applications demands that it should be possible to configure them as flexibly as possible. Settings for data output determine the current state-of-the-art, where the user can generally set the data depth to 8, 10 or 12 bit, specify whether the data is output over 1 or 2 channels and set the data rate. Furthermore, it is possible to upload correction tables to the camera (for flat-field correction for instance). It is also possible to switch between free-running and trigger modes.
The techniques described so far in this article relate to monochrome line cameras, which make up the vast majority of line-scan cameras in use today. However, line camera models are available for colour applications and colour detection. Three different systems have become established in this area. Firstly, there are the high-precision prism cameras with 3 to 4 CCD or CMOS sensors. 4CCD and CMOS cameras are manufactured by JAI and offer advanced color inspection thanks to IR models. Then there are the tri-linear cameras, which almost achieve the same levels of precision. With this type of camera, three RGB sensor lines are arranged in extremely close proximity to capture the image of the moving object. The internal camera electronics then compensate for the resulting line-shift. And for low-cost applications, there are also the monoline colour line-scan cameras which make use of triple-encoded, RGB filter pixels.
It's the combination that counts
However, even the very best line-scan camera is not able to perform an inspection task in an industrial manufacturing environment efficiently and with a high level of quality all on its own. The optimum interaction between the individual system components - from the illumination, lenses and camera, right up to the image capture hardware - is a decisive factor in determining the quality of the overall image. Furthermore, the capabilities of the image processing software responsible for evaluating the image data is equally crucial.
Thanks to the increasing demand for application solutions, there is a wide selection of fundamentally different technologies and products that together, allow smooth interaction between each link in this chain. Although the combination of a suitable line-scan camera with suitable image capture hardware is no longer a major issue (thanks to common standards agreed by the manufacturers of cameras and frame grabbers some years back), intensive consulting services provided by experts are often still necessary in order to select the individual elements which will make up an overall solution that is ideally suited to meet the requirements of any given application.
We at STEMMER IMAGING offer our customers all the components needed to solve image processing challenges - from illumination, lenses, cameras and frame grabbers, right up to the software and of course, all the correct cables and accessory products needed to get it all working. STEMMER IMAGING 's many years of experience, the strength of its international suppliers and our collaboration with a number of experienced integrators means that we have all the tools on hand to create a system that is perfectly tuned to the end-customer’s requirements. Of course, this does not only apply to applications for line-scan cameras, but also to any other tasks where image processing is deployed.
Interesting applications for line-scan cameras
Traditional applications for line-scan cameras usually include scenarios where continuous materials need to be analysed. Industry provides a number of examples of this. Line-scan cameras are, for instance, used in the printing industry, in the manufacture and subsequent processing of paper, in the manufacture of steel plate, glass tape or of textiles in order to identify and classify faults in these materials. In the case of steel plate, image processing systems allow the steel plate to be used and priced according to its quality. Sections with no faults might for instance be used for the visible parts of the vehicle such as the wings, doors or the bonnet whereas sections with minor surface imperfections might be used for hidden parts of the vehicle such as the underbody or parts hidden by a trim.
Interesting applications for line-scan cameras can also be found in the foodstuffs industry, for instance to discard imperfect maize kernels or similar types of food, which cascade past one or more line-scan cameras like a curtain. The image processing system identifies imperfect kernels or foreign bodies and triggers compressed air jets which are installed below the cameras and automatically eject the unwanted material.
These examples only show a small selection of the technical possibilities and the applications for line-scan cameras which have already been implemented.
A practical example of how to determine the line rate and exposure time:
The following parameters are known: Object width: W = 370mm Object speed: v = 3m/s Required resolution: Dx = 0.2mm/pixel
The required number of pixels is thus
n pixels = W/Dx
This means that a resolution of at least 1850 pixels is required, and models with 2048 pixels are commonly available that are sufficient. If the object width of 370mm is now mapped to the 2k line sensor, one pixel will cover 0.18mm of the object. If the horizontal and vertical resolution are to be the same, (a ratio of 1:1), the time Tl for one line and as a result of the required line rate fl can be calculated as follows for nominal operation:
fl = v / Dy
The calculated resolution of Dx = Dy where Dx = 0.18mm thus results in a frequency:
fl = 16,667Hz
This means that the line-scan camera should be designed for a frequency of at least 16.7 kHz.
The SPYDER2 S2-1x-02k40 from Teledyne DALSA with its line rate of up to 18kHz would therefore be an ideal choice for this particular application and still has some performance in reserve: If the production frequency is increased, the object speed can be raised to a maximum of 3.24m/s.
Line-scan cameras offer an economic and extremely powerful technical solution to quality inspection, particularly when dealing with continuous materials.