Improving Industrial Processes

Industrial image processing is one of the most important drivers of manufacturing automation today. The requirements on the cameras differ considerably depending on the application. Different measurement methods (2D, 2.5D, 3D), spatial and temporal resolution and scan rates can be employed. The resolution and dynamic range of the sensor are critical for optical inspection on manufacturing lines, while a scan rate of 50 Hz is generally sufficient.

Other applications require very high scan rates. If image information is output as raw data, very fast interfaces between the sensor and camera are generally needed, and considerable technical resources are required for transmitting, storing and processing the data. In some cases, however, the image data itself is not of primary interest. When image features are desired, such as the position of points (blobs) or the path of lines, it is useful to extract this information right at the chip and only transmit the relevant data. This is referred to as “compressed sensing.” One example is laser triangulation, which is used for 3D scanning of objects. Reducing the data volume at a very early point enables scan rates of 10 kHz and higher at full sensor resolution (2 MPix) with relatively simple camera hardware and only a few digital sensor interfaces (such as 3 LVDS pairs).

Still other applications depend on achieving high scan rates as well as tracking the latency of derived values from image capture to output. Examples here include fast characterization of objects for sorting by color and shape, such as with targeted air jets (pneumatic classification), or analyzing the size and shape of digitized blobs (e.g. molten pool analysis for laser machining). Decisions must be made rapidly based on the detected features to control the activation of air nozzles in the first case or adapt the laser power, focus or scan rate in the second one.
An example of a possible processing cycle with different steps of varying speed is shown in the figure.

Fig. 01: An example of a possible processing cycle with different steps of varying speed. Source: Fraunhofer IIS, EAS.

An image is first captured at very high speed, and a trigger condition is evaluated, such as the presence of analyzable information in the captured image. The relevant decision is made in the second step. In the third step, control values for a mechanical actuator (e.g. activation and deactivation times) are calculated with a very low delay. The actual driving of the actuator takes place in the fourth step, and real image information can be output in the final step of the cycle for a success check, if necessary. Rather than continuously, however, this only takes place at appropriate times, such as when a certain event is expected in the sensor image or if defects have been detected in the data processing.

The steps of this control loop can be realized on a software-programmed vision system-on-a-chip. This solution enables scanning rates in the double-digit kilohertz range, depending on the available light and required exposure time.