Steinbeis Junior Employee Project aimed at developing competitive multicamera systems
Optical measurement systems used in industry are subject to a variety of requirements. As the image resolutions of industrial cameras improve, the size of chips is increasing, and this in turn is raising people’s expectations in terms of the quality of optical output. Another effect is that enhanced optical quality has an immediate impact on technical requirements and this has an influence on prices. As part of a PhD project at Ulm University in Ilmenau, Guido Straube decided to try to develop a system using several low-cost cameras as an alternative to established camera systems. The obvious way to do this and develop his ideas was to work with the experts at the quality assurance and image processing specialist SQB, which is based at the university campus.
Straube set the bar high. The multicamera system he wanted to develop for his project should offer image quality on a par with a single-chip camera system, but it should be cheaper to make and offer significant commercial benefits over established single-unit systems.
To ensure the image resolutions of his new system would be competitive compared to large single-chip cameras, the budding engineer opted to integrate any number of cameras into his system (multicamera solution). The pictures from different cameras would then have to be merged and processed as a single image. What is important with such an approach is how the edges of images from individual cameras overlap – they have to be pieced together. The resolutions of the individual cameras have to be added up, but it still has to be possible to use software to process image overlaps.
Because the individual cameras had to be inexpensive, certain corners had to be cut when it comes to image quality. This is inevitable because cameras in the low cost segment often have fixed-lens systems. To nonetheless match the high resolutions offered by individual cameras, Straube decided to enhance camera output using image processing. The first step when setting up a camera array is to select the right cameras themselves. The priority at this point is the image resolution offered by individual cameras and the interfaces needed for data transfer and the power supply. “One important factor is that there’s no fixed way to pre-process images inside the camera itself. When you process the data, it’s good to have access to the raw data provided by the sensors,” explains Straube. He decided to use two megapixel board-level cameras for his project and these were given a 3D-printed plastic housing. After selecting the required cameras, the next step was to design a layout for controlling the unit and capturing images. “It’s important to find a way to connect several cameras to a computer at the same time so you can adjust settings such as image intensity (gain) and integration times, not just for individual cameras but also for all connected cameras working in unison,” highlights Prof. Dr.-Ing. Gerhard Lins, who as the managing director of SQB also supervised the PhD project. It was also important to the SQB to provide the young scientist with backing, also for financial reasons. The money for funding the research was provided by a European social fund through the Thuringen Aufbaubank, and the overall project was co-funded by SQB.
Not only was it important to work out the right settings for the cameras, one key requirement was that all of the shots taken by the cameras could be synchronized. To piece together individual images and not lose any information, distortions had to be resolved for each image supplied by the cameras. “When you get distorted pictures, the scale of images changes as the distance increases from the optical axis. We make a distinction between pincushion and barrel distortion,” explains Straube, to put it more graphically. Once the image data has been pieced together, Straube has his image. Ideally there will then be no difference between this image and those supplied by a high-resolution camera. Depending on the application and the corresponding resolution requirements, Straube would like to find a way to extend the camera array and reduce the number of required cameras. Processing and evaluating the large volume of image data involved is a challenge for any software. It takes technology to the boundary of big data.
One thing Straube’s sights are firmly fixed on now is the practical application of his research project. His aim is to develop an adaptable system that can be matched to actual usage scenarios and ensure that it is competitive, not just in terms of technological capabilities but also in terms of initial investment. This should make it possible to use his system to take high-definition images of extremely large objects. Possible applications identified by the researchers at the university in Ilmenau include checking car body parts for scratches and quality controls on major components at the end of a production line. The system could also conceivably be extended to include 3D images, and this would expand application options for the solution. To all intents and purposes, the experts seem to be already planning where to take the project next.