Years of research and collaborations with the industry field have enabled us to accumulate extensive theoretical and practical knowledge in different areas of computer vision. Practically, given a specific practical problem, we devise techniques and strategies to solve it in an optimum way by:
- Choosing the most adequate hardware setup in terms of costs and efficiency
- Developing and adapting the suitable set of algorithms and techniques
- Testing, installing and training the technical staff in order to efficiently operate the new system.
The camera represents the workhorse of the field of computer vision, for this, it is critical to choose the optimum one(s) for the given task. In other words, given the objectives and the constrains, camera parameters such as resolution, sensitivity, and optics parameters such as field-of-view, speed, distortions, etc. have to be chosen carefully.
Another aspect that we always take into account is the cost resulting from acquiring the hardware. While linear or high-speed cameras are designed specifically for industrial applications, depending on the application, consumer cameras could provide a much less costly alternative, while offering the same results in many cases.
When designing a vision solution using adequate illumination systems is equally important as choosing the right camera. Here, our experience allows us to choose the optimum illumination setup. For example, choosing the correct wavelength, which influences the light colour, can help us emphasise specific parts of the product, which are important for analysis during the manufacturing process. Moreover, using ultra violet / infrared illumination can help us analyse parts of the structure of the products that are otherwise invisible for the human sight.
On the other hand, using systems where light is cast at a specific angle or lasers can help us detect certain aspects in the shape/structure of the products, enabling reliable quality control or product classification.
Image acquisition processing, be it for industrial, medical, etc. applications are prone to variations induced by illumination changes, changes in the position of the object with respect to the camera, noise, etc. This is even more exaggerated in general-purpose applications, where, the illumination, camera properties cannot be controlled. For these reasons, the development of an image processing strategy that compensates for such changes is crucial for a robust computer vision system.
Also, correctly choosing the right approach for image characterization can further compensate for such changes. Moreover, the characterization techniques have to capture the ideal image properties for the intended application:
- Class particularities, differences between classes for object classification/recognitions;
- The most representative properties for image-based measurements;
- Appropriate image regions for mapping and mosaicing, etc.
The aim here is to develop and implement the appropriate machine learning techniques to solve your specific problem. This is done in two essential steps:
- Designing candidate classification/recognition techniques, taking into account the specific problem, and the type of visual data.
- Devising the testing/evaluation strategy that will allow us to choose the best classification/recognition techniques or set of techniques for your specific problem.