Making machines 'see' in 3D
Efficient machine vision is crucial for automatic inspection and industrial robot guidance. Europe boasts about 3 000 small and medium-sized enterprises (SMEs) working in machine vision. In turn, these SMEs provide services and products to 300 000 other SMEs in the machine building and automation sector. However, current machine vision technology is limited as it relies on 2D cameras that lack depth and contrast. The EU-funded project 'Efficient 3D completeness inspection' (3DCOMPLETE) worked on developing better 3D inspection methods by exploiting two different technologies. The first is 'full 3D' orthoimage technology that calculates arbitrary views of an object with a small number of images of this object. The second is extended 2.5D scanning technology combining 3D shape data with colour and texture information. To achieve its aims, the project conducted tests in the automotive industry and electronics manufacturing to define quality control criteria that would help develop such technology. Using multiple high-precision cameras, virtual sensors and advanced software algorithms, the project team integrated both technologies involved in two demonstrators. It designed a high-cost system with many cameras and a low-cost option with much fewer cameras, aimed at different business needs and budgets. Overall, the project team designed a 'full 3D' prototype that uses oblique images to generate arbitrary views of an industrial process in order to gain precise inspection and improve depth reconstruction. It also produced an 'extended 2.5D' camera prototype that combines profile scanners with colour cameras to generate 'coloured point clouds' to facilitate quality assurance and industrial monitoring. The 3DCOMPLETE project overcame many challenges in machine vision and is expected to have a strong impact on end-of-line inspection systems in assembly lines. The technology could potentially transform conventional inspection systems in assembly lines, providing better data and realising cost savings. Project results were unveiled at the CONTROL fair in Stuttgart, Germany, and published in numerous scientific journals. Plans are underway to fine-tune and exploit the technology, establishing a new paradigm in machine vision. Industrial processes and assembly lines in several sectors stand to benefit radically from this endeavour.
