FPGA-based smart camera : industrial applications
Résumé
For the last two decades and still today, smart cameras offer innovative solutions for industrial vision applications. This kind of system associates a flexible image acquisition with high-speed processing possibilities. Many smart camera designs are based on FPGA components to obtain these two features. Indeed, the FPGA enables the CMOS sensor to be controlled and therefore to propose a configurable acquisition according to the application constraints (i.e. dynamic windowing). The configurable structure of a FPGA represents a key advantage for modifying the embedded processing (even on-the-fly using dynamic reconfiguration). Meanwhile, FPGA components offer a large number of hardware resources, such as multipliers or embedded memory blocks, which enable complex image processing to be implemented and to be performed in real-time. Designers can even consider increasing the spatial image resolution and/or the frame-rate due to the FPGA technology improvements. The new solutions on the prototyping tools as well as the modelling languages available for FPGA design should be considered. Indeed, design methods based on High-Level Synthesis (HLS) enable the time to market to be significantly reduced. Moreover, these improvements enable gains on the smart camera design to be obtained, as for instance quick HW/SW implementations or quick communication interface integrations. After a general presentation of the smart camera structure, the Le2i laboratory's experience on smart camera designs will be used to highlight these gains.. The high processing capacities of a FPGA component at high frame rates, with high resolution images, will be demonstrated The presentation of the impact of co-processing on the smart camera performances, followed by a description of a new data-flow formalism, which enables quick prototyping of HW/SW implementations including communication interfaces to be automatically obtained, will be proposed. Finally a configurable system supporting automatic video compression adaptation in function of event detection will be presented.