ESA's cornerstone mission Gaia aims at autonomously building a billion-star catalogue by detecting them on
board. The scientific and technical requirements make this an engineering challenge. We have devised a prototype
to assess achievable performances and assist in sizing the on-board electronics. It is based on a sequence
of four tasks: calibrating the CCD data, estimating the sky background, identifying the objects and, finally,
characterising them. Although inspired by previous similar studies (APM, Sextractor), this approach has been
thoroughly revisited and finely adapted to Gaia.
A mixed implementation is proposed which deals with the important data flow and the hard real-time
constraints in hardware (FPGA) and entrusts more complex or variable processing to software. This segmentation
also corresponds to subdividing the previous operations in pixel-based and object-based domains. Our hardware
and software demonstrators show that the scientific specifications can be met, as regards completeness, precision
and robustness while, technically speaking, our pipeline, optimised for area and power consumption, allows for
selecting target components. Gaia's prime contractor, inspired by these developments, has also elected a mixed
architecture, so that our R&D has proven relevant for the forthcoming generation of satellites.