A smart imaging camera can perform actions based on the content of video sequences. A good example is a camera that can segment a video sequence into objects, track some of them, and raise an alarm if some of them show an unusual behaviour. Until now all research effort on smart imaging has been carried out on the algorithms, without any hardware requirements. This results in large and complex systems consisting of cameras and separate PCs. This is not suited for mobile applications. The objective of Camellia is to develop a smart imaging hardware block (core) that can be embedded in a camera. The starting point of this core consists of a video compression core, which will be extended with additional hardware to support smart imaging functionality. This core will be low-cost and low power, and offer sufficient flexibility to address a wide range of mobile smart imaging consumer applications. The feasibility of this core will be demonstrated by two automotive and two surveillance applications.
Most smart imaging applications require compressed video as output. Therefore, the Camellia project will merge smart imaging functionality with a video compression core for mobile applications. The work is centred on the mapping of the smart imaging algorithms onto this video compression core. By reusing as much as possible from this core, a low-cost and low-power solution will be achieved. It is expected that the programmable parts (microprocessor) and some of the hardware coprocessors (the motion estimator) can be reused. Only a limited amount of additional, and smart imaging specific, hardware will be required. This additional hardware should be as generic as possible in order to address a wide range of smart imaging applications. On the other hand, the currently available smart imaging algorithms will have to be modified in order to reuse the hardware parts of the video compression core. Furthermore, it may be possible that some of the computations of these algorithms cannot be implemented in a camera. In those cases alternative solutions will begenerated.
After 30 months Camellia will have resulted in a prototyping system that demonstrates the four smart imaging applications with real-time behaviour. This system will consist of a software board for the programmable part, and FPGA boards to realise the hardware coprocessors. The prototyping system will be very close to an actual chip implementation. Other deliverables are: template architecture of the smart imaging core; mapping results of algorithms onto the core; coprocessors for smart imaging.
Funding SchemeCSC - Cost-sharing contracts
75272 Paris 6
92100 Boulogne Billancourt
35001 Las Palmas De Gran Canaria