Description

In image-processing most tasks are very compute intensive. For example for image-pre-processing each middle point of multiple visible objects in a given image is to be found. A very innovative possibility is the Marching Pixels approach: an emergent solution, based on the use of cellular automata computation. Such cellular automata computation is a very compute intensive task, that can be efficiently parallelised. ParCA is therefore a system, that uses a single processor array, that works on a single partition of an input-image, representing a state of the basic cellular automata. So ParCA stands for partitioned cellular automata.

System Characteristics

Camera:	Global Memory:	Processor Array:
delivers binary images directly in global memory Controlled by the CPU CPU: Serves as Conroller for all other components Loads the PE Program DMA Controller for rapid Memoryaccess	Off-Chip Holds all necessary Informations incl. programs for the CPU and PEs Partitioning: Only a small part of the whole CA fits into the chip That is why we need to partition Partitionorder is programmable	Works on a single partition, while the other is being loaded Double-Buffering principle Fully programmable in SIMD manner Own instruction-set for neighborhood operations Serial ALU

Current Results FPGA Prototype Partitions of 4x4 tested on a Xilinx Virtex2 FPGA PowerPC HardIP core as CPU Global Memory implemented as external DDR Ram ASIC Implementation Leon3 SoftIP core as CPU Own DMA controller Layout with 32x32 partition SRAM macrocellblocks for partitions and program memory Timings At industrial resolution of 320x240 we achieve 91 FPS In VGA resolution 23 FPS 40 times faster than an embedded Intel Atom Processor

Bachelor and Master Theses

Contact