2008 · 22 citations · 1 references
EngineeringComputer EngineeringComputer ScienceComputer VisionMachine VisionReconfigurable ComputingDigital Image ProcessingField-programmable Gate Array ArchitectureFpga ArchitectureImage AnalysisImage CompressionImage ProcessingBlob DetectionDigital Image AnalysisImage Processing AcceleratorsImage Processing UnitsMachine Vision SystemsObject PropertiesRun Length Analysis
In the fields of machine vision and image processing applications the blob analysis became a well known method to detect objects in a digital image. Together with the increasing resolutions and frame rates of recent digital video cameras the requirements on the hardware to perform the blob analysis are very high. Software implementations may not be able to accomplish a satisfying performance. Furthermore, existing hardware solutions require a processing of the picture in multiple passes. This paper describes the development of a novel FPGA algorithm, performing a high speed real-time blob analysis using only one single pass. The input data are compressed using run length encoding. This allows the use of a full configuration camera link interface with data rates of up to 680MByte/s. Furthermore, object properties are passed to the host PC on-the-fly which reduces the latency to a minimum while at the same time allows the reuse of object labels thus achieving a memory efficient implementation. 1.
1
Visual pattern recognition by moment invariants
Ming-Kuei Hu · IRE Transactions on Information Theory · 1962 · 7.5K citations