18
Architecture of Specialized Proces in FPGA Structures a Real-Time Video Data Pre-proces Kazimierz Wiatr Institute of Electronics, AGH Technical University of Cracow, POLAND PDPTA ‘99 元元元元 元元元元元 元元元 1999 11.27

MISD Architecture of Specialized Processors

  • Upload
    wood

  • View
    42

  • Download
    0

Embed Size (px)

DESCRIPTION

MISD Architecture of Specialized Processors in FPGA Structures for a Real-Time Video Data Pre-processing Kazimierz Wiatr Institute of Electronics, AGH Technical University of Cracow, POLAND PDPTA ‘99 元智大學 系統實驗室 楊登傑 1999 11.27. Tasks of real-time image analysis. - PowerPoint PPT Presentation

Citation preview

Page 1: MISD Architecture of Specialized Processors

MISD Architecture of Specialized Processors in FPGA Structures

for a Real-Time Video Data Pre-processing

Kazimierz WiatrInstitute of Electronics, AGH

Technical University of Cracow, POLAND

PDPTA ‘99元智大學 系統實驗室 楊登傑

1999 11.27

Page 2: MISD Architecture of Specialized Processors

Tasks of real-time image analysis

• FPGA: Field Programmable Gate Array.

• MISD:Multiple Instruction-stream Single Data-stream

• The vision signal real-time processing for control systems require high computation powers.

• The author took an effort to search different solutions:

– 1.Be related to using specialized hardware structures for implementing various operations.

– 2.To use such intercommunication of those that the architecture is as effective as possible.

Page 3: MISD Architecture of Specialized Processors

Tasks of real-time image analysis(cont.)

• In the algorithms of image analysis,three level are provided:

– (1)Lowest level:called the vision signal pre-processing.

– Function:elimination the interference,drawing the object out of its background,edge detection, adjusting the object greyness level from the histogram,histogram balancing.

– (2)middle level:perform the image segmentation, the object localization, recognizes the image shape and singles out the shape specific features.

Page 4: MISD Architecture of Specialized Processors

Tasks of real-time image analysis(cont.)

– (3)highest level:it is the analysis of the complicated scene:the object movement detection, the object current control, presetting the parameters for low and middle level image processing and analysis.

Page 5: MISD Architecture of Specialized Processors
Page 6: MISD Architecture of Specialized Processors

MISD architecture for video data preprocessing

• The goal of the author was to develop a multiprocessor architecture which -due to the computation elements used and to their interconnection- would result in a very short implementation time of the image preprocessing.

• The efficiency of effective use of the multiprocessor structure is related to the optimized computation tasks to various processors and to the proper data transfer between them.

Page 7: MISD Architecture of Specialized Processors

MISD architecture for video data preprocessing(cont.)

• The most effective here is the multiprocessor pipelined system based on MISD architecture implemented in FPGA structures.

• Pipelined bus enables making use of various independently designed processor modules configurated in a system according to what is needed.

Page 8: MISD Architecture of Specialized Processors

MISD architecture for video data preprocessing(cont.)

• In fig 3,processors P interconnected by a pipelined bus composed of the video data and the control signals.

• Processors are accessible from the VME bus level.

• The logic module IL serves the interrupt signals and their transfer onto VME bus.

Page 9: MISD Architecture of Specialized Processors
Page 10: MISD Architecture of Specialized Processors

Pipeline morphological processor

• In fig 4,a diagram of a morphological processor works in the pipeline architecture.

• The whole logic is placed in an FPGA programmable structure.

• Since a morphological processor realizes context operations, it was necessary to use two external delay lines of 512x1 bit organization.

Page 11: MISD Architecture of Specialized Processors

Pipeline morphological processor(cont.)

• The processor logic consists of three register groups(9 1-bit registers in each group)and two comparator groups(9 1-bit comparators in each group).

• R-I:is to memorize the transformed point together with its environment.

• R-II:includes values of individual points of the structural element,but what is important on this level are only 0 and 1 values.

Page 12: MISD Architecture of Specialized Processors

Pipeline morphological processor(cont.)

• R-III:memorizes,which points of the structural element are not taken into account in the course of comparing.

• C-I:compare values of image points and the structural element.

• C-II:pass the results of these comparisons, which refer to the points disregarded.

• AND:logical product of the second comparator set outputs is performed.

• Fig 5,presents implementation of morphological processor in FPGA structure.

Page 13: MISD Architecture of Specialized Processors
Page 14: MISD Architecture of Specialized Processors

Fig 5,presents implementation of morphological processor in FPGA structure.

Page 15: MISD Architecture of Specialized Processors

Reconfigurable pipelined processors

• The author’s universal reconfigurable pipelined processor is a module comprising three parts:FPAG structure,triple-port memory(TPRAM),two FIFO buffers.

• This enables any sequence of pre-processing operations of the images produced by hardware processors,with on need to physically relocate the dedicated modules of specialized processors.

Page 16: MISD Architecture of Specialized Processors

Reconfigurable pipelined processors(cont.)

• The Triple-port memory TPRAM(MT43c4257)enable the logic operations to be performed on two images,one from the camera,and the other(written by the master processor)from the bus,or it is one of the preceding images.

Page 17: MISD Architecture of Specialized Processors
Page 18: MISD Architecture of Specialized Processors

Conclusion

• The exemplary hardware processors herein described to operate in the pipelined bus developed by the author are not open to the changes in the image processing algorithm.

• New options in this area are related to the use of FPGA programmable systems of high integration scale, their configuration to be written in RAM memory.