Higher Order Autocorrelation Vision Chip

Ishii, Idaku; Yamamoto, Kenkichi; Kubozono, M.

doi:10.1109/ted.2006.878024

Cited by 33 publications

(12 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Although there are dozens of variants for different task solution, the traditional neuronet computings is extremely complex in hardware implementation and require complex settings on the task. As a prototype, it is proposed to use methods of perception organizing, information processing and neurons communication in the retina of the human eye [6,7,[11][12][13][24][25][26][27][28][29][30][31]:…”

Section: Architectures Of Parallel Perception and Processing Of Informentioning

confidence: 99%

Directions of Development of Intelligent Real Time Video Systems

Boyun¹

2017

Appl. Theory Comput. Technol.

View full text Add to dashboard Cite

Real time video systems play a significant role in many fields of science and technology. The range of their applications is constantly increasing together with requirements to them, especially it concerns to real time video systems with the feedbacks. Conventional fundamentals and principles of real-time video systems construction are extremely redundant and do not take into consideration the peculiarities of real time processing and tasks, therefore they do not meet the system requirements neither in technical plan nor in informational and methodical one. Therefore, the purpose of this research is to increase responsiveness, productivity and effectiveness of real time video systems with a feedback during the operation with the high-speed objects and dynamic processes. The human visual analyzer is considered as a prototype for the construction of intelligent real time video systems. Fundamental functions, structural and physical peculiarities of adaptation and processes taking place in a visual analyzer relating to the information processing, are considered. High selectivity of information perception and wide parallelism of information processing on the retinal neuron layers and on the higher brain levels are most important peculiarities of a visual analyzer for systems with the feedback. The paper considers two directions of development of intelligent real time video systems. First direction based on increasing intellectuality of video systems at the cost of development of new information and dynamic models for video information perception processes, principles of control and reading parameters of video information from the sensor, adapting them to the requirements of concrete task, and combining of input processes with data processing. Second direction is associated with the development of new architectures for parallel perception and level-based processing of information directly on a video sensor matrix. The principles of annular and linear structures on the neurons layers, of close-range interaction and specialization of layers, are used to simplify the neuron network.

show abstract

Section: Architectures Of Parallel Perception and Processing Of Informentioning

confidence: 99%

Directions of Development of Intelligent Real Time Video Systems

Boyun¹

2017

Appl. Theory Comput. Technol.

View full text Add to dashboard Cite

show abstract

“…1 计算机视觉技术在最近十几年发展迅速，应用领域不断拓展，针对不同的对象，研究人员提出了多种多样的图像处理算法，这种技术的进步使得计算机视觉正在代替人眼，完成各种繁重复杂的工业现场检测和监控任务 [1][2][3] 。目前，传统的图像传感器 * 留学回国人员基金(M15C300010)和中国铁路总公司科技研究开发计划课题(2014G009-G)资助项目。20150510 收到初稿，20151110 收到修改稿的采样速度不超过 100 帧/s，随着高速智能相机的研发，视觉系统每秒存储数据可达 660 Mb [4] ，图像采集速度可达到 250 帧/s、500 帧/s、1 000 帧/s 甚至更高，这些高速相机可以对高速运动现象进行快速捕捉，清晰的还原人眼无法观测的高速运动过程 [5][6][7][8] 。现代机械工程领域，在大规模工业生产和交通运输设备方面，生产线和载运工具的运动速度越来越快，无法用普通相机进行检测。例如高速全自动贴片机的元件贴片时间可在几 ms 内完成；高速印刷、高速加工等工业生产；汽车发动机、火车轮对和飞机螺旋桨、涡轮发动机的高速旋转等高速机械运动。工业中可执行的高速机械运动，往往以往复式、重复式等周期运动体现，被人类控制用于完成固定的动作。对于这些高速机械系统的运动特性检测，往往利用转速传感器、红外激光等手段进行部分零件的速度和频率检测，但基于视觉技术的高速检测，国内尚无相关研究；大部分针对行人步态识别、动物行为识别等，算法复杂不易用于高速实时检测。目前，针对高速运动现象的检测，国内外研究机构开发了高速视觉平台进行相关研究。NIE 等 [9] 利用 240 帧/s 的高速相机对试验鼠进行长时间的高速录像，提出了试验鼠典型运动自动提取算法，可应用于药物试验中。ISHII 等 [10] 利用高速视觉系统对附着标记点的动物心脏进行 1 000 帧/s 的高速图像采集，离线处理重建心脏三维运动状态。高速智能相机主要利用高速图像处理算法，对每一帧图像进行毫秒级的高速计算，智能判断运动特性，可进行长时间的高速实时检测，或提取特征信息和选择性录取高速视频。马波等 [11] 利用高速摄像机获取医用安瓿的序列图像，结合电气控制系统，设计了视觉检测系统，用于检测安瓿中可见的异物。 GU 等 [12][13] 利用 FPGA 并行处理技术和高速相机实现了多目标特征的实时提取算法，WATANABE 等 [14][15] 开发了高速视觉系统，并应用于高速文档识别，机器手运动控制和物体的高速实时跟踪等领域。毕昕等针对液晶显示器制造中的 Mura 缺陷，提出了基于机器视觉的检测方法，分别研究基于小波滤波、同态变换和主动轮廓模型等理论，建立 Mura 缺陷自动检测方法 [16] 。张正涛等 [17] [18] 。然而，在利用 GPU 处理的过程中，图像数据在 CPU 与 GPU 之间的双向传输占用较多时间，存在丢帧现象，不利于毫秒级的高速实时计算，难以对高速运动现象完成准确检测。因此，本研究主要利用高速视觉平台，设计高速图像处理算法，对每一帧当前图像利用软件程序执行算法，完成高速实时计算。 ( , , ) 0…”

Section: 前言unclassified

Research on Real-time Vision Detection Technology for Motion Feature of High-speed Mechanical System

Wang¹

2016

JME

View full text Add to dashboard Cite

Abstract：A high-speed image sensor based high-speed image processing algorithm is presented, which can real-time detect the period motion feature from mechanical system high-speed motion, and builds a high-speed vision system which can long-term non-contact calculate the motion frequency of complex high-speed mechanical devices. This study considers the high-speed camera as multi-sensors, and perform offline image processing for multi-period high-frame-rate images. Based on the selected key pixels considered as the multi-sensors, converting the images to one dimension vectors, period phase reference-images-sequence is created from pre-sotred original image data. To verify the algorithm and system, this study utilizes a rotation device drove by stepper motor and a mechanical motion of sewing machine for experiments. Experimental results confirm that the algorithms and system can select the key feature pixels, create reference data, and finish the high-speed real-time vision detection for high-speed motion.Key words：high-speed motion；vision sensor；vision detection；motion feature；high-speed camera 0 前言 1

show abstract

“…In contrast, for improving the integration technology, recently, various high speed vision systems have been developed such as vision chips [5]- [8] that integrate sensors and processing elements into one chip and a high-speed vision [9] that processes a mega-pixel image with 1000 fps in real time. In particular, because 3D profile calculations with the lightsection method can actualize with simple processing that is suited for integration, many high-speed smart sensors [10]- [14] which specialize in the 3D measurements have been developed.…”

Section: Conventional Workmentioning

confidence: 99%

A High-Speed 3D Shape Measurement System Using a Multi-Sided Mirror

Yamamoto

Fujii

Ishii

2007

2007 IEEE International Conference on Automation Science and Engineering

View full text Add to dashboard Cite

In this paper, the authors propose a new concept for high spatial resolution 3D shape measurement based on the light-section method by setting a multi-sided mirror between an image sensor and measured object. The authors developed a prototype system for 3D measurement based on a high-speed vision hardware, which can process a 1024×1024 pixel image at 1000 fps, and evaluated its effectiveness by showing the experimental results for a printed circuit board.

show abstract

Higher Order Autocorrelation Vision Chip

Cited by 33 publications

References 10 publications

Directions of Development of Intelligent Real Time Video Systems

Directions of Development of Intelligent Real Time Video Systems

Research on Real-time Vision Detection Technology for Motion Feature of High-speed Mechanical System

A High-Speed 3D Shape Measurement System Using a Multi-Sided Mirror

Contact Info

Product

Resources

About