WO1994003869A1 - Dispositif electronique pour l'analyse d'image et la vision artificielle - Google Patents
Dispositif electronique pour l'analyse d'image et la vision artificielle Download PDFInfo
- Publication number
- WO1994003869A1 WO1994003869A1 PCT/FR1993/000778 FR9300778W WO9403869A1 WO 1994003869 A1 WO1994003869 A1 WO 1994003869A1 FR 9300778 W FR9300778 W FR 9300778W WO 9403869 A1 WO9403869 A1 WO 9403869A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- processors
- processor
- neighboring
- bit
- inputs
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
Definitions
- the present invention relates to an electronic device for image analysis and artificial vision, with parallel operation, organized in a mesh of n-dimensional processors and based on asynchronous global calculations.
- Machine vision can, schematically, be divided into three phases:
- the first category of treatment has led to many effective architectural achievements based in particular on the organization of an array of processors. It is not the same for the second category, however essential, which can only be executed efficiently on highly parallel machines, sophisticated, expensive and complex to program.
- Image analysis is based on diverse and a priori irregular data movements, which are therefore difficult to parallelize. It turns out, however, that most of these movements can be reduced to simple calculation primitives.
- the related components of these graphs represent fundamental basic entities in image analysis. Number of treatments can thus be reduced to iterations of applications of associative operators (logical OR, maximum, addition) on related components of these graphs. We will call hereafter "associations" of such applications.
- the object of the invention is to provide a device making it possible to offer higher performance than devices of the known art, for a lower cost, while being programmable from a calculation model that is simple to master.
- the invention relates to an electronic device for image analysis and artificial vision comprising a set of elementary processors in which each processor, associated with an image pixel and interconnected by serial links to n neighboring processors, has a memory internal of limited size and can operate in SIMD mode on its internal data, characterized in that each processor can be connected or not to its n neighboring processors according to a connection graph which is a subgraph of the physical connection graph of the processors, so as to be able to directly manipulate connected components of the image, in that an associative operator can be applied to these connected components by means of asynchronous communications, in that a local organ makes it possible to determine the moment when the data remains stable on the processor, stability detectors of the various processors being centralized by a network of doors OR towards a central controller which can determine the instant of occurrence of the end of calculation.
- connection of a processor with the n neighboring processors at a given time is materialized in an n-bit register, each bit representing one of the neighboring processors and being positioned if there is an incident arc in this connection graph.
- processor from the neighboring processor n AND gates associated with the registers and with the n connections coming from the neighboring processors making it possible to mask at zero the data coming from the neighboring processors which are not linked by an incident arc towards this processor.
- the associative operator can be a global OR, or a global sum, or an operator making it possible to transform an undirected graph into a graph oriented by asynchronous arbitration.
- the device of the invention makes it possible to solve problems posed by image analysis with superior performance and a cost structure and of dimension greatly reduced compared to the devices of the prior art.
- the device of the invention has a certain number of decisive advantages:
- this device is very easily integrated, in particular if it is based on a mesh organization
- the device of the invention makes it possible to improve the low-level image processing times (convolution, median filtering, etc.) by a factor of one to ten depending on the treatments, and the analysis times. image (segmentation into regions, etc.) of one or two orders of magnitude, with technology and number of processors given compared to current machines;
- the device of the invention enables very efficient vision machines to be produced at relatively low cost. Ultimately (ten years), it will even be possible to integrate the device of the invention in a reduced volume ( ⁇ 1dm3). There are many potential areas of application, including:
- FIG. 3 gives a representation of a tree bit counter according to - Figure 4 gives a representation of an arbitration circuit according to the invention.
- FIG. 4 gives a representation of an arbitration circuit according to the invention.
- Many processing algorithms can be reduced to applications associative operators (or associations) on sets of connected pixels, in the sense of a certain graph representing, depending on the case, a region, a curve, etc.
- the essential object of the invention is to define a device allowing rapid application of these associations.
- the device of the invention is based on an interconnection of elementary processors according to a given topology, for example a two-dimensional mesh.
- Each elementary processor is interconnected by serial links to its nearest n neighbors.
- Each processor which has an internal memory of limited size (for example 256 bits), can simultaneously receive information from its neighbor on a wide bit and transmit it to it, which requires two communication wires 12 and 13. All processors are controlled by a common controller which can operate in SIMD mode ("Single Instruction Multiple Data") on its internal data.
- SIMD mode Single Instruction Multiple Data
- a unit 17 for combining the neighboring processors it recovers the outputs of the masking unit 15 and combines them either by a global OR, or by counting bits, or by arbitration;
- an arithmetic and logic unit 18 it is a conventional arithmetic and logic unit on four bits, operating between the output of the combination unit 17 and a register 19. The least significant bit
- this unit observes the output 20 of the arithmetic and logic unit 18 and detects its variations.
- the different outputs 22 of the convergence detection units of the various elementary processors are centralized by a global OR so as to detect a global convergence of the device.
- Local stability detection can, for example, be done by periodically comparing output 20 of ALU 18 with its previous value. When all the outputs 20 have a stable value, the asynchronous calculation has converged;
- a memory 23 which can read and write to register 19, to the masking units 15 of the neighboring processors and to the output of the arithmetic and logic unit 18.
- Each elementary processor also includes various communications buses between the memory and some of the diagrammed resources, as well as multiplexers making it possible also to use the arithmetic and logical unit to perform operations between local data. These functions, presenting no originality, are not shown in FIG. 2.
- the tree bit counter shown in Figure 3, includes seven adders with three inputs (two bits and one carry) and two outputs (sum and carry).
- the first three adders 25, 26 and 27 receive the eight bits 24 to be considered; the sums of the first three adders 25, 26 and 27 are connected to the inputs of a fourth adder 28, the sum of which constitutes the least significant output bit S1;
- the retentions of the adders 25, 26 and 27 are connected to the inputs of a fifth adder 29; the carry of the fourth adder 28 and the sum of the fifth adder
- the model of the device of the invention makes it possible to gain approximately an order of magnitude on the application of a general convolution, a factor of 3 on the maximum filtering and more than an order of magnitude on the median filtering;
- Objects can be of two types: compact (regions) or filiform (contours). For example, by considering regions, the device of the invention makes it possible to rapidly perform a parallel fusion of these in order to carry out a segmentation of the image (with a constant time per fusion step).
- - oriented graphs most of the considered ones until now were non-oriented.
- each node will r, during Association, the result of the application of the operator to the nodes which are downstream from it.
- a special case of an interesting oriented graph is a chain (two neighbors per node), insofar as it allows to apply in constant time an operator with a parallel prefix ("segmented scan"), whether along an axis or a curve any.
- a similar interest presents itself for a tree which makes it possible to carry out searches quickly;
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
Description
Claims
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP93917839A EP0606458A1 (fr) | 1992-07-31 | 1993-07-29 | Dispositif electronique pour l'analyse d'image et la vision artificielle |
JP6505054A JPH06511586A (ja) | 1992-07-31 | 1993-07-29 | 人工視覚画像解析用電子デバイス |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
FR92/09529 | 1992-07-31 | ||
FR9209529A FR2694430B1 (fr) | 1992-07-31 | 1992-07-31 | Dispositif électronique pour l'analyse d'image et la vision artificielle. |
Publications (1)
Publication Number | Publication Date |
---|---|
WO1994003869A1 true WO1994003869A1 (fr) | 1994-02-17 |
Family
ID=9432512
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/FR1993/000778 WO1994003869A1 (fr) | 1992-07-31 | 1993-07-29 | Dispositif electronique pour l'analyse d'image et la vision artificielle |
Country Status (4)
Country | Link |
---|---|
EP (1) | EP0606458A1 (fr) |
JP (1) | JPH06511586A (fr) |
FR (1) | FR2694430B1 (fr) |
WO (1) | WO1994003869A1 (fr) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109997154A (zh) * | 2017-10-30 | 2019-07-09 | 上海寒武纪信息科技有限公司 | 信息处理方法及终端设备 |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2129589A (en) * | 1982-11-08 | 1984-05-16 | Nat Res Dev | Array processor cell |
-
1992
- 1992-07-31 FR FR9209529A patent/FR2694430B1/fr not_active Expired - Fee Related
-
1993
- 1993-07-29 JP JP6505054A patent/JPH06511586A/ja active Pending
- 1993-07-29 WO PCT/FR1993/000778 patent/WO1994003869A1/fr not_active Application Discontinuation
- 1993-07-29 EP EP93917839A patent/EP0606458A1/fr not_active Withdrawn
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2129589A (en) * | 1982-11-08 | 1984-05-16 | Nat Res Dev | Array processor cell |
Non-Patent Citations (1)
Title |
---|
LI H ET AL.: "reconfigurable simd massively parallel computers", PROCEEDINGS OF THE IEEE, vol. 79, no. 4, April 1991 (1991-04-01), NEW YORK US, pages 429 - 443, XP000238523, DOI: doi:10.1109/5.92038 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109997154A (zh) * | 2017-10-30 | 2019-07-09 | 上海寒武纪信息科技有限公司 | 信息处理方法及终端设备 |
Also Published As
Publication number | Publication date |
---|---|
JPH06511586A (ja) | 1994-12-22 |
EP0606458A1 (fr) | 1994-07-20 |
FR2694430A1 (fr) | 1994-02-04 |
FR2694430B1 (fr) | 1994-09-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP3660849B1 (fr) | Circuit mémoire adapté à mettre en oeuvre des opérations de calcul | |
EP0552074B1 (fr) | Système de traitement de données multiprocesseur | |
EP3053108B1 (fr) | Circuit electronique, notamment apte a l'implementation d'un reseau de neurones, et systeme neuronal | |
EP3084588B1 (fr) | Module de traitement du signal, notamment pour reseau de neurones et circuit neuronal. | |
CA1289261C (fr) | Systeme de traitement d'images pour reseau maille polymorphe | |
Lopich et al. | Asynchronous cellular logic network as a co‐processor for a general‐purpose massively parallel array | |
CA2401422A1 (fr) | Procede et dispositif de perception automatique | |
FR2850768A1 (fr) | Dispositif electronique configurable a granularite mixte | |
EP0206892B1 (fr) | Procédé de traitement de signaux numérisés représentatifs d'une image origine | |
WO1994003869A1 (fr) | Dispositif electronique pour l'analyse d'image et la vision artificielle | |
EP3394799A1 (fr) | Circuit electronique, notamment apte a l'implementation de reseaux de neurones a plusieurs niveaux de precision | |
Liu et al. | A new systolic array algorithm for discrete Fourier transform | |
Galilee et al. | Parallel asynchronous watershed algorithm-architecture | |
EP0540402B1 (fr) | Réseau de résistances binaires et son utilisation pour l'étiquetage de composantes connexes d'images numérisées en vision artificielle | |
CN117292433A (zh) | 一种基于FPGA部署的改进Yolov8行人检测系统 | |
EP0780775A1 (fr) | Architecture d'un système de tableaux de processeurs à structures parallèles multiples | |
EP4202770A1 (fr) | Reseau de neurones avec generation a la volee des parametres du reseau | |
FR2563349A1 (fr) | Multiplieur matriciel systolique de traitement de donnees numeriques | |
EP4242855B1 (fr) | Générateur d'adresses pour un calculateur à architecture de type " instruction unique, données multiples | |
FR3144687A1 (fr) | Architecture de réseau neuronal pour un réseau systolique de processeurs et procédé de traitement de données utilisant un réseau neuronal | |
Dutta et al. | Low Latency and Area Efficient Very Large Scale Integration Architecture of 2-Dimensional Bicubic Interpolation using Carry Save Adder Based Fast Multiplier | |
FR2918190A1 (fr) | Dispositif d'adressage pour processeur parallele. | |
Zhang et al. | Universal Pre-Calculating Structure: Reducing Complexity of Ising Chips with Arbitrary Connectivity | |
Valkenburg et al. | Parallel implementation of vision algorithms on a hybrid pipelined/multitransputer architecture | |
Andreou et al. | Neuromorphic Chiplet Architecture for Wide Area Motion Imagery Processing |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A1 Designated state(s): JP US |
|
AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): AT BE CH DE DK ES FR GB GR IE IT LU MC NL PT SE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1993917839 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref country code: US Ref document number: 1994 211335 Date of ref document: 19940426 Kind code of ref document: A Format of ref document f/p: F |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
WWP | Wipo information: published in national office |
Ref document number: 1993917839 Country of ref document: EP |
|
WWW | Wipo information: withdrawn in national office |
Ref document number: 1993917839 Country of ref document: EP |