CN114584674A - Visual integration system for processing same image - Google Patents
Visual integration system for processing same image Download PDFInfo
- Publication number
- CN114584674A CN114584674A CN202210483316.5A CN202210483316A CN114584674A CN 114584674 A CN114584674 A CN 114584674A CN 202210483316 A CN202210483316 A CN 202210483316A CN 114584674 A CN114584674 A CN 114584674A
- Authority
- CN
- China
- Prior art keywords
- image
- module
- image processing
- convolution
- interface
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/14—Picture signal circuitry for video frequency region
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/50—Constructional details
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/04—Synchronising
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Abstract
The invention discloses a visual integration system for processing the same image, which comprises: the system comprises an image acquisition unit, a controller, an image storage module, a communication module, an image processing module and a visual display module; the controller is respectively connected with the image storage module, the image acquisition unit, the communication module and the image processing module, the image processing module is connected with the communication module through a communication interface, the image acquisition unit is connected with a multi-channel data interface, and the image acquisition unit acquires data information of the camera through the multi-channel data interface; the image processing module is also connected with the visual display module through a communication interface. The invention can synchronously process the same image data information, improves the image processing efficiency through the acceleration module, realizes image processing through different modes, can realize different analysis effects on the same image processing, and further improves the image data information processing and analyzing capability.
Description
Technical Field
The present invention relates to the field of image data processing technology, and more particularly to a vision integration system for processing the same image.
Background
The image processing technology is a technology for realizing image information processing by using a computer technology means, and mainly comprises different modes such as image digitization, image enhancement and restoration, image data coding, image segmentation, image identification and the like. The vision integrated system (also called as an intelligent camera) integrates image acquisition, processing and communication functions, and provides a machine vision solution which has multiple functions, modularization, high reliability and easy realization.
In the image processing process, most of the prior art adopts the modes of denoising and the like to realize data information processing so as to improve the definition of image data information. When the binary gray scale processing is required, the gray scale value calculation and the image data information processing need to be performed on the image data information again, which easily causes the image data information processing efficiency to be low. In the conventional image processing technology, the image processing speed is relatively slow, the size is small, the networking degree is low, and the processing hardware is not easy to carry, so that the image processing capability is delayed.
Disclosure of Invention
Aiming at the defects of the technology, the invention discloses a visual integration system for processing the same image, which can synchronously process the data information of the same image, improve the image processing efficiency through an acceleration module, realize image processing through different modes and further improve the image information analysis and application capability.
In order to realize the technical scheme, the application provides the following technical scheme:
a vision integration system for processing identical images, comprising:
an image acquisition unit; the image acquisition unit comprises an image sensor, a dual-port RAM interface, an SH4 chip processor and an acceleration module, and realizes external data information interaction through an I/O interface;
a controller; the controller is used for controlling other modules to be in a working state and represents a multi-axis integrated motion controller combined by a DSP chip and an FPGA chip; the intelligent control system comprises a DSP + FPGA chip, and an electromagnetic valve, a relay, a servo driver, a servo motor, a monitoring module, a communication module, a human-computer interaction module and a display which are connected with the DSP + FPGA chip;
an image storage module; the storage of image data information is realized;
a communication module; the data information interaction module is used for transmitting the received data information to other modules so as to realize data information interaction;
an image processing module; the image processing device is used for outputting different image information according to different processing rules for the same image data information, and realizing the output of processing results of different image data information; the image processing module comprises a compatible data interface, a first image processing module, a second image processing module and a visual output interface; the output end of the data interface is connected with the input ends of the first image processing module and the second image processing module, and the output ends of the first image processing module and the second image processing module are connected with the input end of the visual output interface; the first image processing module represents an image processing module based on local texture feature extraction of a hyperchaotic image, the second image processing module represents an image identification module based on a direction gradient histogram, and different types of output of the same image are realized through a visual output interface;
the visual display module is used for realizing data output after the same image data information is processed according to different processing rules;
the controller is respectively connected with the image storage module, the image acquisition unit, the communication module and the image processing module, the image processing module is connected with the communication module through a communication interface, the image acquisition unit is connected with a multi-channel data interface, and the image acquisition unit acquires data information of the camera through the multi-channel data interface; the image processing module is also connected with the visual display module through a communication interface.
As a further technical solution of the present invention, the multi-channel interface at least includes an RS232 communication channel interface, an RS485 communication channel interface, a carrier communication channel interface, a TCP/IP communication channel interface, an RS422 communication channel interface, an ethernet communication channel interface, a CAN communication channel interface, a USB communication channel interface, a WIFI communication channel interface, a ZigBee communication channel interface, a bluetooth communication channel interface, or an optical fiber communication channel interface; the multi-channel interface further comprises a cloud data interaction port.
As a further technical scheme of the invention, the acceleration module comprises an ARM processor, an on-chip memory, an off-chip memory, a storage module and 2 mutually cascaded convolution accelerators, wherein the on-chip memory, the off-chip memory and the storage module are connected with the ARM controller.
As a further technical scheme of the invention, the method for realizing convolution acceleration by the convolution accelerator is represented as follows:
(1) the data parameters of the convolution accelerator are set up,
andrespectively representThe length and width of the layer convolution kernel,whereinAndrespectively representing the step size of a convolution kernel, and 2 convolution accelerators which are cascaded mutually output data information results to represent:
in the formula (1), the reaction mixture is,、andthe activation function and the input mapping respectively representing the current layer are subjected to convolution operation and the secondLayer oneBias of individual feature maps; obtaining neurons of a convolutional neural networkAnd then performs an upsampling operation with respect to the downsampled layer,downsampled layer after upsamplingThe function represents:
in the formulae (2) to (3),、andrespectively representing the sampling factor of the convolution acceleratorFirst of a layerUp-sampling information of each neuron and the down-sampling layer;
down-sampling layer learning function representation:
in the formula (4), the reaction mixture is,downsampling information representing an upsampled layer;
(2) the deep convolutional neural network is in a convergence state, and the last full-connection layer of the deep convolutional neural network is used as a new graphImage features, representation;
(3) According toObtaining new cluster labelsAnd the output of the deep convolutional neural network in the next iteration process is represented;
(4) and setting the iteration times to represent 100 times, and when the iteration times are equal to 100, selecting a noise reduction automatic encoder to further obtain the compressed image code, thereby improving the data acceleration capability.
As a further technical solution of the present invention, the human-computer interaction module represents an FX3U-64MR-ES model controller. The first image processing module comprises an image information extraction module, an image texture convolution module, a gray value calculation module and a brightening module, wherein the output end of the image information extraction module is connected with the input end of the image texture convolution module, the output end of the image texture convolution module is connected with the input end of the gray value calculation module, the output end of the gray value calculation module is connected with the input end of the brightening module, and the image information extraction module represents a filter.
As a further technical solution of the present invention, a processing method of the first image processing module includes the following steps:
and (3) extracting image problem features by adopting a filter, and expressing an output image filtering function:
in the formula (5), the reaction mixture is,which represents the center frequency of the filter and,a constant that controls the bandwidth of the radial filter,a representation directional bandwidth determination parameter;
in the formula (6), the reaction mixture is,a feature map representing the texture of the image,is shown inA map of the textural features of the image at scale,is shown inAnd (5) performing image convolution corresponding to the texture feature map of the image under the scale.
the function formula expresses:
in the formula (7), the reaction mixture is,representing the average value of the grey scale of the pixels of the image,the visual constant is represented by a visual constant,which represents the size of the image window and,to representThe gray value of the image pixel under the position;
in the formula (8), the reaction mixture is,represents the local brightness in the hyperchaotic image,which represents the local background brightness of the image,representing the high frequency components of the image,representing an image edge variation parameter.
As a further technical solution of the present invention, the second image processing module includes a gray-scale image normalization module, a pixel gradient calculation module, a pixel normalization calculation module, and a feature vector generation module, wherein an output end of the gray-scale image normalization module is connected to an input end of the pixel gradient calculation module, an output end of the pixel gradient calculation module is connected to an input end of the pixel normalization calculation module, and an output end of the pixel normalization calculation module is connected to an input end of the feature vector generation module.
As a further technical solution of the present invention, a method of processing an image by a second image processing module includes the steps of:
step 1: converting the image data collected by the second image processing module into a gray scale map, and performing normalization processing through the following functions:
in the formula (9), the reaction mixture is,T(a,b) A gray value representing image data information collected by the second image processing module;
step 2: gradient calculation, calculating the image in pixels (a,b) Gradient values of the points, the gradient function formula is:
in the formula (10), the compound represented by the formula (10),L a (a,b) Represents the horizontal gradient values of the image,L b (a,b) Represents the vertical gradient values of the image,H(a,b) Pixel values representing an image at pixel pointsa,b) The gradient vector of (a) is:
in the formulas (11) to (12),L(a,b) A gradient value representing the image is determined,representing the gradient direction of the image;
and step 3: constructing a directional gradient histogram, dividing the image into a plurality of modules, and performing normalization processing;
and 4, step 4: generating characteristic vectors, wherein each normalization module has partially overlapped directional gradient histogram characteristics, extracting the characteristics to generate the characteristic vectors, and after the directional gradient histogram characteristics are obtained, machine learning is carried out
The linear classifier isf(x) Eliminating redundant information in the image, the linear classifier isf(x) The expression function of (a) is:
in the formula (13), the reaction mixture is,Rrepresents one parameter of the linear classifier and is,xrepresents a variation of a sample of the HTGS image,za hidden variable is represented by a number of hidden variables,Z(x) The value space of the hidden variable is represented,P(x,z) Representing the HTGS image sample information.
The invention has the following positive beneficial effects:
the invention can synchronously process the same image data information, improves the image processing efficiency through the accelerating module, realizes image processing through different modes and further improves the image information analysis and application capability. The vision integrated system has the characteristics of easy learning, easy use, easy maintenance, convenient installation and the like, and can construct a reliable and effective machine vision system in a short time.
The invention can quickly realize the functions of positioning, geometric measurement, presence/absence detection, counting, character recognition, bar code recognition, color analysis and the like, and is suitable for most machine vision applications. The visual integration system realizes high integration of the image acquisition unit, the image processing module and the network communication device. The invention has small volume, compact structure and small size, is easy to be installed on industrial production lines and various devices, and is convenient to be assembled, disassembled and moved. The vision integration system of the invention generally provides better network functions, and can be applied to each industrial monitoring point in real time by virtue of network advantages. The invention has low cost, the vision integrated system integrates the collection and the processing, and the PC system and the image collection card are not needed to be configured, thereby greatly reducing the cost of the vision system.
The invention can realize different analysis effects on the same image processing, thereby improving the image data information processing and analyzing capability.
Drawings
In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without inventive exercise, wherein:
FIG. 1 is a schematic diagram of the overall architecture of the present invention;
FIG. 2 is a schematic diagram of an image capture module according to the present invention;
FIG. 3 is a schematic diagram of an acceleration module architecture according to the present invention;
FIG. 4 is a schematic diagram of a control module architecture according to the present invention;
FIG. 5 is a schematic diagram of a human-computer interaction module architecture according to the present invention;
FIG. 6 is a block diagram of an image processing module according to the present invention;
FIG. 7 is a block diagram of a first image processing module according to the present invention;
FIG. 8 is a block diagram of a second image processing module according to the present invention;
FIG. 9 is a schematic diagram of an acceleration method of the accelerator according to the present invention;
FIG. 10 is a diagram illustrating a first image processing module processing an image according to the present invention;
FIG. 11 is a diagram illustrating an image processing module according to a second embodiment of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, and it should be understood that the embodiments described herein are merely for purposes of illustration and explanation, and are not intended to limit the present invention.
A vision integration system that processes identical images, comprising:
an image acquisition unit; the image acquisition unit comprises an image sensor, a dual-port RAM interface, an SH4 chip processor and an acceleration module, and realizes external data information interaction through an I/O interface;
a controller; the controller is used for controlling other modules to be in a working state and represents a multi-axis integrated motion controller combined by a DSP chip and an FPGA chip; the intelligent monitoring system comprises a DSP + FPGA chip, and an electromagnetic valve, a relay, a servo driver, a servo motor, a monitoring module, a communication module, a human-computer interaction module and a display which are connected with the DSP + FPGA chip;
an image storage module; the storage of image data information is realized;
a communication module; the data information interaction module is used for transmitting the received data information to other modules so as to realize data information interaction;
an image processing module; the image processing device is used for outputting different image information according to different processing rules for the same image data information, and realizing the output of processing results of different image data information; the image processing module comprises a compatible data interface, a first image processing module, a second image processing module and a visual output interface; the output end of the data interface is connected with the input ends of the first image processing module and the second image processing module, and the output ends of the first image processing module and the second image processing module are connected with the input end of the visual output interface; the first image processing module represents an image processing module based on local texture feature extraction of a hyperchaotic image, the second image processing module represents an image identification module based on a direction gradient histogram, and different types of output of the same image are realized through a visual output interface;
the visual display module is used for realizing data output after the same image data information is processed according to different processing rules;
the controller is respectively connected with the image storage module, the image acquisition unit, the communication module and the image processing module, the image processing module is connected with the communication module through a communication interface, the image acquisition unit is connected with a multi-channel data interface, and the image acquisition unit acquires data information of the camera through the multi-channel data interface; the image processing module is also connected with the visual display module through a communication interface;
in the above embodiment, the multi-channel interface at least includes an RS232 communication channel interface, an RS485 communication channel interface, a carrier communication channel interface, a TCP/IP communication channel interface, an RS422 communication channel interface, an ethernet communication channel interface, a CAN communication channel interface, a USB communication channel interface, a WIFI communication channel interface, a ZigBee communication channel interface, a bluetooth communication channel interface, or an optical fiber communication channel interface; the multi-channel interface further comprises a cloud data interaction port.
In the above embodiments, it is shown that the data information acquisition capability is improved, and a high-speed and high-resolution camera, a high-speed processor, and the like need to be configured. For the vision integration system, due to the integrated design, the system is limited by the processor, the memory and the like, and complex operations are not performed on large data volumes. The vision integration system can not be compared with PC-BASED at present in occasions requiring high speed and high precision. The PC-Based system is expected to adapt to complex applications and can be configured and controlled more flexibly. Algorithmically, complex operations can be implemented in a variety of high-level languages. When the precision is to be improved, the method can be realized by improving the system configuration and increasing the number of cameras. The invention is provided with the acceleration module, and can improve the data information acquisition capability.
In the specific application process, the image acquisition unit comprises 5 main chips: the system comprises an image acquisition chip OV7620, a high-speed microprocessor SH4, a large-scale programmable array FPGA and a serial port communication control chip MAX 232. Two dual-port RAMs are programmed in the FPGA to generate dot frequency, line field synchronization and other signals required by the image sensor and control the storage time sequence of the dual-port RAMs. The SH4 is responsible for configuring the OV7620 through the I2C, reading and processing image data of the dual-port RAM, and uploading image data or controlling other devices such as a stepping motor through a serial port. The system module is represented by a CMOS image sensor OV7620, and further comprises a condenser lens and other auxiliary components such as a 27MHz crystal oscillator, a resistor, a capacitor and the like.
The cmos image sensor is a new image sensor which is developed rapidly in recent years, and due to the adoption of the same cmos technology, a pixel array and a peripheral supporting circuit can be integrated on the same Chip, so that the cmos image sensor is a complete image system (Camera on Chip). The system uses a CMOS color image sensor OV7620, marketed by ommvision, with a resolution of 640x 480. It can work in progressive scanning mode and also in interlaced scanning mode. It can not only output color images, but also be used as a black and white image sensor. There are many image output formats supported by this chip: 1) YCrCb4:2: 216 bit/8 bit format; 2) ZV port output format; 3) RGB original data 16 bit/8 bit; 4) CCIR601/CCIR656 format. Its functions include contrast, brightness, saturation, white balance and automatic exposure, synchronous signal position and polarity output, frame rate and output format, etc. all can be controlled by I2C bus to program and configure on-chip register.
The FPGA adopts xc2s100, and 10000 logic gates are integrated in the chip. The interface program is written in VHDL (very High Speed Integrated Circuit Hardware Description language). Indicating an increase in the data transfer rate, 2 dual-port RAM buffers, the size of which indicates 127KB, are allocated inside the xc2s100, each of which stores 1 line of image data. The two sets of dual-port RAMs switch odd and even row counters. When a line is stored, an application signal for an interrupt to read the line is immediately transmitted to SH 4. The read-write operation of the dual-port RAM in the FPGA shares the same data bus and address bus, and when the read-write operation is carried out simultaneously, a time sequence problem is generated to cause data errors in writing or reading. In the two processes, the data and address bus conflict is prevented, and a central bus arbiter is designed inside the FPGA. According to the sequence of the public data transmission, the central arbiter receives the bus request of the image sensor firstly, and the central arbiter responds to the read signal request of the single chip microcomputer system after the image is stored in the RAM. The system uses an SH4 chip as a representation processor: the SH4 singlechip is a full 32-bit singlechip with low power consumption, high performance and RISC (reduced instruction set computer) structure, which is released by Hitachi corporation. The processing speed can reach 60M IPS-100 MIPS, the chip can work under the voltage of 2.25v, and a 32-bit multiplier, a 4-path 5KB CACHE, an access management unit MMU, other general interfaces, clock circuits and the like are integrated in a 400MW chip. Hitachi company shows that SH4 series single-chip microcomputers provide c and c + + language integrated compiling tool HIM (Hitachi integrated Manag). By which a source program compiled link in Hitachi C, C + + format can be represented as an assembler or object machine code. The image sensor chip OV7620 has flexible programmable functions that can be programmed through the I2C bus to set the function registers. Because the singlechip does not have an internal hardware I2C bus interface, the I2C bus interface function is realized only by adopting a software simulation method. The two I/O pins representing SH4 are taken to represent the SCL and SDA bus device interfaces of the I2C bus.
The acceleration module comprises an ARM processor, an on-chip memory, an off-chip memory, a storage module and 2 mutually cascaded convolution accelerators, wherein the on-chip memory, the off-chip memory and the storage module are connected with the ARM controller.
In the above embodiment, the method for realizing convolution acceleration by the convolution accelerator represents that:
(1) setting data parameters of a convolution accelerator, wherein in a specific embodiment, a deep convolution neural network learning process comprises two links, namely convolution layer learning and downsampling layer learning. The convolutional layer performs convolution processing such as local link and weight sharing on the data acquired by the previous layer based on the principle and the network structure that video images with the same source belong to one class, so that the number of links and parameters is reduced.Andrespectively representThe length and width of the layer convolution kernel,whereinAndrespectively representing the step size of a convolution kernel, and 2 convolution accelerators which are cascaded mutually output data information results to represent:
in the formula (1), the reaction mixture is,、andthe activation function and the input mapping respectively representing the current layer are subjected to convolution operation and the secondLayer oneBias of individual feature maps;
neurons of layer +1 are calculationsKey of layer neuron, useTo representAnd (4) respectively calculating the product of the activation function of the layer neuron and the weight function and the gradient value. I.e. representing neurons from which a convolutional neural network is acquiredThen, an up-sampling operation is performed on the down-sampling layer, and the down-sampling layer after up-sampling is performedThe function represents:
in the formulae (2) to (3),、andrespectively representing the sampling factor of the convolution acceleratorFirst of a layerUp-sampling information of each neuron and the down-sampling layer;
the problem of excessive learning-induced fitting of the convolutional layer is solved, and the sampling layer is calculated under the condition of maximum poolingAnd the upper limit value of the regional characteristics is used for simplifying the calculation process, improving the stability of the network model and avoiding the problem of overfitting. Down-sampled layer learning function representation:
in the formula (4), the reaction mixture is,downsampling information representing an upsampled layer;
(2) the deep convolutional neural network is in a convergence state, and the last full-connection layer of the deep convolutional neural network is used as a new image characteristic to represent;
(3) According toObtaining new cluster labelsAnd the output of the deep convolutional neural network in the next iteration process is represented;
(4) and setting the iteration times to represent 100 times, and when the iteration times are equal to 100, selecting a noise reduction automatic encoder to further obtain the compressed image code, thereby improving the data acceleration capability.
In the above embodiment, the human interaction module represents an FX3U-64MR-ES model controller.
In a specific embodiment, the block diagram of the human-computer interaction structure has a power supply, a Central Processing Unit (CPU), a memory, an input module, an output module, a communication interface, a power supply module, an expansion interface module, and the like. The power supply part plays an important role in the whole human-computer interaction structure diagram, can provide system operation voltage and ensures the stability of images obtained by the images; the CPU controls and commands the whole human-computer interaction structure; the memory stores the required hardware and software safely, the system code of the original factory is usually solidified in the system memory, the user can not rewrite the system code in the read-only memory, and the quality of the software code also determines the performance of the PLC; the input/output module is a channel for conveniently receiving signals and feedback signals; the expansion interface module of the human-computer interaction C is mainly responsible for the connection between the PLC and the peripheral module, and ensures effective data communication between the controller and the outside.
In a specific embodiment, the multi-axis integrated motion controller combined by the DSP and the FPGA chip is an 'embedded PC and motion control card integrated' controller, can control up to 8 axes simultaneously, has higher reliability and interference immunity, can realize more complex and accurate camera motion control, does not need a professional touch screen (HMI) for a control system module, is integrated in the controller, and only needs to be connected with a common display. The motion control module does not need to install a hardware PLC, the function of PLC control is realized by writing a software program, the program written on the PLC can be modified according to the actual situation on site, and the motion control module has strong universality and good transportability. The servo module adopts a Sanyo (SANYO) alternating current servo motor, the servo motor has the advantages of small volume and good rigidity, the servo module mainly receives motion instructions of motion control, including the rotation angle, the rotation speed and the torque of the servo motor, has the advantages of small inertia, quick response, stable rotation and the like, and the servo module ensures the stable operation of the camera under the high-speed condition, adopts a driver with high rigidity to transmit and considers other overload capacity.
In the above embodiment, the first image processing module includes an image information extraction module, an image texture convolution module, a gray value calculation module, and a brightening module, where an output end of the image information extraction module is connected to an input end of the image texture convolution module, an output end of the image texture convolution module is connected to an input end of the gray value calculation module, an output end of the gray value calculation module is connected to an input end of the brightening module, and the image information extraction module represents a filter.
In the above embodiment, the processing method of the first image processing module includes the following steps:
in an embodiment, the texture is an important index in the image information, which includes the correlation information between the structural organization of the object surface and the environment, and can be an important visual feature for representing and analyzing the hyperchaotic image information. The distribution of energy in the magnitude spectrum over the various frequency bands, which shows a close relationship with textural features, represents a reduction of filtering redundancy,
and (3) extracting image problem features by adopting a filter, and expressing an output image filtering function:
in the formula (5), the reaction mixture is,which represents the center frequency of the filter and,a constant that controls the bandwidth of the radial filter,a representation directional bandwidth determination parameter;
in the formula (6), the reaction mixture is,a feature map representing the texture of the image,is shown inA map of the textural features of the image at scale,is shown inAnd (5) performing image convolution corresponding to the texture feature map of the image under the scale.
the method is characterized in that the partial visibility of an image is measured, an image visibility concept is introduced, and a function formula is represented as follows:
in the formula (7), the reaction mixture is,representing the average of the image pixel gray levels,the visual constant is represented by a visual constant,which represents the size of the image window and,to representThe gray value of the image pixel under the position;
in the formula (8), the reaction mixture is,represents the local brightness in the hyperchaotic image,which represents the local background brightness of the image,representing the high frequency components of the image,representing an image edge variation parameter.
By the method, the slight change of the local contrast of the hyperchaotic image can be clearly expressed.
In the above embodiment, the second image processing module includes a gray-scale image normalization module, a pixel gradient calculation module, a pixel normalization calculation module, and a feature vector generation module, wherein an output end of the gray-scale image normalization module is connected to an input end of the pixel gradient calculation module, an output end of the pixel gradient calculation module is connected to an input end of the pixel normalization calculation module, and an output end of the pixel normalization calculation module is connected to an input end of the feature vector generation module.
In the above embodiment, the method of processing an image by the second image processing module includes the steps of:
step 1: converting the image data collected by the second image processing module into a gray scale map, and performing normalization processing through the following functions:
in the formula (9), the reaction mixture is,T(a,b) A gray scale value representing image data information collected by the second image processing module;
step 2: gradient calculation, calculating the image in pixels (a,b) Gradient values of the points, the gradient function formula is:
in the formula (10), the reaction mixture is,L a (a,b) Represents the horizontal gradient values of the image,L b (a,b) Represents the vertical gradient values of the image,H(a,b) Pixel values representing an image at pixel pointsa,b) The gradient vector of (a) is:
in the formulas (11) to (12),L(a,b) A gradient value representing the image is determined,representing the gradient direction of the image;
and step 3: constructing a directional gradient histogram, dividing the image into a plurality of modules, and performing normalization processing;
each second image processing module outputs 8 x 8 pixels of the module, dividing the gradient direction of the module into 9 blocks. And performing weighted projection on each pixel in the module in the gradient direction histogram to calculate the gradient direction histogram of the module. And combining and normalizing adjacent modules.
And 4, step 4: generating feature vectors, wherein each normalization module has partially overlapped directional gradient histogram features, extracting the features to generate the feature vectors, and separating two kinds of image features to the maximum extent through a machine learning algorithm after the directional gradient histogram features are obtained. The classifier can simplify the classification problem and is divided intof(x) Eliminating redundant information in the image, the linear classifier isf(x) The expression function of (a) is:
in the formula (13), the reaction mixture is,Rrepresents one parameter of the linear classifier and is,xrepresents a variation of a sample of the HTGS image,za hidden variable is represented by a number of hidden variables,Z(x) The value space of the hidden variable is represented,P(x,z) Representing HTGS image sample information;
in particular embodiments, a linear classifier is trained to obtain optimal parameters primarily by minimizing an objective functionRSpecifically defined as:
in the formula (14), M (R) represents an objective function,is shown asAnd (5) training tasks. For solving the objective function minimization, it can be optimized by fixing R to select the optimum hidden variable value for each positive sample. Square blockThe modularization concept is adopted for the gradient histogram feature, and the areas of the current unit and the four surrounding units are directly normalized during normalization processing, so that the image can be rapidly acquired, and an operator can perform corresponding manual control.
After the calculation, the visual display of the image data information is realized through the visual output interface, and the visual visualization degree of the image is greatly improved. Different data can be output for the same image to be displayed, and the analysis and application capability of the image information is greatly improved.
Although specific embodiments of the present invention have been described above, it will be understood by those skilled in the art that these specific embodiments are merely illustrative and that various omissions, substitutions and changes in the form of the detail of the methods and systems described above may be made by those skilled in the art without departing from the spirit and scope of the invention. For example, it is within the scope of the present invention to combine the steps of the above-described methods to perform substantially the same function in substantially the same way to achieve substantially the same result. Accordingly, the scope of the invention is to be limited only by the following claims.
Claims (8)
1. A vision integration system for processing identical images, characterized by: the method comprises the following steps:
an image acquisition unit; the image acquisition unit comprises an image sensor, a dual-port RAM interface, an SH4 chip processor and an acceleration module, and realizes external data information interaction through an I/O interface;
a controller; the controller is used for controlling other modules to be in a working state and represents a multi-axis integrated motion controller combined by a DSP chip and an FPGA chip; the intelligent monitoring system comprises a DSP + FPGA chip, and an electromagnetic valve, a relay, a servo driver, a servo motor, a monitoring module, a communication module, a human-computer interaction module and a display which are connected with the DSP + FPGA chip;
an image storage module; the storage of image data information is realized;
a communication module; the data information interaction module is used for transmitting the received data information to other modules so as to realize data information interaction;
an image processing module; the image processing device is used for outputting different image information according to different processing rules for the same image data information, and realizing the output of the processing results of different image data information; the image processing module comprises a compatible data interface, a first image processing module, a second image processing module and a visual output interface; the output end of the data interface is connected with the input ends of the first image processing module and the second image processing module, and the output ends of the first image processing module and the second image processing module are connected with the input end of the visual output interface; the first image processing module represents an image processing module based on local texture feature extraction of a hyperchaotic image, the second image processing module represents an image identification module based on a direction gradient histogram, and different types of output of the same image are realized through a visual output interface;
the visual display module is used for realizing data output after the same image data information is processed according to different processing rules;
the camera comprises a controller, an image storage module, an image acquisition unit, a communication module and an image processing module, wherein the controller is respectively connected with the image storage module, the image acquisition unit, the communication module and the image processing module; the image processing module is also connected with the visual display module through a communication interface.
2. A vision integration system for processing identical images according to claim 1, characterized in that: the multichannel data interface at least comprises an RS232 communication channel interface, an RS485 communication channel interface, a carrier communication channel interface, a TCP/IP communication channel interface, an RS422 communication channel interface, an Ethernet communication channel interface, a CAN communication channel interface, a USB communication channel interface, a WIFI communication channel interface, a ZigBee communication channel interface, a Bluetooth communication channel interface or an optical fiber communication channel interface; the multi-channel interface further comprises a cloud data interaction port.
3. A vision integration system for processing identical images according to claim 1, characterized in that: the acceleration module comprises an ARM processor, an on-chip memory, an off-chip memory, a storage module and 2 mutually cascaded convolution accelerators, wherein the on-chip memory, the off-chip memory and the storage module are connected with the ARM controller.
4. A vision integration system for processing identical images according to claim 3, wherein: the convolution accelerator realizes convolution acceleration by the following method:
(1) the data parameters of the convolution accelerator are set up,
andrespectively representThe length and width of the layer convolution kernel,whereinAndrespectively representing the step size of a convolution kernel, and 2 convolution accelerators which are cascaded mutually output data information results to represent:
in the formula (1), the reaction mixture is,、andthe activation function and the input mapping respectively representing the current layer are subjected to convolution operation and the secondLayer oneBiasing of the individual feature maps;
obtaining neurons of a convolutional neural networkThen, an up-sampling operation is performed on the down-sampling layer, and the down-sampling layer after up-sampling is performedThe function represents:
in the formulae (2) to (3),、andrespectively representing the sampling factor of the convolution acceleratorFirst of a layerUp-sampling information of each neuron and the down-sampling layer;
down-sampling layer learning function representation:
in the formula (4), the reaction mixture is,downsampling information representing an upsampled layer;
(2) the deep convolutional neural network is in a convergence state, and the last full-connection layer of the deep convolutional neural network is used as a new image characteristic to represent;
(3) According toObtaining new cluster labelsAnd the output of the deep convolutional neural network in the next iteration process is represented;
(4) and setting the iteration times to represent 100 times, and when the iteration times are equal to 100, selecting a noise reduction automatic encoder to further obtain the compressed image code, thereby improving the data acceleration capability.
5. A vision integration system for processing identical images according to claim 1, characterized in that: the human-computer interaction module represents an FX3U-64MR-ES model controller; the first image processing module comprises an image information extraction module, an image texture convolution module, a gray value calculation module and a brightening module, wherein the output end of the image information extraction module is connected with the input end of the image texture convolution module, the output end of the image texture convolution module is connected with the input end of the gray value calculation module, the output end of the gray value calculation module is connected with the input end of the brightening module, and the image information extraction module represents a filter.
6. A visual integration system for processing identical images according to claim 1 or 5, characterized in that: the processing method of the first image processing module comprises the following steps:
step 1, extracting data information of the acquired image through an image information extraction module, and extracting texture of the content representation image;
and (3) extracting image problem features by adopting a filter, and expressing an output image filtering function:
in the formula (5), the reaction mixture is,which represents the center frequency of the filter and,a constant that controls the bandwidth of the radial filter,a representation directional bandwidth determination parameter;
step 2, realizing convolution of image data information through an image texture convolution module, wherein the extracted image data information has the same bandwidth, and the texture features of the image are defined as follows:
in the formula (6), the reaction mixture is,a feature map representing the texture of the image,is shown inA map of the textural features of the image at scale,is shown inThe image convolution result corresponding to the texture feature map of the image under the scale;
step 3, realizing gray map calculation through a gray value calculation module;
the function formula expresses:
in the formula (7), the reaction mixture is,representing the average value of the grey scale of the pixels of the image,the visual constant is represented by a visual constant,which represents the size of the image window and,to representThe gray value of the image pixel under the position;
step 4, brightness updating of the extracted data information is achieved through the brightening module; so that the input image information has clear brightness, the local brightness contrast is introduced into the local texture feature extraction, and through brightness calculation, the calculation function is as follows:
7. A vision integration system for processing identical images according to claim 5, wherein: the second image processing module comprises a grey scale icon normalization module, a pixel gradient calculation module, a pixel normalization calculation module and a feature vector generation module, wherein the output end of the grey scale icon normalization module is connected with the input end of the pixel gradient calculation module, the output end of the pixel gradient calculation module is connected with the input end of the pixel normalization calculation module, and the output end of the pixel normalization calculation module is connected with the input end of the feature vector generation module.
8. A vision integration system for processing the same image according to claim 1 or 7, characterized in that: the method of processing an image by a second image processing module includes the steps of:
step 1: converting the image data collected by the second image processing module into a gray scale map, and performing normalization processing through the following functions:
in the formula (9), the reaction mixture is,T(a,b) A gray value representing image data information collected by the second image processing module;
and 2, step: gradient calculation, calculating the image in pixels (a,b) Gradient values of the points, the gradient function formula is:
in the formula (10), the compound represented by the formula (10),L a (a,b) Represents the horizontal gradient values of the image,L b (a,b) Represents the vertical gradient values of the image,H(a,b) Pixel values representing an image at pixel pointsa,b) The gradient vector of (a) is:
in the formulas (11) to (12),L(a,b) A gradient value representing the image is determined,representing the gradient direction of the image;
and step 3: constructing a directional gradient histogram, dividing the image into a plurality of modules, and performing normalization processing;
and 4, step 4: generating characteristic vectors, wherein each normalization module has partially overlapped directional gradient histogram characteristics, extracting the characteristics to generate the characteristic vectors, and after the directional gradient histogram characteristics are obtained, machine learning is carried out
The linear classifier isf(x) Eliminating redundant information in the image, the linear classifier isf(x) The expression function of (a) is:
in the formula (13), the reaction mixture is,Rrepresents one parameter of the linear classifier and is,xrepresents one of the HTGS image sample variables,a hidden variable is represented by a number of hidden variables,Z(x) The value space of the hidden variable is represented,P(x,z) Representing the HTGS image sample information.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210483316.5A CN114584674A (en) | 2022-05-06 | 2022-05-06 | Visual integration system for processing same image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210483316.5A CN114584674A (en) | 2022-05-06 | 2022-05-06 | Visual integration system for processing same image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114584674A true CN114584674A (en) | 2022-06-03 |
Family
ID=81778366
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210483316.5A Pending CN114584674A (en) | 2022-05-06 | 2022-05-06 | Visual integration system for processing same image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114584674A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115504187A (en) * | 2022-09-22 | 2022-12-23 | 中信重工开诚智能装备有限公司 | Intelligent speed regulation and protection system for mining belt and control method |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102044063A (en) * | 2010-12-23 | 2011-05-04 | 中国科学院自动化研究所 | FPGA (Field Programmable Gate Array) and DSP (Digital Signal Processor) based machine vision system |
US20190042923A1 (en) * | 2017-08-07 | 2019-02-07 | Intel Corporation | System and method for an optimized winograd convolution accelerator |
CN210476955U (en) * | 2019-05-10 | 2020-05-08 | 深圳市领略数控设备有限公司 | Multi-axis manipulator controller based on ZYNQ |
WO2021249233A1 (en) * | 2020-06-10 | 2021-12-16 | 中铁四局集团有限公司 | Image processing method, target recognition model training method, and target recognition method |
-
2022
- 2022-05-06 CN CN202210483316.5A patent/CN114584674A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102044063A (en) * | 2010-12-23 | 2011-05-04 | 中国科学院自动化研究所 | FPGA (Field Programmable Gate Array) and DSP (Digital Signal Processor) based machine vision system |
US20190042923A1 (en) * | 2017-08-07 | 2019-02-07 | Intel Corporation | System and method for an optimized winograd convolution accelerator |
CN210476955U (en) * | 2019-05-10 | 2020-05-08 | 深圳市领略数控设备有限公司 | Multi-axis manipulator controller based on ZYNQ |
WO2021249233A1 (en) * | 2020-06-10 | 2021-12-16 | 中铁四局集团有限公司 | Image processing method, target recognition model training method, and target recognition method |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115504187A (en) * | 2022-09-22 | 2022-12-23 | 中信重工开诚智能装备有限公司 | Intelligent speed regulation and protection system for mining belt and control method |
CN115504187B (en) * | 2022-09-22 | 2023-06-13 | 中信重工开诚智能装备有限公司 | Intelligent speed regulation and protection system and control method for mining belt |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2018112833A1 (en) | Efficient transferring of human experiences to robots and other autonomous machines | |
CN107545263B (en) | Object detection method and device | |
US20190114546A1 (en) | Refining labeling of time-associated data | |
CN103424409A (en) | Vision detecting system based on DSP | |
AU2018319215A1 (en) | Electronic apparatus and control method thereof | |
CN114584674A (en) | Visual integration system for processing same image | |
CN100465994C (en) | Method and apparatus for downscaling a digital matrix image | |
Chalimbaud et al. | Embedded active vision system based on an FPGA architecture | |
CN113409355A (en) | Moving target identification system and method based on FPGA | |
CN111695408A (en) | Intelligent gesture information recognition system and method and information data processing terminal | |
CN214751405U (en) | Multi-scene universal edge vision motion control system | |
CN112613425B (en) | Target identification system for small sample underwater image | |
Sun et al. | Data-driven technology in event-based vision | |
CN116449947B (en) | Automobile cabin domain gesture recognition system and method based on TOF camera | |
CN111860361A (en) | Green channel cargo scanning image entrainment automatic identifier and identification method | |
WO2023043001A1 (en) | Attention map transferring method and device for enhancement of face recognition performance of low-resolution image | |
CN112926277B (en) | Design method of miniaturized focal plane array test data acquisition and display system | |
CN113920165A (en) | Robot pose estimation method and system based on multi-sensor feature fusion | |
CN112689121A (en) | Motion tracking system based on FPGA | |
CN116348905A (en) | Electronic apparatus and control method thereof | |
CN112947304A (en) | Intelligent camera multi-core heterogeneous on-chip integration system and visual control method | |
CN113962842B (en) | Dynamic non-polar despinning system and method based on high-level synthesis of large-scale integrated circuit | |
CN114760414B (en) | Image acquisition and processing system for CMV4000 camera | |
CN113132554A (en) | Real-time image acquisition and processing system based on FPGA | |
CN115167666A (en) | But interactive AR intelligence helmet device of gesture |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220603 |
|
RJ01 | Rejection of invention patent application after publication |