Space-time image texture angle detection and discrimination method and system

Abstract

The invention discloses a space-time image texture angle detection and discrimination method and system, and belongs to the field of visual flow measurement instruments. According to the design, a Zynq series processor with an ARM + FPGA architecture is used as a core, and a system module comprises a PS host control end, a PL end texture enhancement module, a two-dimensional Fourier transform module, a frequency spectrum logarithm transform module, a Gaussian filter module and an image segmentation and texture angle detection and judgment module. Wherein the texture enhancement module inhibits noise interference and enhances the continuity of effective textures; the two-dimensional Fourier transform module converts the information of the spatial domain image to a frequency domain; the frequency spectrum logarithmic transformation module is used for carrying out logarithmic magnitude spectrum calculation on the transformed spectrogram so as to improve the visibility of a low-amplitude component; and the image segmentation and texture angle detection and discrimination module carries out angle detection and discrimination after effective and invalid signals are separated, and finally outputs an angle detection result. The method can be applied to a river surface flow velocity measurement system based on a space-time image velocity measurement method, and the adaptability and efficiency of flow measurement in complex scenes can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of visual flow measuring instrument, in particular to a space-time image texture angle detection and discrimination method and system. TECHNICAL BACKGROUND

[0002] The accurate measurement of river flow velocity and flow is crucial for hydrological research, which is related to the prevention of flood disasters, the effective management of water resources, and the rational construction of water projects and other aspects. With the development of technology, image flow measurement technology has been widely used in river surface flow measurement due to its non-contact full-field characteristics, but it still has the problem of insufficient anti-interference ability.

[0003] The space-time image velocity measurement technology is also a method for measuring the surface flow velocity of water based on video images. This method analyzes the surface images of the river to be measured in a continuous time sequence. The information parallel to the water edge line on the water surface to be measured is extracted, and the sequence time and image space are combined. If there are obvious ripples or tracers on the water surface to be measured, there will be obvious texture characteristics on the space-time image. The texture direction reflects the water movement vector at the position of the velocity measurement line, so as to calculate the flow velocity value. Therefore, how to accurately extract the direction of the texture in the image preprocessing stage is extremely important for improving the accuracy of river flow velocity measurement.

[0004] In complex conditions such as strong environmental noise, glare, standing wave, etc., or in turbulent flow scenes with flow velocity pulsation, the texture is disturbed, causing the spectrum to have side lobes or the effective spectrum line to be broadened, resulting in false detection. A discrimination mechanism needs to be introduced based on texture angle detection to reduce the interference of invalid angles and improve robustness. In addition, image traversal is a highly repetitive operation, and the characteristics of CPU serial execution result in low efficiency in image processing, and real-time performance is insufficient in flow measurement applications.

[0005] To this end, the present application designs a space-time image texture angle detection and discrimination method and system, which uses the Zynq series processor with ARM+FPGA architecture as the core to realize hardware acceleration. FPGA is a programmable and customizable chip with parallel processing capability. The Zynq chip architecture is flexible, has low running power consumption, and has reconfigurability and portability. Using the Zynq system can effectively solve the problems of low computing efficiency and poor real-time performance of traditional processors, and can enhance the efficiency and robustness of flow measurement to adapt to various complex field scenes. It has important theoretical significance and application value for improving the accuracy and real-time performance of river surface flow velocity measurement. SUMMARY

[0006] Invention purposes: The application provides a spatiotemporal image texture angle detection and discrimination method and system. The effective and ineffective signals in the frequency spectrum are separated through image segmentation, the local features of the segmented signals are used for statistical discrimination while detecting the texture angle, so as to reduce the noise interference of the false angle. The algorithm is deployed on Zynq to realize hardware acceleration, meet the effectiveness and reliability requirements of the texture angle detection, and reduce the resource consumption and speed up the operation efficiency.

[0007] Technical scheme: The hardware system of the spatiotemporal image texture angle detection and discrimination method comprises the following steps:

[0008] Step 1: the PS host control end reads out the 8-bit gray spatiotemporal image stored in the SD card and stores it into the DDR4, and after the reading is completed, the image signal processing part of the PL end processes the input image according to the sequence of the texture enhancement module, the two-dimensional Fourier transform module, the spectrum logarithmic transform module, the Gaussian filter module, the image segmentation and texture angle detection and discrimination module;

[0009] Step 2: the texture enhancement module uses a 5*5 directional filter to enhance the inclined texture representing the motion of the flow tracer in the image;

[0010] Step 3: the two-dimensional Fourier transform module converts the information of the spatial domain image to the frequency domain;

[0011] Step 4: the spectrum logarithmic transform module reduces the dynamic range of the spectrum through amplitude operation and logarithmic operation and improves the visibility of the low-amplitude component;

[0012] Step 5: the Gaussian filter module enhances the low-frequency effective component of the spectrum after logarithmic transformation, suppresses the high-frequency noise, enhances the effect of subsequent image segmentation, and further improves the accuracy and effectiveness of detection and discrimination;

[0013] Step 6: the image segmentation and texture angle detection and discrimination module detects and discriminates the texture angle, filters out the false angle, and detects the main direction angle;

[0014] Further, in step 1, the PS end reads the image data from the SD card and stores it into the DDR4 using the FATFS file system, and after the reading is completed, the DMA transmission is started.

[0015] Further, in step 2, the following 5*5 size directional filter template is used: if F represents the original image, G represents the gray image, and H is the defined operator template; the calculation formula is as follows:

[0016]

[0017] In the column direction of the operator template, the third row is all 0, so when performing the matrix multiplication operation on the image, the third row of data can be set to 0; for the pixel point multiplication operation in the matrix, a left shift operation of one bit is performed on the pixel value of the point to save the calculation resources and accelerate the calculation; the first and second rows of operators have negative signs, and the negative signs are not considered first when calculating, and the four column product values G1, G2, G4 and G5 are directly calculated, and G n represents the negative number part, and G4 and G5 are added to obtain G p represent the positive number part. The calculation formula is as follows:

[0018] G1 = f 11 +f 12 <<1+f 13 ×3+f 14 <<1+f 15

[0019] G2 = f 21 +f 22 <<1+f 23 ×3+f 24 <<1+f 25

[0020] G3 = 0

[0021] G4 = f 41 +f 42 <<1+f 43 ×3+f 44 <<1+f 45

[0022] G5 = f 51 +f 52 <<1+f 53 ×3+f 54 <<1+f 55

[0023] G n = G1 + G2

[0024] G p = G4 + G5

[0025] By comparing the size of G n and G p , the absolute value of the difference is calculated to complete the calculation of G, and the texture enhanced image data is obtained; the calculation formula is as follows:

[0026]

[0027] Further, in step 3, the two-dimensional FFT transform of the image is realized by multiplexing the FFT IP core and storing the intermediate result matrix using the Ultra RAM resource, which saves the Block RAM and DSP resources and efficiently converts the spatial image information to the frequency domain. For an image f(x, y) with size M x N, the two-dimensional Fourier transform formula in the rectangular coordinate system is:

[0028]

[0029] Further, in step 4, the high gray value part is compressed and the image dark details are enhanced by using the characteristics of the logarithmic curve, i.e., the slope is larger in the low pixel area and the slope is smaller in the high pixel area. The calculation process is realized by the lookup table method, which calculates the logarithmic relationship between the input and output in advance and writes it into the LUT resource of the FPGA, without real-time calculation;

[0030] Further, in step 5, the main low-frequency part energy of the spectrum is preserved and the high-frequency noise energy is filtered out by using the Gaussian filter window function. The calculation process is realized by the lookup table method, which calculates the multiple relationship between the address and the output in advance and writes it into the LUT resource of the FPGA, without real-time calculation;

[0031] Further, in step 6, the image segmentation is completed by retaining the high brightness part of the image after the spectrum logarithmic transformation through threshold processing, achieving the purpose of separating the effective and ineffective information. Let s(x, y) represent the pixel value after threshold processing, T be the threshold, and P be the spectrum retention pixel ratio. Then, the texture angle detection and discrimination are performed through angle traversal and statistics. The calculation formula is as follows:

[0032]

[0033] The FPGA first performs statistics on the gray value of each pixel in the image, then calculates the threshold and starts the threshold processing, and the result is stored in 8 Block RAM resources for subsequent parallel calculation.

[0034] Further, in step 6, let θ be the angle between the search line and the y-axis, and Δ be the threshold. The energy accumulation operation is realized by using the rectangular search line as follows:

[0035]

[0036] This step uses the parallelism of FPGA to realize the calculation of the image stored in 8 Block RAMs at the same time, which accelerates the efficiency of image processing and improves the real-time performance.

[0037] Further, in step 6, a statistical window with size Q is set in the low-frequency region of the spectrum along the angle to be measured for texture angle discrimination. Let Qxm , Q ym are the maximum x and y values in the statistical window, respectively. In the angle traversal process, texture angle detection and discrimination are performed simultaneously. For each angle, the number of highlighted pixels V in the statistical window is counted, and if V is less than a preset number comparison value C, it is immediately determined that the angle is invalid and stepped to the next angle, otherwise the energy accumulation operation of the current angle is continued:

[0038] the number of highlighted pixels V in the statistical window is counted, and if V is less than a preset number comparison value C, it is immediately determined that the angle is invalid and stepped to the next angle, otherwise the energy accumulation operation of the current angle is continued:

[0039]

[0040] This step uses the parallelism of FPGA to realize the simultaneous execution of texture angle detection and discrimination, uses the characteristics of the hardware system to speed up the efficiency of image processing, and improves the real-time performance.

[0041] Advantages: Compared with the prior art, the present application has the following remarkable advantages: (1) a texture angle detection and discrimination fusion method is adopted, the effective and invalid signals in the frequency spectrum are separated through image segmentation, the local features of the segmented signals are used for statistical discrimination during texture angle detection, so as to reduce the noise interference of false angles; (2) the noise interference is reduced and the detection accuracy is improved through intermediate steps such as texture enhancement module, two-dimensional Fourier transform module, frequency spectrum logarithmic transformation module and Gaussian filter module; (3) the powerful computing power and control ability of the PS end of the development board are combined with the high parallelism and pipeline operation logic resources of the PL end, the operation speed is much higher than that of using 12th Gen Intel(R)Core(TM)i7-12700H CPU for calculation, the texture angle detection time of a space-time image is shortened from several seconds to tens of milliseconds, and the high real-time requirement is ensured.

[0042] The present application realizes a space-time image texture angle detection and discrimination method and system based on the Zynq series processor with ARM+FPGA architecture as the core, compares and evaluates the calculation results, speed and resource consumption, optimizes the data transmission and calculation time, shortens the texture angle detection time of a space-time image from several seconds to tens of milliseconds, greatly improves the accuracy and execution efficiency of the algorithm, and provides guarantee for the speed and stability of the real-time image processing system. Ensure that all space-time images with effective texture can be detected for effective texture direction. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is the system block diagram of the texture angle detection and discrimination system of the present application;

[0044] Figure 2 is the RTL view of the present application;

[0045] Figure 3 is the original space-time image;

[0046] Figure 4 is a structure diagram of a DMA IP core;

[0047] Figure 5 is an effect diagram of image enhancement processing of the present application;

[0048] Figure 6 is an effect diagram of image two-dimensional Fourier transform and frequency spectrum logarithmic transform of the present application;

[0049] Figure 7 is an effect diagram of image Gaussian filtering of the present application;

[0050] Figure 8 is a relationship between image coordinates and ROM table addresses of the present application; DETAILED DESCRIPTION

[0051] The space-time image texture enhancement method of the present application is further described below in combination with the accompanying drawings.

[0052] As shown in Figure 1 , it is a system block diagram of the whole application, including a PS host control end, a PL end texture enhancement module, a two-dimensional Fourier transform module, a frequency spectrum logarithmic transform module, a Gaussian filtering module, an image segmentation and texture angle detection and discrimination module; the texture enhancement module suppresses noise interference and enhances the continuity of effective texture; the two-dimensional Fourier transform module converts the information of the spatial domain image to the frequency domain; the frequency spectrum logarithmic transform module calculates the logarithmic amplitude spectrum of the transformed frequency spectrum graph to improve the visibility of low amplitude components; the Gaussian filtering module retains the main low frequency part energy and filters out high frequency noise; the image segmentation and texture angle detection and discrimination module separates the effective and ineffective signals, detects and discriminates the angle, and finally outputs the angle detection result. The Block Design RTL view of the present application is shown in Figure 2 . The AXI Direct Memory Acess module is a DMA module, which accepts the control information of the PS end and sends an interrupt signal to the PS end, and is responsible for the high-speed transmission of data between the DDR4 and the PL end; the Processor System Reset module is a system reset module, which accepts the reset button information and sends a reset signal to all system modules; the Concat module converts two interrupt signals into one signal, thereby integrating the two interrupt events of carrying data from the DDR4 to the PL end and carrying data from the PL end to the DDR4 into one interrupt source; the AXI SmartConnect and AXI Interconnect are AXI on-chip bus protocol modules, which are responsible for the bidirectional data transmission between the PS end and the PL end; the IAD_v9_16 is the address of the present application to the AXI Interconnect module, which is responsible for the address translation between the PS end and the PL end; the AXI DMA IP core is the core of the AXI Direct Memory Acess module, which is responsible for the data transmission between the DDR4 and the PL end. Figure 1The PL end data processing module shown in the PL end function design and packaged into the ip core, input DMA from the DDR4 transmission of the measured space-time image, output detection texture angle results to the DMA and stored in the DDR4; Zynq UltraScale + MPSoC is a PS end module, provides various data lines and control line interface with PS end interaction.

[0053] Next, the implementation of the sub-modules in the design is described in detail:

[0054] Step 1: the original space-time image as shown in Figure 3 First stored in the SD card, the PS control end reads out the image through the FATFS file system and stores it in the DDR4, and after all the reading is completed, it is sent to the PL end through the DMA and AXI bus for subsequent processing.

[0055] The PL end provides a soft core AXIDMAip defined by Xilinx officials, and its structure diagram is shown in Figure 4 This implementation is more friendly to hardware developers and can achieve higher transmission rate. AXIDMA transmits data through AXI-HP interface, uses four HP interfaces in ZYNQ platform, and provides maximum total bandwidth. Each HP interface contains control and data FIFO, which provides buffer for burst transmission of large data, making HP interface an ideal high-speed data transmission channel. In addition, AXIDMA can be configured through AXI4-Lite protocol and AXI4-GP interface, and can obtain state information in the transmission process through interrupt mechanism.

[0056] Step 2: image enhancement processing effect as shown in Figure 5 The texture enhancement module uses a 5x5 directional filter to enhance the inclined texture representing the motion of the tracer in the image. If F represents the original image, G represents the gray image, and H is the defined operator template; its calculation formula is as follows:

[0057]

[0058] In the column direction of the operator template, the third row is all 0, so when performing matrix multiplication operation on the image, the third row data can be set to 0; for the pixel point multiplication operation in the matrix, the pixel value of the point is uniformly left shifted one bit to achieve the effect of saving calculation resources and accelerating calculation; the first and second rows of operators have negative signs, but the negative signs are not considered first when calculating, and the four column product values G1, G2, G4, G5 are calculated directly, and G n represents the negative part, and G4 and G5 are added to obtain G p represent the positive part:

[0059] G1 = f 11 + f 12 <<1 + f 13 ×3 + f 14 <<1 + f 15

[0060] G2 = f 21 + f 22 <<1 + f 23 ×3 + f 24 <<1 + f 25

[0061] G3 = 0

[0062] G4 = f 41 + f 42 <<1 + f 43 ×3 + f 44 <<1 + f 45

[0063] G5 = f 51 + f 52 <<1 + f 53 ×3 + f 54 <<1 + f 55

[0064] G n = G1 + G2

[0065] G p = G4 + G5

[0066] By comparing the size of G n and G p , the absolute value of the difference is calculated, the calculation of G is completed, and the texture enhanced image data is obtained; the calculation formula is as follows:

[0067]

[0068] Step 3: The two-dimensional Fourier transform module converts the information of the spatial domain image to the frequency domain.

[0069] By multiplexing the FFT ip core and using the Ultra RAM resource to store the intermediate result matrix, the FPGA implementation of the two-dimensional FFT transform of the image is realized, and the Block RAM and DSP resources are saved, and the information of the spatial domain image is efficiently converted to the frequency domain. For an image with a size of 128*128, the two-dimensional Fourier transform formula in the rectangular coordinate system is:

[0070]

[0071] Step 4: The image two-dimensional Fourier transform and frequency spectrum logarithmic transform effect is as follows: Figure 6The log curve has a larger slope in the area with lower pixel value and a smaller slope in the area with higher pixel value. The high gray value part is compressed to enhance the details of the dark part of the image. The calculation process is realized by table lookup method. The log relationship between the input and output is calculated in advance and written in the LUT resource of the FPGA, so that real-time calculation is not needed.

[0072] Step 5: The image Gaussian filtering effect is shown in Figure 7 The Gaussian filtering module enhances the low-frequency effective components of the spectrum after log transformation, suppresses the high-frequency noise, enhances the effect of subsequent image segmentation, and thus further improves the accuracy and effectiveness of detection and discrimination.

[0073] The Gaussian window filtering module is used for filtering the image after log transformation. The module includes a ROM table module and a pixel processing module. The ROM table is actually realized by a more flexible LUT, which contains the coefficients of the Gaussian window of each pixel coordinate point of the image. For a two-dimensional Gaussian window, its expression is as follows: where r is the radius of the Gaussian window, (x, y) is the pixel coordinate, (c x , c y ) is the image center, and σ is the standard deviation of the Gaussian function, which is used to control the width and attenuation degree of the Gaussian window. The coefficient parameter needs to be quantized to a 16-bit unsigned fixed-point integer through linear mapping for FPGA processing. The module realizes single-cycle delay coefficient reading by responding to the row and column count signals of the pixel processing module. The pixel processing module reads the current pixel coordinate and controls the generation of the address for reading ROM data. The coefficient of the current pixel coordinate point is read from the ROM and multiplied with the current coordinate pixel value. In order to match the output image of the previous module, the Gaussian window is quadrant shifted, i.e., the first quadrant and the third quadrant of the two-dimensional window function are exchanged, and the second quadrant and the fourth quadrant are exchanged, so that the center of the Gaussian window is shifted to the low-frequency position of the four corners of the image, and a half-window structure is generated to reduce the storage space requirement by 50%. The relationship between the image coordinate and the ROM table address is shown in Figure 8 .

[0074] Step 6: The image segmentation and texture angle detection and discrimination module detects and discriminates the texture angle, filters out the wrong angle, and detects the main direction angle. Through threshold processing, the part of the pixels with high brightness after the spectrum log transformation is reserved to complete the image segmentation, so as to separate the effective and ineffective information. s(x, y) represents the pixel value after threshold processing, T is the threshold, and P is the spectrum reserved pixel ratio. Then, the texture angle is detected and discriminated through angle traversal and statistics. The calculation formula is as follows:

[0075]

[0076] The FPGA first counts the gray value of each pixel in the image, then calculates the threshold value and starts the threshold processing, and the result is stored in 8 block RAM resources for subsequent parallel calculation.

[0077] Let θ be the angle between the search line and the y-axis, and Δ be the threshold value. The energy accumulation operation is implemented by using the rectangular search line as follows:

[0078]

[0079] This step uses the parallelism of the FPGA to simultaneously calculate the images stored in the 8 block RAMs, uses the characteristics of the hardware system to speed up the image processing efficiency, and improves the real-time performance.

[0080] A statistical window with a size of Q is set along the angle direction in the low-frequency region of the spectrum for texture angle discrimination. Let Q xm , Q ym be the maximum x value and y value in the statistical window, respectively. During the angle traversal process, the texture angle detection and discrimination are simultaneously performed. For the current angle, the number of highlighted pixels V in the statistical window is counted. If V is less than the preset comparison value C, it is immediately determined that the angle is invalid and stepped to the next angle. Otherwise, the energy accumulation operation for the current angle is continued.

[0081]

[0082] This step uses the parallelism of the FPGA to simultaneously perform the texture angle detection and discrimination, uses the characteristics of the hardware system to speed up the image processing efficiency, and improves the real-time performance.

Claims

1. A method and system for spatiotemporal image texture corner detection discrimination, characterized by: With ARM+FPGA architecture Zynq series processor as the core, the system module includes PS host control end, PL end texture enhancement module, two-dimensional Fourier transform module, spectrum logarithmic transformation module, Gaussian filter module, image segmentation and texture angle detection and discrimination module. Among them, the texture enhancement module suppresses noise interference and enhances the continuity of effective texture; the two-dimensional Fourier transform module converts the information of the spatial image to the frequency domain; the spectrum logarithmic transformation module calculates the logarithmic amplitude spectrum of the transformed frequency spectrum to improve the visibility of the low amplitude component; the Gaussian filter module retains the main low frequency part energy and filters out the high frequency noise; the image segmentation and texture angle detection and discrimination module separates the effective and invalid signals, detects and discriminates the angle, and finally outputs the angle detection result; the steps of the hardware system implementation method are: Step 1, the PS host control end reads out the 8-bit gray space-time image stored in the SD card and stores it in the DDR4, and after all reading is completed, it is sent to the PL end image signal processing part according to the order of the texture enhancement module, two-dimensional Fourier transform module, spectrum logarithmic transformation module, Gaussian filter module, image segmentation and texture angle detection and discrimination module to process the input image; Step 2, the texture enhancement module uses a 5x5 directional filter to enhance the inclined texture representing the movement of the tracer in the image; Step 3, the two-dimensional Fourier transform module converts the information of the spatial image to the frequency domain; Step 4, the spectrum logarithmic transformation module reduces the dynamic range of the spectrum through amplitude operation and logarithmic operation, and improves the visibility of the low amplitude component; Step 5, the Gaussian filter module enhances the low frequency effective component of the spectrum after logarithmic transformation, suppresses high frequency noise, enhances the effect of subsequent image segmentation, and further improves the accuracy and effectiveness of detection and discrimination; Step 6, the image segmentation and texture angle detection and discrimination module detects and discriminates the texture angle, filters out the wrong angle, and detects the main direction angle.

2. The hardware system based on the spatio-temporal image texture angle detection and discrimination method according to claim 1, characterized in that: In step 1, the PS end uses FATFS file system to read image data from SD card and store it in DDR4, and after all reading is completed, it starts DMA transmission.

3. The hardware system based on the method for detecting and distinguishing texture angles of spatio-temporal images according to claim 1, wherein: In step 2, the following 5x5 size directional filter template is used: if F represents the original image, G indicates the gray image, and H is the defined operator template; the calculation formula is as follows: In the column direction of the operator template, the third row is all 0, so when the matrix multiplication operation is performed on the image, the third row of data can be set to 0; for the pixel point multiplication operation in the matrix, a left shift operation of one bit is uniformly performed on the pixel value of the point to achieve the effect of saving computing resources and accelerating calculation; the first and second rows of operators have negative signs, and the negative signs are not considered first when calculating, and four column product values G1, G2, G4 and G5 are directly calculated, and G n represents the negative number part, and G4 and G5 are added to obtain G p represent the positive number part. The calculation formula is as follows: G1 = f 11 + f 12 <<1 + f 13 x 3 + f 14 <<1 + f 15 G2 = f 21 + f 22 <<1 + f 23 x 3 + f 24 <<1 + f 25 G3＝0 G4 = f 41 + f 42 <<1 + f 43 x 3 + f 44 <<1 + f 45 G5 = f 51 + f 52 <<1 + f 53 x 3 + f 54 <<1 + f 55 G n = G1+ G2 G p = G4+ G5 By comparing the size of G n and G p , the absolute value of the difference is calculated, the calculation of G is completed, and the image data after texture enhancement is obtained; the calculation formula is as follows:

4. The hardware system based on the method for detecting and distinguishing texture angle of space-time image according to claim 1, wherein: In step 3, by reusing the FFT ip core and using the Ultra RAM resource to store the intermediate result matrix, the FPGA implementation of two-dimensional FFT transformation of the image is realized, and the Block RAM and DSP resources are saved, and the information of the spatial image is efficiently converted to the frequency domain. For an image f(x, y) with size MxN, the two-dimensional Fourier transform formula in the rectangular coordinate system is:

5. The hardware system based on the method for detecting and distinguishing texture angle of space-time image according to claim 1, wherein: In step 4, the logarithmic curve has the characteristics that the slope is large in the area with low pixel value and the slope is small in the area with high pixel value, so the high gray value part is compressed and the details of the dark part of the image are enhanced. The calculation process is realized by lookup table method, which calculates the logarithmic relationship between input and output in advance and writes it into the LUT resource of FPGA, without real-time calculation.

6. The hardware system based on the method for detecting and distinguishing texture angle of space-time image according to claim 1, wherein: In step 5, the energy of the low frequency part of the spectrum is reserved by a Gaussian filter window function, and the energy of the high frequency noise is filtered out. The calculation process is realized by table lookup method, and the relationship between the address and the output multiple is calculated in advance and written into the LUT resource of FPGA, without real-time calculation.

7. The hardware system based on the method for detecting and distinguishing texture angle of space-time image according to claim 1, wherein: In step 6, the image segmentation is completed by retaining the part of the pixels with high brightness in the image after the spectrum logarithmic transformation through threshold processing, so as to achieve the purpose of separating the effective and invalid information. S(x, y) represents the pixel value after threshold processing, T is the threshold, and P is the spectrum reserved pixel ratio. Then, the texture angle detection and discrimination are realized by angle traversal and statistics. The calculation formula is as follows: FPGA first calculates the threshold value by counting the gray value of each pixel in the image, and then starts the threshold processing. The result is stored in 8 block RAM resources for subsequent parallel calculation.

8. The hardware system based on the method for detecting and distinguishing texture angle of space-time image according to claim 1, wherein: In step 6, let θ be the angle between the search line and the y-axis, and Δ be the threshold. The energy accumulation operation is realized by rectangular search line using the following formula: In this step, the parallelism of FPGA is used to realize the simultaneous calculation of the images stored in 8 block RAMs. The characteristics of the hardware system are used to speed up the efficiency of image processing and improve the real-time performance.

9. The hardware system based on the method for detecting and distinguishing texture angle of space-time image according to claim 1, wherein: In step 6, a statistical window with size Q is set along the angle direction in the low frequency region of the spectrum for texture corner detection, where Q xm , Q ym are the maximum x and y values in the statistical window, respectively. During the angle iteration, the texture corner detection and discrimination are performed simultaneously. For the current angle, the number of highlighted pixels V in the statistical window is counted. If V is less than a preset comparison value C, the current angle is immediately determined to be invalid and the next angle is stepped to. Otherwise, the energy accumulation operation for the current angle is continued. In this step, the parallelism of FPGA is used to realize the simultaneous execution of texture angle detection and discrimination. The characteristics of the hardware system are used to speed up the efficiency of image processing and improve the real-time performance.

Images