Image processing method and system for adaptive optimization detection of floating ball position

Abstract

This invention belongs to the technical field of image data processing and relates to an image processing method and system for adaptive optimization detection of a float position. The method includes the following steps: generating a spatial kinematic tensor based on basic geometric center data; generating a hydrodynamic spatial distortion phase array by modifying the preset standard point matrix distribution coordinates using addressing scaling factors; superimposing and merging the hydrodynamic spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer; merging the distortion spatial feature data to generate a target adaptive filtering pixel matrix; calculating the anti-interference centroid target position based on the entity's vertical contour boundary; generating a defect pointer and triggering a hardware abnormal interrupt level to reset the visual acquisition parameters. This invention solves the problem of limited target detection accuracy and insufficient tracking stability caused by using a fixed image sampling and processing model under dynamic fluid disturbance environments.

Description

Technical Field

[0001] This invention belongs to the technical field of image data processing, and relates to an image processing method and system for adaptive optimization detection of float position. Background Technology

[0002] In fields such as industrial automation and environmental monitoring, real-time positioning and tracking of floating targets in liquids are often required using visual methods to infer liquid level height. The fluid environment is dynamic and uncertain; liquid surface fluctuations and the complex reflections and refractions of light on the water surface, coupled with the rapid and irregular motion of the buoy itself due to wave impacts, all contribute to interference with visual detection algorithms. These factors can cause drastic changes in the target's shape, brightness, and other apparent characteristics in the image, posing a challenge to stable and continuous tracking.

[0003] To address these challenges, the industry commonly employs machine vision-based solutions for target detection and tracking. For example, patent document CN113744325B discloses a liquid level detection device and method based on image recognition technology. Such solutions typically utilize image segmentation techniques to set fixed color or grayscale thresholds to separate the float from the background and calculate its geometric center.

[0004] The aforementioned existing technical solutions have certain limitations when dealing with complex fluid disturbances. Schemes based on fixed threshold segmentation are highly sensitive to changes in lighting and water surface reflections. When ambient light changes or bright areas appear on the water surface, incomplete segmentation or misidentification of reflective areas as targets can easily occur.

[0005] Therefore, the technical problem to be solved by this invention is how to overcome the problem of limited target detection accuracy and insufficient tracking stability caused by using a fixed image sampling and processing model in dynamic fluid disturbance environments. Summary of the Invention

[0006] In a first aspect, the present invention provides an image processing method for adaptive optimization detection of float position, comprising the following steps: S1. Receive the raw dynamic video data stream and write it into the video memory space to construct the raw pixel buffer area. Extract the basic geometric center data of three consecutive frames in the raw pixel buffer area and generate the spatial kinematic tensor based on the basic geometric center data. S2. Calculate the addressing scaling factor based on the spatial kinematic tensor, and use the addressing scaling factor to modify the preset standard lattice distribution coordinates to generate a fluid dynamic spatial distortion phase array. S3. Obtain the default center reference base address, and superimpose and merge the fluid dynamics spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer; S4. Extract distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and combine the distortion spatial feature data to generate the target adaptive filter pixel matrix; S5. Using a preset high-pass edge enhancement operator, perform a convolution operation on the target adaptive filtering pixel matrix to extract the vertical contour boundary of the entity, and calculate the position of the anti-interference centroid target based on the vertical contour boundary of the entity. S6. Calculate the rank index variable of the internal algebraic matrix of the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, generate a defect pointer and trigger a hardware abnormal interrupt level to reset the vision acquisition parameters.

[0007] A further aspect of the present invention involves extracting the fundamental geometric center data of three consecutive frames of images from the original pixel buffer, including the following steps: Activate baseline locking operation within the original pixel buffer and capture three consecutive adjacent frames from the startup sequence, where the startup sequence is the initial video sequence when the system starts. Perform grayscale thresholding with preset fixed parameters on three consecutive frames of images to generate a binary mask image; Extract the centroid coordinates of the connected domain of the floating sphere in the binary mask image as the basic geometric center data.

[0008] A further aspect of the present invention generates a spatial kinematic tensor based on fundamental geometric center data, comprising the following steps: Discrete difference equations are input for the basic geometric center data on the horizontal transverse dimension axis and the vertical gravity dimension axis, respectively; The second-order central difference equation is executed to obtain the horizontal acceleration component caused by wave impact and the vertical acceleration component caused by the float's own buoyancy. The horizontal acceleration components are combined with the vertical acceleration components to generate a spatial kinematic tensor that expresses the intensity of physical disturbances in the underlying environment.

[0009] A further aspect of the present invention involves calculating the addressing scaling factor based on the spatial kinematic tensor, including the following steps: Extract the horizontal and vertical acceleration components from the spatial kinematic tensor; After dividing the absolute value of the horizontal acceleration component by the system's preset reference acceleration constant and performing dimensionless processing, it is substituted into the preset first logarithmic distortion mapping function to perform forward conversion and output the horizontal addressing stretching ratio coefficient. After dividing the absolute value of the vertical acceleration component by the reference acceleration constant, the result is substituted into the second logarithmic distortion mapping function with an additional zero overflow prevention factor for reverse conversion to output the vertical addressing compression ratio coefficient. The horizontal addressing stretching ratio coefficient and the vertical addressing compression ratio coefficient together constitute the addressing ratio coefficient.

[0010] A further aspect of the present invention utilizes an addressing scaling factor to modify a preset standard lattice distribution coordinate to generate a hydrodynamic spatial distortion phase array, comprising the following steps: Based on the preset central grid radial geometric topology rules, the horizontal addressing stretching ratio coefficient and the vertical addressing compression ratio coefficient are jointly applied to modify the preset standard lattice distribution coordinates in a directional manner, thereby deriving the non-uniform sampling bias parameter of the specified step size neighborhood. All non-uniform sampling bias parameters are compiled to generate a hydrodynamic spatial distortion phase array for intervening in the underlying pre-stored logic.

[0011] A further aspect of this invention involves obtaining a default center reference base address, and then superimposing and merging the hydrodynamic spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer, comprising the following steps: Release the sampling instruction blocking state of the main controller inside the deformable kernel addressing convolutional engine, and extract a set of regularized third-order distributed two-dimensional grid coordinates from the internal read-only memory area to serve as the default center reference base address; The fluid dynamics spatial distortion phase array is written as a series of continuous physical control level encoded information into the address offset control register of the direct memory accessor; The underlying half-adder module and multiplexed operation circuit unit of the hardware logic gate are activated, and the gate-level physical superposition and merging mechanism is performed on the data presented in the default center reference base address and address offset control register to generate deformed address offset read and write pointers.

[0012] A further aspect of this invention involves extracting distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and then concatenating the distortion spatial feature data to generate a target adaptive filtering pixel matrix, comprising the following steps: Capture the sequence of non-contiguous memory jump access instructions consisting of deformable address offset read and write pointers, and submit a data retrieval request to the memory controller front queue pool to start the non-standard memory block retrieval thread; By using a sequence of stepping non-contiguous memory jump access instructions, data page reads are performed within the original pixel buffer area that are forcibly expanded based on the physical interval size, crossing invalid environmental reflective interference pixel segments to pick up distorted spatial feature data at isolated data nodes. The distortion spatial feature data is guided into the first buffer delay line channel of the deformation kernel addressing convolution engine to perform temporal re-densification and realignment operations, and the data after eliminating coordinate gaps are combined to form the target adaptive filter pixel matrix.

[0013] A further aspect of the present invention involves calculating the anti-interference centroid target position based on the vertical contour boundary of the entity, including the following steps: Apply a second-order moment geometric centroid estimation operation to the closed graphic range of the entity's vertical contour boundary in the current frame to analyze the analytical coordinate values ​​and obtain the anti-interference centroid target position; Feeding back the anti-interference centroid target position to the end of the basic historical coordinate sequence queue provides a dependency source for subsequent secondary recursive periodic operations.

[0014] A further aspect of this invention involves generating a defect pointer and triggering a hardware exception interrupt to reset the visual acquisition parameters, including the following steps: Enable the coprocessor to perform real-time calculation of the rank index variable of the internal algebraic matrix of multiple linearly independent data rows in the hydrodynamic spatial distortion phase array; When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, it is determined that the address index has lost its directionality, and the hardware bus is forcibly locked to generate a homogeneous defect pointer in which all spatial data bits point to the same read-only reserved physical memory segment. Based on the defect pointer, a division-to-zero overflow exception is triggered in the division operation of the addressing convolution engine, and a hardware exception interrupt level is output to the reset pin of the image acquisition device to reset the visual acquisition parameters.

[0015] Secondly, the present invention provides an image processing system for adaptive optimization detection of float position, comprising the following modules: The kinematics analysis module receives the raw dynamic video data stream and writes it into the video memory space to construct the raw pixel buffer. It extracts the basic geometric center data of three consecutive frames in the raw pixel buffer and generates a spatial kinematic tensor based on the basic geometric center data. The distortion array generation module calculates the addressing scaling factor based on the spatial kinematic tensor, and uses the addressing scaling factor to modify the preset standard lattice distribution coordinates to generate a fluid dynamic spatial distortion phase array. The deformation pointer generation module obtains the default center reference base address and superimposes and merges the fluid dynamics spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer. The adaptive sampling module extracts distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and then combines the distortion spatial feature data to generate the target adaptive filtered pixel matrix. The position update and feedback module uses a preset high-pass edge enhancement operator to perform a convolution operation on the target adaptive filtering pixel matrix to extract the vertical contour boundary of the entity, and calculates the anti-interference centroid target position based on the vertical contour boundary of the entity. The system stability monitoring module calculates the rank index variable of the internal algebraic matrix of the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, a defect pointer is generated and a hardware abnormal interrupt level is triggered to reset the vision acquisition parameters.

[0016] In summary, the present invention has the following beneficial technical effects: 1. To address the interference of wave disturbances and water surface reflections on target features in dynamic hydrological environments, a hydrodynamic spatial distortion phase array is generated by calculating spatial kinematic tensors. This transforms the macroscopic physical motion of the target into an intervention in the microscopic image sampling mode. This allows the spatial distribution of pixel sampling points to dynamically adapt to image deformations caused by waves, guiding the sampling process to avoid noise areas such as high-frequency water surface reflections during the physical sampling stage. An adaptive filtering mechanism is implemented at the data input source, providing a higher signal-to-noise ratio set of original pixels for subsequent processing and improving the stability of target feature extraction.

[0017] 2. By utilizing deformable address offset read / write pointers to perform step-by-step, non-contiguous memory jump accesses, and temporally reorganizing the acquired irregularly distributed data, a standard target adaptive filtering pixel matrix is ​​synthesized. This mechanism completes the reading and reassembly of non-contiguous data at the front end of the hardware architecture, enabling the system to achieve image content perception without modifying the backend standard arithmetic logic unit. It reduces the impact of large-area interference regions on detection without introducing complex algorithm compensation or increasing additional computational load.

[0018] 3. By monitoring the validity of the address index through real-time calculation of the matrix rank of the distorted phase array, when target loss causes kinematic tensor divergence and triggers matrix rank convergence collapse, the system can promptly identify logical calculation anomalies and throw a hardware physical emergency stop level to perform a low-level reset of the image acquisition device. This forms a fast response loop independent of the main software processing flow, enabling the system to have autonomous detection and rapid recovery capabilities under extreme conditions, which is beneficial to ensuring the long-term reliable operation of the system in unattended scenarios. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0020] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0021] Figure 2 A schematic diagram of the framework in the embodiments of this application is disclosed. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1 - Figure 2 A preferred description of the present invention is provided below.

[0023] See attached document Figure 1 This invention proposes an image processing method for adaptive optimization detection of float position, comprising the following steps: S1. Receive the raw dynamic video data stream and write it into the video memory space to construct the raw pixel buffer area. Extract the basic geometric center data of three consecutive frames in the raw pixel buffer area and generate the spatial kinematic tensor based on the basic geometric center data. S2. Calculate the addressing scaling factor based on the spatial kinematic tensor, and use the addressing scaling factor to modify the preset standard lattice distribution coordinates to generate a fluid dynamic spatial distortion phase array. S3. Obtain the default center reference base address, and superimpose and merge the fluid dynamics spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer; S4. Extract distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and combine the distortion spatial feature data to generate the target adaptive filter pixel matrix; S5. Using a preset high-pass edge enhancement operator, perform a convolution operation on the target adaptive filtering pixel matrix to extract the vertical contour boundary of the entity, and calculate the position of the anti-interference centroid target based on the vertical contour boundary of the entity. S6. Calculate the rank index variable of the internal algebraic matrix of the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, generate a defect pointer and trigger a hardware abnormal interrupt level to reset the vision acquisition parameters.

[0024] In one embodiment of the present invention, step S1 includes the following steps: Discrete difference equations are input into the basic geometric center data on the horizontal and vertical gravity axes, respectively; the second-order central difference equation is executed to obtain the horizontal acceleration component caused by wave impact and the vertical acceleration component caused by the float's own buoyancy; the horizontal and vertical acceleration components are combined to generate a spatial kinematic tensor to express the intensity of physical disturbance in the underlying environment.

[0025] Specifically, this step begins with the central processing unit initiating a communication session with the environmental sensing image acquisition device via an industry-standard gigabit Ethernet interface. The environmental sensing image acquisition device incorporates a megapixel-level complementary metal-oxide-semiconductor (CMOS) image sensor. The central processing unit then instructs the environmental sensing image acquisition device to acquire images at a rate of [data rate] per second. The frame rate continuously captures digital images of the target drifting scene. Frame rate It is set according to the typical frequency range of the measured liquid surface fluctuations. To satisfy the Nyquist sampling theorem, the typical setting value is 60 frames per second. The basis for this setting is that the typical dominant frequency of gravity waves on the surface of natural water bodies is usually between 0.5Hz and 5Hz. According to the Nyquist sampling theorem, the sampling frequency must be greater than twice the highest frequency of the signal. Setting it to 60 frames per second can leave sufficient margin to effectively capture high-frequency instantaneous liquid splashes and rapid shaking of the float, and prevent waveform aliasing.

[0026] Each frame of image data is encapsulated into a data packet conforming to the Giant Eye protocol and sent through a gigabit Ethernet interface. After receiving the data packet, the central processing unit parses out the raw pixel data and calls the memory access controller to write the raw pixel data along with its hardware timestamp into a pre-allocated system video memory space, thereby constructing a raw pixel buffer containing continuous video frames.

[0027] The central processing unit performs a baseline locking operation within the raw pixel buffer, which involves locating and extracting the three consecutive frames that arrive at the buffer first in the start sequence, sorted by timestamp, and denoted as images. ,image With images For each of these three image frames, the central processing unit first performs a color space conversion, converting it from RGB format to an 8-bit grayscale image, and then applies a fixed grayscale threshold. Perform binarization to generate a binary mask image containing only two pixel values, 0 and 255.

[0028] Gray threshold This is a static value pre-calibrated based on the lighting conditions of the experimental scenario and the color contrast between the float and the liquid surface. Assuming the application scenario is an indoor stable light source environment, the float is a highly reflective white sphere, and the liquid surface is a dark, opaque liquid, based on statistical analysis of 256 levels of grayscale images, this grayscale threshold is set in the range of 180 to 220. The reason is that the surface of the float is usually coated with a bright paint with high reflectivity, and under stable lighting, the grayscale values ​​of the pixels in its reflective area are highly concentrated in this bright range; while the grayscale of the dark liquid surface and background is usually below 120. This range setting can not only filter the diffuse reflection noise of water surface ripples, but also ensure the extraction of the core bright area of ​​the float with a complete shape.

[0029] On this binary mask image, the central processing unit runs a connected component labeling algorithm based on two traversals to identify and separate the largest area connected component representing the sphere's outline. For the identified largest area connected component, its fundamental geometric center data, i.e., the centroid coordinates, are determined by calculating the ratio of its first-order spatial moment to its zeroth-order spatial moment. Specifically, the centroid coordinates are obtained by analyzing the binarized connected domain of the floating sphere. The calculation formula for the image moments is as follows:

[0030] in, The zeroth moment represents the area of ​​the connected region. and The first moment is calculated as follows:

[0031] Here, For pixel coordinates Pixel value at; Represents the horizontal column coordinates of a pixel in the image, such as the X-axis; This represents the vertical row coordinate of a pixel in the image, such as the Y-axis; for binary images, its value is 1 or 0. This operation is applied consecutively to three frames of images, thereby constructing a sequence of three two-dimensional coordinate points as a time base data stream reflecting the initial physical state of the system. The time base data stream consists of three time points. , , The corresponding three centroid coordinates , , Composition. Among them, This represents the time interval between adjacent frames, and its value is the frame rate. The reciprocal of.

[0032] The central processing unit accesses the time reference data stream and inputs discrete second-order central difference equations for its coordinate components along the horizontal horizontal axis and the vertical gravity axis. Through this approximation operation, the horizontal acceleration component characterizing the lateral impact of waves is obtained. and the vertical acceleration component characterizing the float's own vertical buoyancy. The two acceleration components mentioned above are obtained by approximating the centroid coordinates in the time base data stream using a second-order central difference equation. The calculation formula is as follows:

[0033] In the above acceleration calculation formula, the numerator term has the dimension of length units, such as pixels, and the denominator term... The dimension of is the square of time, for example, the square of a second. Therefore, the calculated and The unit of measurement is length divided by the square of time, such as pixels per second².

[0034] The central processing unit combines the horizontal and vertical acceleration components into a two-dimensional vector, thereby generating a spatial kinematic tensor capable of quantifying the intensity of physical disturbances in the bottom water. The generated spatial kinematic tensor... It is constructed as a two-dimensional column vector:

[0035] Among them, subscript Represents the current operation position. A discrete time step or frame sequence index; in the subsequent inline text representation of this article, it will be superscripted. The shape of The expression, its superscript Both represent the transpose operation of a matrix or vector. Spatial kinematic tensor. It is a first-order tensor, i.e., a vector. The motion tensor represents the state of motion at time t. The magnitude of the projection of the net external force acting on the center of mass of the buoy onto two orthogonal coordinate axes indirectly reflects the instantaneous state of the liquid surface disturbance.

[0036] For example, assuming the environmental perception image acquisition device has a frame rate of 50 frames per second, then the time interval between adjacent frames... The timeframe is 0.02s. In the startup sequence, the system acquired and processed three consecutive frames of images. Through grayscale thresholding and centroid calculation, a time-based data stream reflecting the initial physical state of the system was obtained. This stream contains three fundamental geometric center data points, respectively... Pixels Pixels Pixels. The system then uses these coordinates to calculate intermediate times. The acceleration component. Horizontal acceleration component. The calculation process involves substituting the horizontal coordinate components 120.0, 121.0, and 122.5 into the second-order central difference equation, resulting in... Pixels per second².

[0037] Similarly, the vertical acceleration component The calculation process is as follows: substituting the vertical coordinate components 250.0, 252.5, and 256.0 into the equation, we obtain... Pixels per second². The system combines the two calculated acceleration components to generate a spatial kinematic tensor that represents the intensity of physical disturbances in the underlying environment. Its specific value is .

[0038] In one embodiment of the present invention, step S2 includes the following steps: Based on the preset central grid radial geometric topology rules, the horizontal addressing stretching ratio coefficient and the vertical addressing compression ratio coefficient are jointly applied to modify the preset standard point distribution coordinates in a directional manner, thereby deriving non-uniform sampling bias parameters for a specified step size neighborhood; all non-uniform sampling bias parameters are compiled to generate a fluid dynamics spatial distortion phase array for intervening in the underlying pre-stored logic.

[0039] Specifically, the central processing unit (CPU) accesses and locks the spatial kinematics tensor stored in the high-speed computing stack by calling kernel-level memory scheduling interface functions. Through pointer dereferencing, the CPU retrieves the horizontal acceleration component representing the magnitude of the lateral impact disturbance of the liquid surface from the spatial kinematics tensor. and the vertical acceleration component characterizing the magnitude of the sinking and lifting displacement of the measured object. They were separated and loaded into two independent floating-point registers.

[0040] The central processing unit substitutes the absolute value of the horizontal acceleration component into the first preset logarithmic distortion mapping function stored in the hardware lookup table for forward conversion. The result of this calculation is designated as the horizontal addressing stretching coefficient. Simultaneously, the absolute value of the vertical acceleration component is substituted into the second preset logarithmic distortion mapping function for inverse conversion to generate the vertical addressing compression ratio coefficient. Its calculation follows the nonlinear equivalent mapping algorithm:

[0041]

[0042] in, | and | | These are the absolute values ​​of the horizontal and vertical acceleration components, respectively. The reference acceleration constant for system calibration has the same dimensions as the acceleration components, for example, set to pixels / second². The dimensions are canceled out by division to ensure that the independent variable of the logarithmic function is a dimensionless pure number. A very small positive number set to prevent overflow when dividing by zero, for example, set to 0. .

[0043] The logarithmic base in the first and second preset logarithmic distortion mapping functions mentioned above and and proportionality constant and It was derived from statistical calibration based on a large amount of offline simulation experimental data. Assuming the application scenario is an open water body such as a reservoir, where the wave disturbance acceleration range is large, in order to ensure that the mapping function has good sensitivity and nonlinear characteristics within its effective working range, and It is usually set to the base of the common logarithm of 10. (Proportionality constant) Set to a value less than 1, the scaling constant. Setting it to 0.2-0.8 is beneficial because horizontal wave impacts typically cause frequent high-frequency oscillations. This range helps to attenuate and suppress the horizontal stretching amplitude, preventing the sampling grid from exceeding the effective field of view due to excessive stretching. The range is set to 1.5-3.0. Since the vertical buoyancy of the float is the main physical quantity observed by the system, this range can appropriately amplify the sampling resolution weight in the direction of gravity and enhance the sensitivity to changes in liquid level.

[0044] After obtaining the two scaling factors, the central processing unit generates a grid around the central point according to the radial geometric topology rules of the central grid. of A set of standard lattice distribution coordinates. Among them, the grid size is defined in the radial geometric topology rules of the central grid. It is an odd number to ensure the existence of a unique center point, typically 3 or 5. This size is set so that it precisely matches the standard register block size of the underlying hardware multiply-accumulate array, enabling data throughput within a single clock cycle. If the size is too large, the overhead of the hardware multiplier will increase exponentially; if it is too small, sufficient edge gradient features cannot be extracted. or The neighborhood of.

[0045] For each standard coordinate point in this set The central processing unit performs a directional modification operation, that is, by jointly applying the horizontal addressing stretching ratio and the vertical addressing compression ratio, it calculates the new coordinates of the point in the distortion space. By subtracting the new coordinates from the original standard coordinates, the non-uniform sampling bias parameter of that point is derived, which is a two-dimensional displacement vector. For a Any coordinate point in the standard lattice Its corresponding non-uniform sampling bias parameter Calculated as:

[0046] The non-uniform sampling bias parameter essentially defines the offset of each sampling point of the convolution kernel or sampling template relative to its standard grid position. The central processing unit assembles the non-uniform sampling bias parameters of all coordinate points according to their spatial order in the original grid, generating a hydrodynamic spatial distortion phase array composed of multiple vector elements with strong directional weights, used to intervene in the underlying pre-stored logic. It is made by all The resulting matrix, whose data structure is a two-dimensional array, describes the spatial morphological distortion of the sampling neighborhood, and its expression is:

[0047] For example, following the calculation results from the previous step, the central processing unit extracts the spatial kinematics tensor... Extracting the horizontal acceleration component pixels per second² and vertical acceleration component Pixels per second². Assuming in this embodiment, the system's preset mapping function parameters are... Horizontal addressing stretching ratio factor The calculation process is as follows Vertical addressing compression ratio The calculation process is as follows .

[0048] System basis Calculate the corner coordinates based on the radial geometric topology rules of the central grid. The non-uniform sampling bias parameter, the original coordinates of this point are Its bias parameters Calculated as Through the analysis of The offset of all 9 points within the grid, including the center point, is . Performing the same calculations, the final assembly generates a hydrodynamic spatial distortion phase array containing nine two-dimensional vector elements. This array defines a rhombic sampling neighborhood that is stretched in the horizontal direction and compressed in the vertical direction.

[0049] In one embodiment of the present invention, step S3 includes the following steps: The sampling instruction blocking state of the main controller inside the deformation kernel addressing convolution engine is released, and a set of regularized third-order distributed two-dimensional grid coordinates is extracted from the internal read-only memory area to serve as the default center reference base address; the hydrodynamic spatial distortion phase array is written as a series of continuous physical control level encoded information into the address offset control register of the direct memory accessor; the underlying half-adder module and multiplexed operation circuit unit of the hardware logic gate are activated, and the gate-level physical superposition and merging mechanism operation is performed on the data presented in the default center reference base address and the address offset control register to generate the deformation address offset read and write pointer.

[0050] Specifically, the central processing unit sends a preset hardware activation instruction to the main controller inside the deformable kernel addressing convolution engine. This instruction releases the sampling instruction blocking state of the main controller's internal registers, switching it from a dormant mode with floating addresses to an executable state. The deformable kernel addressing convolution engine is a dedicated integrated circuit module designed to perform convolution operations with non-uniformly distributed sampling points.

[0051] Simultaneously, the central processing unit reads a set of regularized third-order distributed two-dimensional grid coordinates from its internal read-only memory. This set of coordinates, centered on the centroid position of the floating sphere determined in the previous time cycle, constitutes a set of nine standard integer coordinate points. This set is loaded into the base address register array, serving as the default center reference base address for the current computation cycle. The default center reference base address uses the floating sphere's centroid coordinates calculated in the previous frame as its origin. And an extension based on this The absolute memory address representation of a standard Cartesian coordinate grid. For example, for a The grid contains relative coordinate points as .

[0052] The central processing unit (CPU) enables a board-level memory configuration access tunnel, treating the hydrodynamic spatial distortion phase array generated in the previous technical step as a continuous data block. Through specific memory-mapped input / output ports, its data content is forcibly converted into a string of physical control level encoded information. This level information bypasses the operating system's virtual memory management and logical layer masking, and is written to the address offset control register of the direct memory accessor coupled to the deformation kernel addressing convolution engine via the chip's internal hardware bus. This address offset control register is a set of hardware registers, the number of which corresponds to the grid size; each register stores two floating-point components of a two-dimensional offset vector.

[0053] The central processing unit (CPU) triggers the hardware logic gates within the deformable kernel-addressable convolutional engine. Specifically, it inputs each coordinate component in the default central reference base address and the corresponding offset vector component in the address offset control register into a set of parallel hardware low-level half-adder modules and multiplexed arithmetic circuit units. These hardware units perform bitwise vector addition on the two sets of input data using a gate-level physical superposition and merging mechanism. This gate-level physical superposition and merging mechanism means that the addition operation is not performed by a general-purpose arithmetic logic unit, but by a combinational logic circuit without a clock cycle, to achieve a single-clock-cycle level operation delay of approximately several nanoseconds.

[0054] Its output is a set of physical memory addresses. This set of addresses together constitutes a deformable address offset read / write pointer that can locate the next timing target memory physical bit segment. Specifically, this deformable address offset read / write pointer... The generation is achieved by adjusting the default central reference base address. With fluid dynamics spatial distortion phase array This is achieved through physical superposition of offset vectors. For any standard grid point coordinate in the default center reference base address... The corresponding new address after deformation The calculation is as follows:

[0055] in, It represents the position of the first Line number The standard coordinates of the column; From the hydrodynamic spatial distortion phase array Extract the corresponding The non-uniform sampling bias parameter of the location. (In the formula) Representing a two-dimensional coordinate, the generated deformation address offset read / write pointer It is all these new addresses The set that defines all subsequent sampling operations on the image data. Location of each point:

[0056] in, Constructing symbols for mathematical sets; The universal quantifier symbol represents that for all, This indicates that the element belongs to the set; Indicates from 1 to The formula represents the closed interval of integers. It is composed of all the rows that meet the requirements. With column The numbers are all between 1 and... Deformation address within range A set that is made up of each other.

[0057] For example, following the calculation results from the previous step, the system has generated a hydrodynamic spatial distortion phase array. Assume the buoy's center of mass calculated in the previous frame is located in the image coordinates... The system generates a central point based on this. The default center reference base address. For the corner points of the grid. Its standard absolute address for The system will use the non-uniform sampling bias parameters calculated in the previous technical step corresponding to that corner point. Write to the address offset control register. The hardware logic gate physically superimposes these two inputs to calculate the new deformable address. .

[0058] Since the pixel coordinates are integers, the system uses its built-in nearest neighbor interpolation logic to round the result to obtain the final physical address. By repeating this process for all nine points in the default central reference base address, the system eventually generates a deformable address offset read / write pointer, which is a set of nine independently offset and rounded two-dimensional physical addresses used to guide the next data fetch operation.

[0059] In one embodiment of the present invention, step S4 includes the following steps: The system captures a sequence of step-by-step non-contiguous memory jump access instructions consisting of deformable address offset read / write pointers, and submits a data retrieval request to the memory controller's front queue pool to initiate an unconventional memory block extraction thread. Utilizing this sequence, it performs data page read operations within the original pixel buffer area, forcibly expanding the data page based on the physical interval size, traversing invalid environmental reflection interference pixel segments to pick up distorted spatial feature data at isolated data nodes. It guides the distorted spatial feature data into the first buffer delay line channel of the deformable kernel addressing convolution engine to perform temporal re-densification and realignment operations, converging the data after eliminating coordinate gaps to synthesize the target adaptive filtered pixel matrix.

[0060] Specifically, the central processing unit's memory controller captures a sequence of skipped, non-contiguous memory access instructions, consisting of variadic address offset read / write pointers, on the rising edge of each system clock. The skipping and non-contiguous nature stems from the fact that the addresses in the variadic address offset read / write pointers do not increment linearly in row or column order. This sequence is not a single instruction, but rather a group, for example, nine independent read requests pointing to non-adjacent memory addresses.

[0061] The memory controller submits this batch of requests to its internal front queue pool and initiates an unconventional memory block fetching thread specifically designed for handling non-contiguous memory access. This thread parses each address in the sequence of non-contiguous memory jump access instructions and, for the image data corresponding to the current timestamp, performs a data page read operation within the original pixel buffer, forcibly expanded based on the physical interval size. This means that the read operation is no longer a continuous access to a rectangular pixel block, but rather a jump-based fetching of pixel data from different physical pages of video memory based on discrete addresses provided by the deformed address offset read / write pointer. Through this physical page jump operation, the read process naturally traverses large areas of high-frequency reflective, invalid environmental glare interference pixel segments located between sampling points. It should be noted that invalid environmental glare interference pixel segments specifically refer to areas on the liquid surface with high brightness and high spatial frequency characteristics formed by light source reflection; these areas produce edge artifacts in traditional convolution operations.

[0062] Through the aforementioned skip reads, the system only picks up and retains the distortion spatial feature data located at isolated data nodes. Distortion spatial feature data refers to the set of pixel grayscale values ​​obtained through non-uniform sampling, which logically constitutes a local image patch distorted according to hydrodynamic characteristics. All picked-up distortion spatial feature data, i.e., the grayscale values ​​of nine pixels, are guided through the external system data bus and flow into the first buffer delay cable channel of the distortion kernel addressing convolution engine. The buffer delay cable channel is a set of physically existing circuit connections with controllable transmission delays, ensuring that pixel data arriving from different memory addresses at different times can be aligned and latched into the target register array simultaneously.

[0063] Within this buffer delay cable channel, data from different memory addresses undergoes timing re-densification and realignment, i.e., it is rearranged and stored in a compact register array, thereby eliminating coordinate gaps caused by non-contiguous reads. This register array, filled with data, is then reassembled into a target adaptive filtering pixel matrix capable of matrix operations by subsequent arithmetic units. The target adaptive filtering pixel matrix has a standard data structure. The matrix contains pixel values ​​that do not originate from a physically rectangular region of the original image, but rather from a sampling region that dynamically deforms according to hydrodynamic properties.

[0064] For example, following the deformed address offset read / write pointers containing nine non-uniform physical addresses generated in the previous step, the memory controller receives these nine addresses, for example... Nine independent single-pixel read requests are issued to the original pixel buffer. Assuming the original image contains... The central area contains a patch of saturated gray-toned horizontal water wave reflections, which cover the area from... arrive The pixel range. Traditional Convolution will read a portion of the pixels in this reflective area. However, because the sampling points in this embodiment are stretched horizontally, for example, the address of one sampling point is calculated as... This read operation physically skips the entire 10-pixel wide reflective band. The system sequentially picks up the pixel grayscale values ​​corresponding to these nine addresses, assuming they are respectively... These grayscale values ​​are fed into the buffer delay cable channel and reassembled to obtain a... The target adaptive filtering pixel matrix has the following contents:

[0065] Although the matrix has a regular structure, its pixel values ​​are derived from the physically stretched and compressed diamond-shaped regions in the image, thus avoiding the incorporation of horizontal reflective noise.

[0066] In one embodiment of the present invention, step S5 includes the following steps: The second-order moment geometric centroid estimation operation is applied to the closed graphic range of the entity's vertical contour boundary in the current frame to analyze the analytical coordinate values ​​and obtain the anti-interference centroid target position. The anti-interference centroid target position is fed back to the end of the basic historical coordinate sequence queue to provide a dependency source for subsequent secondary recursive cycle operations.

[0067] Specifically, the central processing unit (CPU) connects the pipeline of its internal arithmetic logic unit (ALU) to load the target adaptive filtering pixel matrix output by the deformable kernel-addressable convolution engine as a complete data block from its output register array into a set of input buffer registers in the core ALU module. The core ALU module consists of a set of parallel-configured digital signal processor cores or logic blocks from a field-programmable gate array (FPGA). The original, unmodified grayscale response mapping values ​​carried by each physical bit within the target adaptive filtering pixel matrix are fully preserved.

[0068] The hardware multiply-accumulate array within the core arithmetic logic unit module is activated. This array loads a preset high-pass edge enhancement operator from read-only memory and performs hardware convolution calculations on the specific asymmetric reshaped mesh segment represented by the target adaptive filter pixel matrix. To highlight the vertical contour boundary of the entity, the high-pass edge enhancement operator typically employs an operator sensitive to vertical edges, such as one with the same size as the target adaptive filter pixel matrix. Vertical Sobel operator.

[0069] This process generates a convolution matrix by element-wise multiplying a high-pass edge enhancement operator with the target adaptive filtering pixel matrix and summing the results. This operation effectively enhances vertical edges and suppresses horizontal noise signals. The hardware multiply-accumulate array asymmetric summation convolution operation is mathematically equivalent to a standard two-dimensional discrete convolution. Let the target adaptive filtering pixel matrix be... Qualcomm edge enhancement operator is The convolution result matrix Each element in The calculation is as follows:

[0070] in, and To correspond to the Qualcomm edge enhancement operator Local spatial summation index of the inner matrix dimension, operator This represents a two-dimensional discrete convolution operation. The convolution result matrix is ​​processed by a fixed threshold comparator. All pixels with values ​​higher than a preset edge threshold are set to 1, and the rest are set to 0, thereby extracting the vertical contour boundaries of entities in the current frame that have been enhanced to resist interference.

[0071] Specifically, when passing the threshold After processing, a binary image of the vertical contour boundary of the entity in the current frame is generated. The pixels with a value of 1 delineate the vertical edge of the floating ball after removing background ripple interference. The calculation method is as follows:

[0072] in, Binary image representing the vertical contour boundary of an entity In coordinates The pixel grayscale value at that location, with the absolute value sign. This indicates that the amplitude of the convolution result is taken. A second-order moment geometric centroid estimation operation is applied to the binary closed graphic region formed by the vertical contour boundaries of entities in the current frame. This operation is a noise-insensitive centroid calculation method that can provide sub-pixel level positioning results. The analytical coordinate values ​​are obtained by calculating the ratio of the first-order moment to the zero-order moment, thus deriving the latest generation of anti-interference centroid target position. Latest generation anti-interference centroid target position. The calculation is the same as in step S1, but the target is updated to a binary image. :

[0073] in, , , In this set of formulas, To be applied to the updated binary image The zeroth moment on, and This is the corresponding first moment.

[0074] The central processing unit performs a queue operation, pushing the latest generation of anti-interference centroid target position as the new coordinate point to the end of the fixed-length basic historical coordinate sequence queue. Simultaneously, it pops the earliest timestamped historical coordinate point from the head of the queue, thus providing updated motion trajectory data for subsequent secondary recursive cycle calculations. The basic historical coordinate sequence queue is a first-in-first-out (FIFO) data structure, typically 3 bytes long, used to store the centroid positions of the latest three moments, serving the needs of the next round of spatial kinematic tensor calculation, thereby forming a closed-loop adaptive control system.

[0075] For example, following the calculation results from the previous step, the system loads... The target adaptive filtering pixel matrix has the following specific values:

[0076] The lower grayscale values ​​in the left column 90 85 92 This represents the edge of the float. The system loads from read-only memory. The vertical Sober operator as a high-pass edge enhancement operator:

[0077] The hardware multiply-accumulate array performs convolution operations, and the calculation process of the resulting matrix is ​​as follows: Assuming a preset edge threshold It is 200, because Therefore, a pixel with a value of 1 is generated at the corresponding position, forming the vertical contour boundary of the entity.

[0078] Assuming that this operation is performed on a larger, complete image, resulting in a closed contour composed of multiple pixels with a value of 1, the system performs second-order moment geometric centroid estimation on this contour region, calculating the target position of the latest generation anti-interference centroid as follows: The system will assign these coordinates. The data is pushed into the basic historical coordinate sequence queue to replace the earliest data point. The updated data in this queue will be used for acceleration calculation in step S1 of the next image processing cycle.

[0079] In one embodiment of the present invention, step S6 includes the following steps: The coprocessor is enabled to perform real-time calculation of the rank index variable of the internal algebraic matrix of multiple linearly independent data in the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, it is determined that the address index has lost its directionality. The hardware bus is forcibly locked to generate a homogeneous defect pointer in which all spatial data bits point to the same read-only reserved physical storage segment. Based on the defect pointer, the division-to-zero overflow exception of the addressing convolution engine is triggered, and the hardware exception interrupt level is output to the reset pin of the image acquisition device to reset the visual acquisition parameters.

[0080] Specifically, the coprocessor in this method continuously monitors the data exchange between the central processing unit and the direct memory accessor at each computation cycle node. The parallel processing coprocessor is a logic unit implemented in a field-programmable gate array (FPGA), and its purpose is to solve the monitoring metrics without consuming the main CPU's computing resources.

[0081] When the hydrodynamic spatial distortion phase array is flushed from the memory cache to the address offset control register of the direct memory accessor, the coprocessor retrieves a copy of the array. The coprocessor reorganizes the multiple two-dimensional offset vectors in the array into a matrix and calculates the absolute value of the determinant of the matrix in real time. This value is defined as the rank index variable of the internal algebraic matrix that can characterize the linearly independent dimensions of multiple rows of data.

[0082] Specifically, the rank index variable of the internal algebraic matrix By selecting a hydrodynamic spatial distortion phase array The values ​​in the middle correspond to the basic orthogonal base points, such as the two offset vectors at coordinates (1,0) and (0,1). and Add these values ​​to the original standard coordinate vectors respectively to reconstruct the actual deformation space span vector. and, It approximates the result by calculating the area of ​​the parallelogram it spans. middle, and respectively Corresponding to the two selected basic orthogonal base points, such as the position vectors with coordinates (1,0) and (0,1); subscript This corresponds to the time step index in the aforementioned spatial kinematics tensor. The area of ​​this parallelogram is equivalent to the absolute value of the determinant of the matrix formed by these two deformed spatial span vectors:

[0083] Among them, operators The outermost matrix represents the determinant operation. This variable represents the absolute value of the result of the determinant calculation; In a physical sense, it quantifies the two-dimensional spatial extensibility of the sampling neighborhood. The larger the value, the more dispersed the sampling points are, and vice versa, the distribution tends to be linear or point-like.

[0084] Meanwhile, in the main processing flow, when the object under test, such as a buoy, is completely submerged or obstructed, causing loss of image tracking, the spatial kinematic tensor calculated in the preceding step S1 will exhibit infinite gain divergence due to the jump in input coordinates. Spatial kinematic tensor The infinite gain divergence phenomenon occurs when the absolute value of any component exceeds a preset maximum acceleration threshold. To determine:

[0085] Among them, logical operators This indicates "logical OR", meaning that the entire condition is met as long as the absolute value of any acceleration component in the horizontal or vertical direction exceeds the limit.

[0086] Once the coprocessor detects that the offset vectors calculated from the rank index variable of the internal algebraic matrix collectively approach zero due to input divergence, causing its value to collapse and fall below a preset minimum threshold, this phenomenon is judged as the address index completely losing its directionality. That is, when... The value is lower than the preset zero lower limit. ,Right now When this occurs, the system determines that rank collapse has occurred. Zero lower limit. It is set to a positive decimal close to machine zero based on the system's floating-point arithmetic precision, for example... This is used to reliably determine rank collapse with finite precision. Based on the IEEE 754 standard for single-precision floating-point numbers, the effective mantissa precision is approximately... , set as It can effectively filter out truncation errors and rounding noise caused by floating-point operations, and can also determine whether the area of ​​a polygon has truly undergone mathematical singularity convergence, that is, the area has degenerated to zero.

[0087] This judgment will trigger the system to lock on the hardware bus and generate a homogeneous defect pointer where all spatial data bits point to the same read-only reserved physical memory segment. A homogeneous defect pointer is a program state where all address vectors within the deformable address offset read / write pointer converge to the same value. When this homogeneous defect pointer is subsequently passed to the deformable kernel addressing convolution engine, its internal geometric normalization module will calculate a zero area due to the overlap of all sampling point coordinates. This will cause the passive denominator to overflow and short-circuit during the normalized division operation, thus throwing an abnormal hardware physical emergency stop level. Specifically, the normalized division operation inside the deformable kernel addressing convolution engine can be represented as:

[0088] in, It is the original, unweighted data value output by the deformable kernel-addressable convolutional engine before the area normalization operation. It is a deformable address offset read / write pointer The area of ​​the sampling region formed by the address points in the code. It is a very small positive number set to prevent accidental division by zero. When rank collapse occurs, such as in the prime defect pointer state, The value approaches 0, resulting in an extremely small denominator that triggers a hardware overflow.

[0089] An abnormal hardware physical emergency stop level is a high-priority, non-maskable hardware interrupt signal, such as a logic high level with a width of not less than 100 ns. This is sufficient to meet the timing requirements of the hardware reset pin of most complementary metal-oxide-semiconductor (CMOS) image sensors. The reason for setting this width range is that the RESET pin of a CMOS sensor typically requires the level toggling state to be maintained for at least tens of nanoseconds before it can be latched and recognized by the internal register. A pulse width of 100 ns ensures the absolute reliability of the signal triggering and solves the problem of system recovery delay caused by excessively long high levels.

[0090] The abnormal signal, transmitted via physical circuitry independent of the software operating system, impacts and triggers the hardware reset pin of the connected environmental perception image acquisition device. This initializes the device's internal exposure logic gates and gain register, thus hard-setting the visual acquisition operating parameters. Resetting these parameters includes restoring the image sensor's exposure time to its default intermediate value, resetting the automatic gain control (AGC) to its initial range, and setting the white balance to automatic mode.

[0091] For example, consider a scenario where the buoy is completely covered by a large, opaque floating object. The centroid detection algorithm in the preceding steps provides results in three frames. , , The disordered coordinates cause the components of the spatial kinematic tensor calculated in step S1 to be inconsistent. achieve Pixels per second², far exceeding the system's maximum acceleration threshold. Pixels per second². When this acceleration value is input into the hardware lookup table in step S2, the over-limit overflow protection logic is triggered, forcing the hardware to output a very small scaling factor close to 0. This causes the offset vector of each point in the generated hydrodynamic spatial distortion phase array to be forcibly approximated to the coordinate components that cancel out their relative base address. Based on this, the coprocessor extracts the two reconstructed deformation space span vectors. and The absolute values ​​of all components decayed to Assuming extraction calculation and Calculate the rank index variable of the internal algebraic matrix. This value is much smaller than the set lower limit of zero. The system determined that the order had collapsed.

[0092] Therefore, since the offset vector almost completely cancels out the relative coordinate stretching, the physical addresses of all nine sampling points in the generated deformable address offset read / write pointer converge and collapse to the same base address, for example... When the deformable kernel-addressable convolutional engine attempts to normalize the region formed by these nine overlapping points, the calculated area... If the value is 0, the hardware divider executes. If a floating-point overflow occurs, a 5V high-level signal lasting 200 ns is immediately generated on a specific pin of the processor as an emergency hardware stop level. This signal is sent to the RESET pin of the environmental perception image acquisition device via a physical wire, restoring all its internal configuration registers to their factory default values. This completes a hard reset of the visual acquisition operation parameters and prepares for a new round of scene capture.

[0093] See appendix Figure 2 The present invention also proposes an image processing system for adaptive optimization detection of float position, comprising the following modules: The kinematics analysis module receives the raw dynamic video data stream and writes it into the video memory space to construct the raw pixel buffer. It extracts the basic geometric center data of three consecutive frames in the raw pixel buffer and generates a spatial kinematic tensor based on the basic geometric center data. The distortion array generation module calculates the addressing scaling factor based on the spatial kinematic tensor, and uses the addressing scaling factor to modify the preset standard lattice distribution coordinates to generate a fluid dynamic spatial distortion phase array. The deformation pointer generation module obtains the default center reference base address and superimposes and merges the fluid dynamics spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer. The adaptive sampling module extracts distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and then combines the distortion spatial feature data to generate the target adaptive filtered pixel matrix. The position update and feedback module uses a preset high-pass edge enhancement operator to perform a convolution operation on the target adaptive filtering pixel matrix to extract the vertical contour boundary of the entity, and calculates the anti-interference centroid target position based on the vertical contour boundary of the entity. The system stability monitoring module calculates the rank index variable of the internal algebraic matrix of the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, a defect pointer is generated and a hardware abnormal interrupt level is triggered to reset the vision acquisition parameters.

[0094] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0095] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An image processing method for adaptive optimization detection of float position, characterized in that, Includes the following steps: S1. Receive the raw dynamic video data stream and write it into the video memory space to construct the raw pixel buffer area. Extract the basic geometric center data of three consecutive frames in the raw pixel buffer area and generate the spatial kinematic tensor based on the basic geometric center data. S2. Calculate the addressing scaling factor based on the spatial kinematic tensor, and use the addressing scaling factor to modify the preset standard lattice distribution coordinates to generate a fluid dynamic spatial distortion phase array. S3. Obtain the default center reference base address, and superimpose and merge the fluid dynamics spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer; S4. Extract distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and combine the distortion spatial feature data to generate the target adaptive filter pixel matrix; S5. Using a preset high-pass edge enhancement operator, perform a convolution operation on the target adaptive filtering pixel matrix to extract the vertical contour boundary of the entity, and calculate the position of the anti-interference centroid target based on the vertical contour boundary of the entity. S6. Calculate the rank index variable of the internal algebraic matrix of the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, generate a defect pointer and trigger a hardware abnormal interrupt level to reset the vision acquisition parameters.

2. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, Extracting the fundamental geometric center data of three consecutive frames from the original pixel buffer includes the following steps: Activate baseline locking operation within the original pixel buffer and capture three consecutive adjacent frames from the startup sequence, where the startup sequence is the initial video sequence when the system starts. Perform grayscale thresholding with preset fixed parameters on three consecutive frames of images to generate a binary mask image; Extract the centroid coordinates of the connected domain of the floating sphere in the binary mask image as the basic geometric center data.

3. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, Generating a spatial kinematic tensor based on fundamental geometric center data includes the following steps: Discrete difference equations are input for the basic geometric center data on the horizontal transverse dimension axis and the vertical gravity dimension axis, respectively; The second-order central difference equation is executed to obtain the horizontal acceleration component caused by wave impact and the vertical acceleration component caused by the float's own buoyancy. The horizontal acceleration components are combined with the vertical acceleration components to generate a spatial kinematic tensor that expresses the intensity of physical disturbances in the underlying environment.

4. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, The addressing scaling factor is calculated based on the spatial kinematic tensor, including the following steps: Extract the horizontal and vertical acceleration components from the spatial kinematic tensor; After dividing the absolute value of the horizontal acceleration component by the system's preset reference acceleration constant and performing dimensionless processing, it is substituted into the preset first logarithmic distortion mapping function to perform forward conversion and output the horizontal addressing stretching ratio coefficient. After dividing the absolute value of the vertical acceleration component by the reference acceleration constant, the result is substituted into the second logarithmic distortion mapping function with an additional zero overflow prevention factor for reverse conversion to output the vertical addressing compression ratio coefficient. The horizontal addressing stretching ratio coefficient and the vertical addressing compression ratio coefficient together constitute the addressing ratio coefficient.

5. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, Generating a hydrodynamic spatial distortion phase array by modifying the preset standard lattice distribution coordinates using addressing scaling factors includes the following steps: Based on the preset central grid radial geometric topology rules, the horizontal addressing stretching ratio coefficient and the vertical addressing compression ratio coefficient are jointly applied to modify the preset standard lattice distribution coordinates in a directional manner, thereby deriving the non-uniform sampling bias parameter of the specified step size neighborhood. All non-uniform sampling bias parameters are compiled to generate a hydrodynamic spatial distortion phase array for intervening in the underlying pre-stored logic.

6. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, Obtain the default center reference base address, and then superimpose and merge the hydrodynamic spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer, including the following steps: Release the sampling instruction blocking state of the main controller inside the deformable kernel addressing convolutional engine, and extract a set of regularized third-order distributed two-dimensional grid coordinates from the internal read-only memory area to serve as the default center reference base address; The fluid dynamics spatial distortion phase array is written as a series of continuous physical control level encoded information into the address offset control register of the direct memory accessor; The underlying half-adder module and multiplexed operation circuit unit of the hardware logic gate are activated, and the gate-level physical superposition and merging mechanism is performed on the data presented in the default center reference base address and address offset control register to generate deformed address offset read and write pointers.

7. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, The distortion spatial feature data is extracted from the original pixel buffer based on the distortion address offset read / write pointer, and the distortion spatial feature data is then combined to generate the target adaptive filtering pixel matrix, including the following steps: Capture the sequence of non-contiguous memory jump access instructions consisting of deformable address offset read and write pointers, and submit a data retrieval request to the memory controller front queue pool to start the non-standard memory block retrieval thread; By using a sequence of stepping non-contiguous memory jump access instructions, data page reads are performed within the original pixel buffer area that are forcibly expanded based on the physical interval size, crossing invalid environmental reflective interference pixel segments to pick up distorted spatial feature data at isolated data nodes. The distortion spatial feature data is guided into the first buffer delay line channel of the deformation kernel addressing convolution engine to perform temporal re-densification and realignment operations, and the data after eliminating coordinate gaps are combined to form the target adaptive filter pixel matrix.

8. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, The location of the anti-interference centroid target is calculated based on the vertical contour boundary of the entity, including the following steps: Apply a second-order moment geometric centroid estimation operation to the closed graphic range of the entity's vertical contour boundary in the current frame to analyze the analytical coordinate values ​​and obtain the anti-interference centroid target position; Feeding back the anti-interference centroid target position to the end of the basic historical coordinate sequence queue provides a dependency source for subsequent secondary recursive periodic operations.

9. The image processing method for adaptive optimization detection of float position according to claim 1, characterized in that, Generate a defect pointer and trigger a hardware exception interrupt level to reset the vision acquisition parameters, including the following steps: Enable the coprocessor to perform real-time calculation of the rank index variable of the internal algebraic matrix of multiple linearly independent data rows in the hydrodynamic spatial distortion phase array; When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, it is determined that the address index has lost its directionality, and the hardware bus is forcibly locked to generate a homogeneous defect pointer in which all spatial data bits point to the same read-only reserved physical memory segment. Based on the defect pointer, a division-to-zero overflow exception is triggered in the division operation of the addressing convolution engine, and a hardware exception interrupt level is output to the reset pin of the image acquisition device to reset the visual acquisition parameters.

10. An image processing system for adaptive optimization detection of float position, used to implement the image processing method for adaptive optimization detection of float position as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The kinematics analysis module receives the raw dynamic video data stream and writes it into the video memory space to construct the raw pixel buffer. It extracts the basic geometric center data of three consecutive frames in the raw pixel buffer and generates a spatial kinematic tensor based on the basic geometric center data. The distortion array generation module calculates the addressing scaling factor based on the spatial kinematic tensor, and uses the addressing scaling factor to modify the preset standard lattice distribution coordinates to generate a fluid dynamic spatial distortion phase array. The deformation pointer generation module obtains the default center reference base address and superimposes and merges the fluid dynamics spatial distortion phase array with the default center reference base address to generate a deformation address offset read / write pointer. The adaptive sampling module extracts distortion spatial feature data from the original pixel buffer based on the distortion address offset read / write pointer, and then combines the distortion spatial feature data to generate the target adaptive filtered pixel matrix. The position update and feedback module uses a preset high-pass edge enhancement operator to perform a convolution operation on the target adaptive filtering pixel matrix to extract the vertical contour boundary of the entity, and calculates the anti-interference centroid target position based on the vertical contour boundary of the entity. The system stability monitoring module calculates the rank index variable of the internal algebraic matrix of the hydrodynamic spatial distortion phase array. When the rank index variable of the internal algebraic matrix is ​​lower than the preset zero value lower limit, a defect pointer is generated and a hardware abnormal interrupt level is triggered to reset the vision acquisition parameters.

Images