Image feature calibration method based on motion offset

Abstract

The invention discloses an image feature calibration method based on motion offset, and relates to the field of track detection image processing. The method comprises the steps of motion image acquisition, initial image calibration, fuzzy feature calibration and calibration image output. Acquiring real-time acceleration data through a train acceleration sensor, and converting the real-time acceleration data into pixel displacement deviation to realize preliminary calibration so as to obtain a reference processing image; based on historical database target features, separating a target body from an environment area, generating an environment point processing curve in combination with motion parameters, and screening a to-be-calibrated pixel set through feature fitting; a fuzzy feature calibration strategy is adopted to calibrate pixels and then fuse the calibrated pixels with a reference image, and the calibration precision is dynamically optimized through curve calibration and effect verification steps. Layered calibration is realized, the detail reduction degree of the moving image is improved, the environmental adaptability is enhanced, the target feature integrity is guaranteed, and the method is suitable for a rail transit image monitoring scene.

Description

Technical Field

[0001] The present invention relates to the field of track detection image processing, and in particular to an image feature calibration method based on motion offset. Background Art

[0002] With the rapid development of intelligent rail transit, image acquisition equipment onboard trains has been widely used in key scenarios such as track environment monitoring, line inspections, and safety warnings. However, during high-speed train operation, image acquisition equipment can experience periodic or sudden motion offsets due to factors such as acceleration during startup, deceleration during braking, and centrifugal forces on curves. This can lead to quality issues in captured motion images, such as pixel shifting, blurred edges, and motion artifacts.

[0003] Existing calibration methods for moving images fall into two main categories: one is based on pure vision algorithms, achieving offset correction through image feature point matching. However, these methods are highly dependent on image texture features and are prone to failure in low-texture scenes such as rail tunnels and single backgrounds. Their high computational complexity makes them difficult to meet the real-time processing requirements of trains. The other relies on motion sensor-assisted calibration, using devices such as accelerometers and gyroscopes to obtain motion parameters and apply them to image correction. However, these traditional methods often employ simple linear displacement compensation models, which are unable to cope with the nonlinear image offsets caused by complex train motion. In particular, they are ineffective in correcting motion blur and non-uniform afterimages generated by large acceleration rates. Furthermore, existing calibration methods generally fail to distinguish between the target subject and the surrounding area in the image, applying a unified calibration strategy for both, which can easily lead to distortion of the target subject's features. Furthermore, they lack a dynamic optimization mechanism for the calibration model. Calibration accuracy decreases significantly when the train's operating environment or motion state exceeds a preset range, making it difficult to adapt to the complex and variable nature of rail transit scenarios.

[0004] Therefore, in order to solve the problems existing in the prior art, the present invention proposes an image feature calibration method based on motion offset. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention aims to provide an image feature calibration method based on motion offset.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method for calibrating image features based on motion offset includes the following steps: During the motion image acquisition step, the train uses an encoder mounted on the vehicle to record running speed information in real time, while a camera device is installed trackside. Through real-time train-to-ground communication, the train transmits the speed information recorded by the encoder to a public machine near the camera. After receiving the speed information, the public machine parses the speed information and determines the appropriate acquisition frequency. This captures the train's running motion image and simultaneously reads the ideal environment image from the historical database. an initial image calibration step, obtaining the acceleration of the train at the time of shooting, determining the basic displacement deviation of the image based on the acceleration, and performing preliminary image calibration on the motion image to obtain a reference processed image; Fuzzy feature calibration step: identifying the target subject and the environmental area in the reference processed image based on the target features in the historical database, and extracting the environmental pixels in the environmental area; generating an environmental point processing curve containing expected features by combining the motion parameters of the train at the time of shooting and the ideal environmental image, extracting the actual features of the environmental pixels and performing feature-level fitting on the environmental point processing curve, and selecting pixels whose deviations exceed a preset threshold as the pixel set to be calibrated; The calibration image output step calibrates the pixel set to be calibrated by a fuzzy feature calibration strategy, and fuses the calibrated pixel set with the reference processed image to obtain a processed standard motion image.

[0007] As a further improvement of the present invention, the initial image calibration step includes obtaining real-time acceleration data at the time of shooting through an acceleration sensor carried by the train, the real-time acceleration data including the magnitude and direction of acceleration; converting the real-time acceleration data into a basic displacement deviation of the image in the pixel dimension based on a preset acceleration-displacement conversion model, the basic displacement deviation including pixel offset values ​​in the horizontal and vertical directions; performing reverse compensation correction on the pixel coordinates of the motion image according to the basic displacement deviation, filling the pixel missing areas caused by displacement with neighborhood pixel interpolation, and outputting the image after preliminary calibration as a reference processed image.

[0008] As a further improvement of the present invention, the fuzzy feature calibration step includes: based on the target features in the historical database, performing feature comparison on the benchmark processed image, identifying and separating the target subject and the environmental area in the image, extracting environmental pixel points from the environmental area, and the environmental pixel points are the pixel points of the non-target subject in the benchmark processed image; combining the motion parameters of the train at the time of shooting and the ideal environmental image, calculating and generating an environmental point processing curve through the environmental point processing strategy, the environmental point processing curve includes the expected spatial distribution characteristics and grayscale change characteristics of the environmental pixel points when there is no motion offset, the motion parameters include the motion direction and the acceleration change rate, extracting the actual features of the environmental pixel points, the actual features include the spatial position offset, the pixel grayscale attenuation rate and the gradient direction; performing feature-level fitting on the actual features of the environmental pixel points and the expected features of the environmental point processing curve, screening out the pixel points whose deviation between the actual features and the expected features exceeds a preset threshold, and outputting them as the pixel set to be calibrated.

[0009] As a further improvement of the present invention, the ideal environment image is an environment image collected by the train under preset ideal working conditions and manually annotated and processed without motion offset calibration. The image acquisition device carried by the train performs periodic image acquisition when the train is in the preset ideal working conditions, and the acquisition frequency is not lower than a preset threshold; the target features are obtained by manually annotating the collected ideal environment images, extracting the characteristic parameters of key objects, establishing a feature model and storing it in a historical database, and updating and optimizing the feature model to obtain a target feature database.

[0010] As a further improvement of the present invention, the environmental point processing curve is a multidimensional feature set that describes the characteristic change law of environmental pixel points in a state without motion offset, including a pixel spatial position distribution curve, a grayscale value attenuation curve and a gradient direction change curve. The environmental point processing curve extracts the characteristic parameters of all environmental pixel points in the environmental area based on the ideal environmental image in the historical database, establishes a baseline change curve of each characteristic parameter through statistical analysis, and obtains the motion parameters of the train at the time of shooting, including motion direction, instantaneous speed, and acceleration change rate. The baseline change curve is dynamically corrected based on a preset motion parameter feature influence model, and the corrected curve is compared with the environmental image characteristics without motion offset within a preset time in the historical database, the deviation is calculated, and the curve parameters are optimized through an iterative algorithm to generate the environmental point processing curve in the current motion state.

[0011] As a further improvement of the present invention, the motion parameter feature influence model is obtained by training the historical data of the train motion state and the corresponding image feature changes through a machine learning algorithm to obtain a mapping relationship between the motion parameters and the image feature changes, and the model input parameters and output parameters are determined. When the motion parameters of the train at the moment of shooting are obtained, the motion parameter feature influence model is input to obtain the corresponding feature change correction coefficient, and the baseline change curve is corrected pixel by pixel according to the feature change correction coefficient to obtain the corrected environment point processing curve.

[0012] As a further improvement of the present invention, the fuzzy feature calibration strategy includes, for each pixel point in the pixel set to be calibrated, based on its deviation from the environmental point processing curve, the train motion parameters and the pixel features of the corresponding position in the ideal environmental image, calculating the pixel position compensation value, grayscale correction coefficient and edge sharpening parameters; performing coordinate offset compensation, grayscale value correction and edge contour enhancement processing on the pixel points according to the calibration parameters to generate calibrated pixel points; setting a spatial distance weight factor to assign different fusion weights to the calibrated pixel points and the surrounding non-calibrated pixel points, the closer the distance, the higher the weight ratio; using a weighted average algorithm to fuse the calibrated pixel points with the pixels of the corresponding area in the benchmark processed image, and performing smooth transition processing on the fusion boundary to generate a standard motion image.

[0013] As a further improvement of the present invention, it also includes an environmental point processing curve calibration step. When the proportion of the pixel set to be calibrated exceeds a preset ratio threshold after feature fitting of several consecutive images, a preset number of ideal environmental images and motion parameters at corresponding moments are selected from the historical database as calibration samples; the current environmental point processing curve is applied to the calibration sample, and the overall deviation value between the actual characteristics of the environmental pixel points in the sample and the expected characteristics of the curve is calculated; based on the overall deviation value, the characteristic parameters of the environmental point processing curve are corrected by the least squares method, and the correction amplitude is positively correlated with the deviation value. The adjusted environmental point processing curve is applied to a new verification sample. If the feature fitting deviation of the verification sample is lower than the preset qualified threshold, the curve is updated; otherwise, the curve calibration is continued.

[0014] As a further improvement of the present invention, it also includes a calibration effect verification step, setting verification indicators, including pixel deviation rate, feature matching degree and image clarity score, wherein the pixel deviation rate is the average value of the pixel position deviation between the calibrated image and the ideal environment image, and the feature matching degree is the overlap degree between the target subject feature and the standard feature in the historical database; the processed standard motion image is subjected to indicator detection, and the actual value of each verification indicator is calculated; the actual value is compared with the preset qualified threshold value, and the image is output if all the standards are met; if the standards are not met, the calibration step corresponding to the failed indicator is triggered to perform parameter backtracking adjustment, and the whole process from initial image calibration to image fusion is repeated until the verification indicator meets the standards.

[0015] The beneficial effects of the present invention are: (1) Through the initial image calibration step, the basic displacement deviation correction is realized based on the acceleration parameters. Then, through the fuzzy feature calibration step, the target body and the environmental area are distinguished. Feature-level fitting and screening are performed on the environmental pixels, and the calibration accuracy is layered. This effectively solves the problem of insufficient correction of nonlinear offset and dynamic afterimage in traditional methods, and significantly improves the detail restoration of motion images.

[0016] (2) Based on the ideal environment image and target feature model in the historical database, a dynamically adjusted environment point processing curve is generated in combination with real-time motion parameters. The model is adaptively optimized through a special curve calibration step to ensure that the calibration effect can be maintained stably when the train motion state changes or the environment differs, thus overcoming the limitation of traditional methods that are highly dependent on fixed scenes.

[0017] (3) By comparing target features, the target body and the environmental area are separated, and the environmental pixels are calibrated in a targeted manner, avoiding interference with the features of key targets such as track signs and traffic lights during the calibration process, ensuring the integrity and accuracy of the target body features, and providing reliable image data support for subsequent track monitoring, safety warning and other applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a method flow chart of an image feature calibration method based on motion offset according to the present invention; Figure 2 is a flow chart of the fuzzy feature calibration steps according to an embodiment of the present invention; Figure 3 is a flow chart of steps for constructing an environmental point processing curve according to an embodiment of the present invention; Figure 4 is a flow chart of the environmental point processing curve calibration steps according to an embodiment of the present invention; Figure 5 4 is a flow chart of the calibration effect verification steps of an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom," "top," "inner," and "outer" refer to directions toward or away from the geometric center of a particular component, respectively.

[0020] The embodiment of the present invention proposes an image feature calibration method based on motion offset, such as Figures 1 to 5 As shown, the following steps are included: A motion image acquisition step is to acquire the motion image of the train while it is running and read the ideal environment image from the historical database; The train's onboard image acquisition device captures real-time images of the train's movement. This device, which can be a high-definition industrial camera or visual sensor, covers the area in front of and surrounding the train. Simultaneously, an ideal environment image matching the current route is retrieved from a pre-built historical database. This ideal environment image is captured and processed when the train is operating under preset ideal conditions, including constant speed, stationary status, and standard lighting conditions. These images are manually annotated and calibrated for motion offset, serving as a baseline reference for image feature comparison.

[0021] an initial image calibration step, obtaining the acceleration of the train at the time of shooting, determining the basic displacement deviation of the image based on the acceleration, and performing preliminary image calibration on the motion image to obtain a reference processed image; The train's onboard accelerometers collect real-time acceleration data at the moment the image is captured. This data includes the specific magnitude and direction of the acceleration. Based on a pre-set acceleration-displacement conversion model, the real-time acceleration data is converted into the image's basic displacement deviation in the pixel dimension, which includes horizontal and vertical pixel offset values. Based on this basic displacement deviation, the pixel coordinates of the moving image are inversely compensated. Neighborhood pixel interpolation is used to fill in missing pixels caused by displacement. After coordinate correction and missing pixel filling, a preliminarily calibrated image is output as the baseline processed image.

[0022] Fuzzy feature calibration step: identifying the target subject and the environmental area in the reference processed image based on the target features in the historical database, and extracting the environmental pixels in the environmental area; generating an environmental point processing curve containing expected features by combining the motion parameters of the train at the time of shooting and the ideal environmental image, extracting the actual features of the environmental pixels and performing feature-level fitting on the environmental point processing curve, and selecting pixels whose deviations exceed a preset threshold as the pixel set to be calibrated; First, feature comparison analysis is performed on the baseline processed image based on target features from a historical database to identify and separate the target subject and surrounding areas in the image. Target features are generated by manually annotating key objects in the idealized environmental image, extracting feature parameters, and then building and optimizing a feature model. Key objects include track markings, traffic lights, surrounding buildings, and other objects of particular interest. Environmental pixels are extracted from the separated environmental areas. Environmental pixels are defined as those pixels in the baseline processed image that do not belong to the target subject. An environmental point processing strategy is used to calculate and generate an environmental point processing curve, combining the train's motion parameters at the time of capture with the idealized environmental image. Motion parameters include motion direction and acceleration rate. The environmental point processing curve contains the expected spatial distribution and grayscale variation characteristics of the environmental pixels in a state without motion offset. Actual features of the environmental pixels are extracted, including spatial position offset, pixel grayscale decay rate, and gradient direction. Feature-level fitting is performed between the actual features of the environmental pixels and the expected features of the environmental point processing curve. Pixels whose deviation between the actual and expected features exceeds a preset threshold are selected and integrated into the pixel set to be calibrated.

[0023] The calibration image output step calibrates the pixel set to be calibrated by a fuzzy feature calibration strategy, and fuses the calibrated pixel set with the reference processed image to obtain a processed standard motion image.

[0024] A fuzzy feature calibration strategy is used to calibrate the pixels in the pixel set to be calibrated. For each pixel in the pixel set to be calibrated, the pixel position compensation value, grayscale correction coefficient, and edge sharpening parameter are calculated by combining the deviation between the pixel and the environmental point processing curve, the train motion parameters, and the pixel features of the corresponding position in the ideal environmental image. Based on these calibration parameters, the pixel points are subjected to coordinate offset compensation, grayscale value correction, and edge contour enhancement to generate calibrated pixel points. A spatial distance weight factor is set, and different fusion weights are assigned to the calibrated pixel points and the surrounding non-calibrated pixels according to this factor, where the closer the distance, the higher the weight of the pixel points. The calibrated pixels are fused with the pixels of the corresponding area in the reference processed image through a weighted average algorithm, and the fusion boundary is smoothed to generate a processed standard motion image.

[0025] Specifically, such as Figures 1 to 5As shown, the initial image calibration step includes obtaining real-time acceleration data at the time of shooting through an acceleration sensor carried by the train, the real-time acceleration data including acceleration magnitude and direction; converting the real-time acceleration data into a basic displacement deviation of the image in the pixel dimension based on a preset acceleration-displacement conversion model, the basic displacement deviation including pixel offset values ​​in the horizontal and vertical directions; performing reverse compensation correction on the pixel coordinates of the motion image according to the basic displacement deviation, filling the pixel missing areas caused by displacement with neighborhood pixel interpolation, and outputting the image after preliminary calibration as a reference processed image.

[0026] The train uses an accelerometer to continuously collect real-time acceleration data at the moment the image is captured. This sensor can be integrated into the train's onboard control system and can accurately capture the acceleration information of the train during movement. The real-time acceleration data obtained not only includes the specific numerical value of the acceleration, but also the direction information of the acceleration, thereby comprehensively reflecting the changes in the train's motion state at the moment of capture.

[0027] Real-time acceleration data is processed based on a pre-defined acceleration-displacement conversion model, converting physical acceleration parameters into basic image displacement deviations in the pixel dimension. This conversion model, developed by analyzing extensive historical data on train motion states and corresponding image displacements, quantifies the mapping between acceleration and pixel displacement. The resulting basic displacement deviations are expressed as horizontal and vertical pixel offsets, corresponding to the degree of lateral and longitudinal image displacement, respectively.

[0028] Based on the calculated basic displacement deviation, the pixel coordinates of the moving image are reversely compensated. During the correction process, the coordinate position of each pixel is adjusted in a direction opposite to the displacement to offset the initial offset caused by the train's movement. Neighborhood pixel interpolation is used to fill in missing pixel areas caused by pixel displacement during the correction process. Neighborhood pixel interpolation extracts feature information from valid pixels surrounding the missing area and generates fill pixels based on spatial distribution patterns to ensure image continuity in the missing area. After coordinate correction and gap filling, the output is a preliminarily calibrated image, which serves as the baseline processed image for subsequent fine-tuning.

[0029] Specifically, such as Figures 1 to 5As shown, the fuzzy feature calibration step includes: based on the target features in the historical database, performing feature comparison on the benchmark processed image, identifying and separating the target subject and the environmental area in the image, extracting environmental pixel points from the environmental area, and the environmental pixel points are the pixel points of the non-target subject in the benchmark processed image; combining the motion parameters of the train at the time of shooting and the ideal environmental image, generating an environmental point processing curve through environmental point processing strategy calculation, the environmental point processing curve includes the expected spatial distribution characteristics and grayscale change characteristics of the environmental pixel points when there is no motion offset, the motion parameters include motion direction and acceleration change rate, extracting the actual features of the environmental pixel points, the actual features include spatial position offset, pixel grayscale attenuation rate and gradient direction; performing feature-level fitting on the actual features of the environmental pixel points and the expected features of the environmental point processing curve, screening out pixels whose deviation between the actual features and the expected features exceeds a preset threshold, and outputting them as the pixel set to be calibrated.

[0030] Based on pre-constructed target features in the historical database, feature comparison analysis is performed on the baseline processed image. Target features are generated by extracting and training features of key objects in the ideal environment image, encompassing characteristic parameters such as color, shape, and texture. During the feature comparison process, feature information from the baseline processed image is extracted and matched against the target features to accurately identify the target subject in the image. The remaining image regions, excluding the target subject, are then separated into environmental regions. Environmental pixels are extracted from the separated environmental regions; these pixels belong to the portion of the baseline processed image that does not belong to the target subject. The extraction process involves pixel-level region segmentation of the image. Based on the boundary features between the target subject and the environmental region, all pixels within the environmental region are included in the extraction range, forming a set of environmental pixels. The environmental point processing curve is generated by combining the train's motion parameters at the time of capture with the ideal environment image using an environmental point processing strategy. Motion parameters specifically include the train's direction and rate of acceleration, reflecting its dynamic motion at the moment of capture. The ideal environment image provides a baseline for environmental features without motion offset. The environmental point processing strategy integrates the influence of motion parameters on image features with the characteristic distribution of an ideal environmental image to construct a multidimensional feature variation model, thereby generating an environmental point processing curve. This curve contains the expected spatial distribution characteristics and expected grayscale variation characteristics of environmental pixels in a non-motion-offset state. The expected spatial distribution characteristics describe the ideal position distribution of pixels in the image coordinate system, while the expected grayscale variation characteristics reflect the normal variation of pixel grayscale values ​​with spatial position. The actual features of environmental pixels are extracted, including spatial position offset, pixel grayscale decay rate, and gradient direction. The spatial position offset is calculated by calculating the difference between the actual and ideal coordinates of the environmental pixel, reflecting the positional deviation of the pixel due to motion offset. The pixel grayscale decay rate is determined by analyzing the ratio of the difference between the pixel grayscale value and the grayscale value at the corresponding position in the ideal environmental image, reflecting the degree of grayscale distortion caused by motion blur. The gradient direction is obtained by calculating the direction of the maximum grayscale change in the pixel neighborhood, representing the actual direction of the pixel edge. The actual characteristics of the environmental pixel are then fitted to the expected characteristics of the environmental point processing curve at the feature level. The fitting process establishes a feature mapping relationship, compares the degree of fit between the actual and expected features one by one, and calculates the difference between the two. A preset threshold is set as the judgment standard, and pixels where the difference between the actual and expected features exceeds the threshold are screened out. These pixels are determined to be significantly affected by motion offset, and are integrated into the pixel set to be calibrated and output.

[0031] Specifically, such as Figures 1 to 5As shown, the ideal environment image is an environment image collected by the train under preset ideal working conditions and manually annotated and processed without motion offset calibration. The image acquisition device carried by the train performs periodic image acquisition when the train is in the preset ideal working conditions, and the acquisition frequency is not less than a preset threshold; the target features are obtained by manually annotating the collected ideal environment images, extracting the characteristic parameters of key objects, establishing a feature model and storing it in a historical database, and updating and optimizing the feature model to obtain a target feature database.

[0032] The ideal environment image is captured when the train is in preset ideal operating conditions and is manually annotated and calibrated for motion offset. These conditions refer to scenarios where the train is operating stably and external environmental interference is minimal, including situations where the train is traveling at a constant speed, parked, or under standard lighting conditions. Under these conditions, image offset caused by train motion is minimal, and the captured image accurately reflects the original characteristics of the environment. The ideal environment image is acquired through periodic image acquisition using an onboard image acquisition device. This device must possess stable imaging performance to ensure image clarity and color consistency. The acquisition process continues at regular intervals, with a frequency no less than a preset threshold, to ensure a sufficient number of sample images under various environmental conditions and road sections, providing sufficient data support for subsequent feature extraction and model building. Target features are acquired through manual annotation of the captured ideal environment image. During the annotation process, technicians accurately identify and select key objects in the image, including track markings, traffic lights, buildings along the route, and other environmental elements requiring key monitoring. Based on the annotation results, the feature parameters of key objects are extracted. These parameters include information that can characterize the uniqueness of the object, such as color distribution range, geometric shape characteristics, and surface texture patterns. The extracted feature parameters are integrated to establish a feature model, and the model is stored in a historical database. To adapt to environmental changes and subtle changes in object characteristics, the feature model needs to be updated and optimized regularly. The update process adjusts the feature parameter weights in the model by introducing newly collected ideal environment image data; the optimization process improves the model's recognition accuracy for key objects by eliminating abnormal samples and correcting feature deviations. The continuously updated and optimized feature model collection together constitutes the target feature database, providing a stable and reliable benchmark for the recognition of target subjects in images.

[0033] Specifically, such as Figures 1 to 5As shown, the environmental point processing curve is a multidimensional feature set that describes the characteristic change law of environmental pixel points in the state without motion offset, including a pixel spatial position distribution curve, a gray value attenuation curve and a gradient direction change curve. The environmental point processing curve extracts the characteristic parameters of all environmental pixel points in the environmental area based on the ideal environmental image in the historical database, establishes a baseline change curve of each characteristic parameter through statistical analysis, and obtains the motion parameters of the train at the time of shooting, including motion direction, instantaneous speed, and acceleration change rate. The baseline change curve is dynamically corrected based on a preset motion parameter feature influence model, and the corrected curve is compared with the environmental image features without motion offset within a preset time in the historical database. The deviation is calculated and the curve parameters are optimized through an iterative algorithm to generate the environmental point processing curve in the current motion state.

[0034] The environmental point processing curve is a multidimensional feature set used to describe the characteristic variation of environmental pixel points in a non-motion-offset state. It includes a pixel spatial position distribution curve, a grayscale value decay curve, and a gradient direction variation curve. The pixel spatial position distribution curve characterizes the ideal position distribution of environmental pixels in the image coordinate system, the grayscale value decay curve reflects the normal decay trend of the grayscale value of environmental pixels as they change with spatial position, and the gradient direction variation curve describes the natural distribution of gradient directions along the edges of environmental pixels. Together, these three curves form a complete characterization of the non-motion-offset environmental image. The generation process of the environmental point processing curve begins with an ideal environmental image in a historical database. Feature parameters of all environmental pixels within the environmental area are extracted from the ideal environmental image. These parameters include key information such as spatial coordinates, grayscale value, and gradient direction. These feature parameters are processed through statistical analysis to calculate the mean, distribution probability, and variation trend of the feature parameters at different spatial positions. Based on this, a baseline variation curve for each feature parameter is established. The baseline variation curve reflects the inherent variation pattern of environmental pixel characteristics in the absence of motion interference. After obtaining the motion parameters at the time of train capture, the baseline variation curve is dynamically corrected. Motion parameters include the train's direction of motion, instantaneous velocity, and rate of acceleration. These parameters accurately reflect the train's dynamic motion state at the moment of capture. Based on a pre-set motion parameter feature influence model, the motion parameters are converted into feature correction coefficients. This model learns the correlation between historical motion data and image feature changes to establish a quantitative mapping between motion parameters and feature deviations. This model then adjusts the feature parameters of the baseline change curve point by point based on the real-time motion parameters, achieving dynamic correction of the curve. The corrected curve then undergoes optimization. The corrected curve is compared with the motion-free environmental image features within a preset time period from a historical database. These motion-free environmental images are reliable samples verified through pre-calibration and reflect the recent state of environmental features. An iterative algorithm is used to optimize the curve parameters by calculating the deviation between the corrected curve and the sample features. During the iteration process, the correction amplitude is dynamically adjusted based on the deviation, gradually narrowing the difference between the curve and the sample features until the deviation falls within a preset range. This ultimately generates an environmental point processing curve that adapts to the current motion state.

[0035] Specifically, such as Figures 1 to 5 As shown, the motion parameter feature influence model is obtained by training the historical data of the train motion state and the corresponding image feature changes through a machine learning algorithm to obtain the mapping relationship between the motion parameters and the image feature changes, and the model input parameters and output parameters are determined. When the motion parameters of the train at the time of shooting are obtained, the motion parameter feature influence model is input to obtain the corresponding feature change correction coefficient, and the baseline change curve is corrected pixel by pixel according to the feature change correction coefficient to obtain the corrected environment point processing curve.

[0036] The construction of the motion parameter feature impact model begins with an in-depth analysis of historical data. This historical data covers train motion information under different operating conditions, as well as feature change data from images captured under these conditions. Motion state information includes parameters such as acceleration, instantaneous velocity, motion direction, and acceleration rate of change during operation. Image feature change data includes characteristic parameters such as pixel position offset, grayscale value change amplitude, and gradient direction deviation caused by motion offset. Training this historical data with a machine learning algorithm allows the algorithm to autonomously learn the inherent correlation between motion parameters and image feature changes, thereby establishing a quantitative mapping relationship between motion parameters and image feature changes. This mapping relationship is the core of the motion parameter feature impact model. During the model training process, the model's input and output parameters are clearly defined. The input parameters are selected as key parameters that fully reflect the train's dynamic motion state, including motion direction, instantaneous velocity, and acceleration rate of change. These parameters directly influence the degree of image feature offset and its variation pattern. The output parameter is set as a feature change correction coefficient, which quantifies the impact of motion parameters on image features and provides a precise adjustment basis for subsequent curve correction. After obtaining the real-time motion parameters of the train at the time of shooting, these parameters are input into the trained motion parameter feature impact model. The model calculates and processes the input motion parameters based on the preset mapping relationship and outputs the corresponding feature change correction coefficients. These correction coefficients are associated with the pixel features at different spatial positions and can specifically reflect the differences in the impact of motion parameters on the features of each pixel in the image. Based on the feature change correction coefficients, the baseline change curve is corrected pixel by pixel. For each pixel, the expected value of the feature parameter is adjusted according to its corresponding correction coefficient, so that the baseline change curve can adapt to the image feature change law under the current train motion state, and finally the corrected environment point processing curve is obtained.

[0037] Specifically, such as Figures 1 to 5 As shown, the fuzzy feature calibration strategy includes, for each pixel point in the pixel set to be calibrated, based on its deviation from the environmental point processing curve, the train motion parameters and the pixel features of the corresponding position in the ideal environmental image, calculating the pixel position compensation value, grayscale correction coefficient and edge sharpening parameter; performing coordinate offset compensation, grayscale value correction and edge contour enhancement processing on the pixel points according to the calibration parameters to generate calibrated pixel points; setting a spatial distance weight factor to assign different fusion weights to the calibrated pixel points and the surrounding non-calibrated pixel points, the closer the distance, the higher the weight ratio; using a weighted average algorithm to fuse the calibrated pixel points with the pixels of the corresponding area in the reference processed image, and performing smooth transition processing on the fusion boundary to generate a standard motion image.

[0038] The fuzzy feature calibration strategy calculates the calibration parameters for each pixel in the pixel set to be calibrated through multi-dimensional parameter analysis. The analysis process comprehensively considers the deviation between the pixel point and the environmental point processing curve, the motion parameters of the train at the time of shooting, and the pixel features of the corresponding position in the ideal environmental image. The deviation reflects the degree of deviation between the actual characteristics of the pixel and the expected characteristics. The motion parameters provide a reference for the dynamic motion state of the train, and the ideal environmental image features serve as a benchmark under the no-offset state. Based on these input information, three sets of key calibration parameters are calculated through a preset algorithm model: the pixel position compensation value is used to correct the coordinate deviation caused by motion offset, the grayscale correction coefficient is used to adjust the grayscale distortion caused by dynamic blur, and the edge sharpening parameter is used to enhance the pixel edge features weakened by motion smear. Targeted processing is performed on the pixel points according to the calculated calibration parameters. Coordinate offset compensation adjusts the offset pixels to their ideal spatial positions by superimposing a position compensation value on the actual pixel coordinates, restoring their correct distribution within the image. Grayscale correction linearly or nonlinearly adjusts the pixel grayscale values ​​based on a grayscale correction coefficient, bringing the corrected grayscale values ​​closer to the true grayscale in the undisturbed state. Edge enhancement enhances the grayscale gradients of the pixel neighborhood based on edge sharpening parameters, enhancing the contrast in edge regions and sharpening blurred contours. After these three processes, calibrated pixels are generated, whose features more closely resemble the ideal state without motion offset. To achieve a natural fusion of the calibrated pixels with the original image, a spatial distance weighting factor is defined. This factor dynamically assigns fusion weights based on the spatial distance between pixels, assigning different weights to the calibrated pixel and its surrounding uncalibrated pixels. Non-calibrated pixels closer to the calibrated pixel receive a higher weight to ensure a coherent feature transition in the fused area. Pixels farther away receive a lower weight to minimize interference with the core features of the calibrated area. A weighted average algorithm is used to fuse the calibrated pixels with the corresponding pixels in the baseline processed image. During the fusion process, the final value of each pixel is calculated by weighting the calibrated pixel value and the surrounding non-calibrated pixel values. This method preserves the precise features of the calibration while maintaining the regional correlation of the original image. Simultaneously, the boundaries of the fused areas are smoothed, and gradient adjustment techniques are used to reduce grayscale abrupt changes at the boundaries and eliminate splicing artifacts between different processed areas. After fusion and boundary processing, the output is a standard motion image with coherent overall features and clear details.

[0039] Specifically, such as Figures 1 to 5As shown, it also includes an environmental point processing curve calibration step. When the proportion of the pixel set to be calibrated exceeds a preset ratio threshold after feature fitting of several consecutive images, a preset number of ideal environmental images and motion parameters at corresponding moments are selected from the historical database as calibration samples; the current environmental point processing curve is applied to the calibration sample, and the overall deviation value between the actual characteristics of the environmental pixel points in the sample and the expected characteristics of the curve is calculated; based on the overall deviation value, the characteristic parameters of the environmental point processing curve are corrected by the least squares method, and the correction amplitude is positively correlated with the deviation value. The adjusted environmental point processing curve is applied to a new verification sample. If the feature fitting deviation of the verification sample is lower than the preset qualified threshold, the curve is updated; otherwise, the curve calibration is continued.

[0040] This step optimizes and adjusts the environmental point processing curve when curve accuracy degrades. Its activation mechanism is based on continuous monitoring of image feature fitting results. When the proportion of the pixel set to be calibrated relative to the total pixels in the environmental area after feature fitting for several consecutively acquired images exceeds a preset threshold, the system determines that the adaptability of the current environmental point processing curve has degraded, triggering the curve calibration process. The first step in curve calibration is to select calibration samples. A preset number of ideal environmental images are retrieved from the historical database. These images are rigorously verified to ensure no motion offset and reliable features. Train motion parameters corresponding to these images, including direction, instantaneous velocity, and acceleration rate, are extracted to form the calibration sample set. The calibration sample set must cover typical scenarios across different road sections and lighting conditions to ensure comprehensive calibration. The currently used environmental point processing curve is applied to the selected calibration samples. Feature comparison is performed to calculate the overall deviation between the actual characteristics of the environmental pixels in the sample and the expected characteristics of the curve. This overall deviation, calculated by integrating the feature deviations of all environmental pixels in the sample, comprehensively reflects the degree of deviation between the current curve and the actual environmental characteristics, providing a quantitative basis for subsequent corrections. Based on the calculated overall deviation, the least squares method is used to correct the characteristic parameters of the environmental point processing curve. This method constructs an optimization model that minimizes the sum of squared deviations to determine the parameter adjustment values ​​that best approximate the curve to the sample characteristics. During the correction process, the magnitude of the characteristic parameter adjustment is positively correlated with the overall deviation; that is, the larger the deviation, the larger the parameter adjustment, ensuring that the curve effectively compensates for the deviation. After correction, the adjusted environmental point processing curve is validated. The adjusted curve is applied to a new validation sample, which is an idealized environmental image and corresponding motion parameters that were not calibrated. The fitting deviation between the environmental pixel features in the validation sample and the expected curve features is calculated. If this deviation is below a preset acceptance threshold, the curve accuracy has been restored, and the environmental point processing curve in the system is updated. If the deviation does not reach the acceptance threshold, the system returns to the deviation calculation step and recalibrates the curve characteristic parameters until the verification deviation meets the required accuracy.

[0041] Specifically, such as Figures 1 to 5 As shown, it also includes a calibration effect verification step, setting verification indicators, including pixel deviation rate, feature matching and image clarity score, where the pixel deviation rate is the average value of the pixel position deviation between the calibrated image and the ideal environment image, and the feature matching is the overlap between the target subject feature and the standard feature in the historical database; the processed standard motion image is tested for indicators, and the actual value of each verification indicator is calculated; the actual value is compared with the preset qualified threshold, and if all meet the standards, the image is output; if not, the calibration step corresponding to the failed indicator is triggered to perform parameter backtracking adjustment, and the whole process from initial image calibration to image fusion is repeated until the verification indicators meet the standards.

[0042] This step quantitatively evaluates and optimizes the quality of the processed standard motion images. First, multi-dimensional verification metrics are set, including pixel deviation rate, feature matching, and image clarity score. The pixel deviation rate is calculated by averaging the positional deviations between all pixels in the calibrated image and the corresponding pixels in the ideal environment image, reflecting the accuracy of the spatial calibration of the overall image. The feature matching is determined by comparing the degree of overlap between the target subject's features and the standard features in the historical database, reflecting the accuracy of the restoration of key object features. The image clarity score is generated by analyzing the image's edge gradients and the richness of detailed textures, assessing the overall visual quality of the image. Metrics are then tested on the processed standard motion images. An image analysis algorithm is used to extract the actual coordinates of each pixel in the image and compare them with the corresponding pixel coordinates in the ideal environment image to calculate the pixel deviation rate. Feature matching techniques are used to match the target subject's feature parameters with the standard features in the historical database, and the percentage of overlapping features is calculated to determine the feature matching score. A clarity assessment model is used to quantitatively analyze the image's edge sharpness and texture detail to generate an image clarity score. Through these testing processes, the actual values ​​of each verification metric are obtained. The actual values ​​of each metric are then compared with the preset acceptance threshold. The qualified threshold is set based on the accuracy requirements of the rail transit image application scenario to ensure that the image quality meets the requirements of subsequent monitoring, identification, and other applications. If the actual values ​​of all indicators meet or exceed the qualified threshold, the image calibration effect meets expectations and the standard motion image is directly output. If any indicators do not meet the standards, the corresponding calibration steps are triggered according to the type of indicator that failed, and the parameters are retroactively adjusted. For example, if the pixel deviation rate does not meet the standard, the displacement compensation parameters in the initial image calibration are retroactively adjusted, and if the feature matching degree is insufficient, the feature fitting threshold in the fuzzy feature calibration is optimized. After the adjustment is completed, the entire process from initial image calibration to image fusion is re-executed until all verification indicators meet the qualified threshold.

[0043] The above shows and describes the basic features, principles, and advantages of the present invention. It should be noted that the present invention is not limited to the above embodiments, which are only some embodiments. Without departing from the spirit and scope of the present invention, various improvements and supplements made are considered to be within the scope of protection of the present invention.

Claims

1. A method for calibrating image features based on motion offset, characterized in that: The steps include: A motion image acquisition step is to acquire the motion image of the train while it is running and read the ideal environment image from the historical database; an initial image calibration step, obtaining the acceleration of the train at the time of shooting, determining the basic displacement deviation of the image based on the acceleration, and performing preliminary image calibration on the motion image to obtain a reference processed image; Fuzzy feature calibration step: identifying the target subject and the environmental area in the reference processed image based on the target features in the historical database, and extracting the environmental pixels in the environmental area; generating an environmental point processing curve containing expected features by combining the motion parameters of the train at the time of shooting and the ideal environmental image, extracting the actual features of the environmental pixels and performing feature-level fitting on the environmental point processing curve, and selecting pixels whose deviations exceed a preset threshold as the pixel set to be calibrated; The calibration image output step calibrates the pixel set to be calibrated by a fuzzy feature calibration strategy, and fuses the calibrated pixel set with the reference processed image to obtain a processed standard motion image.

2. The image feature calibration method based on motion offset according to claim 1, characterized in that: The initial image calibration step includes obtaining real-time acceleration data at the time of shooting through an acceleration sensor carried by the train, the real-time acceleration data including acceleration magnitude and direction; converting the real-time acceleration data into a basic displacement deviation of the image in the pixel dimension based on a preset acceleration-displacement conversion model, the basic displacement deviation including pixel offset values ​​in the horizontal and vertical directions; performing reverse compensation correction on the pixel coordinates of the motion image based on the basic displacement deviation, filling the pixel missing areas caused by displacement using neighborhood pixel interpolation, and outputting the image after preliminary calibration as the reference processed image.

3. The image feature calibration method based on motion offset according to claim 1, characterized in that: The fuzzy feature calibration step includes: based on the target features in the historical database, performing feature comparison on the benchmark processed image, identifying and separating the target subject and the environmental area in the image, extracting environmental pixel points from the environmental area, and the environmental pixel points are the pixel points of the non-target subject in the benchmark processed image; combining the motion parameters of the train at the time of shooting and the ideal environmental image, calculating and generating an environmental point processing curve through the environmental point processing strategy, the environmental point processing curve includes the expected spatial distribution characteristics and grayscale change characteristics of the environmental pixel points when there is no motion offset, the motion parameters include the motion direction and the acceleration change rate, extracting the actual characteristics of the environmental pixel points, the actual characteristics include the spatial position offset, the pixel grayscale attenuation rate and the gradient direction; performing feature-level fitting on the actual characteristics of the environmental pixel points and the expected characteristics of the environmental point processing curve, screening out the pixel points whose deviation between the actual characteristics and the expected characteristics exceeds a preset threshold, and outputting them as the pixel set to be calibrated.

4. The image feature calibration method based on motion offset according to claim 1, characterized in that: In the historical database, the ideal environmental image is an environmental image collected by the train under preset ideal working conditions and manually annotated and processed without motion offset calibration. The image acquisition device carried by the train performs periodic image acquisition when the train is in the preset ideal working conditions, and the acquisition frequency is not less than a preset threshold; the target features are obtained by manually annotating the collected ideal environmental images, extracting the characteristic parameters of key objects, establishing a feature model and storing it in the historical database, and updating and optimizing the feature model to obtain a target feature database.

5. The image feature calibration method based on motion offset according to claim 3, characterized in that: The environmental point processing curve is a multidimensional feature set that describes the characteristic change law of environmental pixel points in a state without motion offset, including a pixel spatial position distribution curve, a grayscale value attenuation curve and a gradient direction change curve. The environmental point processing curve extracts the characteristic parameters of all environmental pixel points in the environmental area based on the ideal environmental image in the historical database, establishes a baseline change curve for each characteristic parameter through statistical analysis, and obtains the motion parameters of the train at the time of shooting, including motion direction, instantaneous speed, and acceleration change rate. The baseline change curve is dynamically corrected based on a preset motion parameter feature influence model, and the corrected curve is compared with the environmental image features without motion offset within a preset time in the historical database. The deviation is calculated and the curve parameters are optimized through an iterative algorithm to generate the environmental point processing curve in the current motion state.

6. The image feature calibration method based on motion offset according to claim 5, characterized in that: The motion parameter feature influence model is obtained by training the historical data of the train motion state and the corresponding image feature changes through a machine learning algorithm to obtain the mapping relationship between the motion parameters and the image feature changes, and the model input parameters and output parameters are determined. When the motion parameters of the train at the moment of shooting are obtained, the motion parameter feature influence model is input to obtain the corresponding feature change correction coefficient, and the baseline change curve is corrected pixel by pixel according to the feature change correction coefficient to obtain the corrected environment point processing curve.

7. The image feature calibration method based on motion offset according to claim 1, characterized in that: The fuzzy feature calibration strategy includes calculating, for each pixel point in the pixel set to be calibrated, a pixel position compensation value, a grayscale correction coefficient, and an edge sharpening parameter based on the deviation between the pixel point and the environmental point processing curve, the train motion parameters, and the pixel features at the corresponding position in the ideal environmental image; performing coordinate offset compensation, grayscale value correction, and edge contour enhancement on the pixel point according to the calibration parameters to generate a calibrated pixel point; The spatial distance weight factor is set to assign different fusion weights to the calibrated pixels and the surrounding non-calibrated pixels. The closer the distance, the higher the weight. The calibrated pixels are fused with the pixels in the corresponding area of ​​the benchmark processed image through the weighted average algorithm, and the fusion boundary is smoothly transitioned to generate a standard motion image.

8. The image feature calibration method based on motion offset according to claim 1, characterized in that: It also includes an environmental point processing curve calibration step. When the proportion of the pixel set to be calibrated exceeds a preset ratio threshold after feature fitting of several consecutive images, a preset number of ideal environmental images and motion parameters at corresponding moments are selected from the historical database as calibration samples; the current environmental point processing curve is applied to the calibration sample, and the overall deviation value between the actual characteristics of the environmental pixel points in the sample and the expected characteristics of the curve is calculated; based on the overall deviation value, the characteristic parameters of the environmental point processing curve are corrected by the least squares method, and the correction amplitude is positively correlated with the deviation value. The adjusted environmental point processing curve is applied to a new verification sample. If the feature fitting deviation of the verification sample is lower than the preset qualified threshold, the curve is updated; otherwise, the curve calibration continues.

9. The image feature calibration method based on motion offset according to claim 1, characterized in that: It also includes a calibration effect verification step, setting verification indicators, including pixel deviation rate, feature matching and image clarity score, where the pixel deviation rate is the average value of the pixel position deviation between the calibrated image and the ideal environment image, and the feature matching is the degree of overlap between the target subject feature and the standard feature in the historical database; the processed standard motion image is subjected to indicator detection, and the actual value of each verification indicator is calculated; the actual value is compared with the preset qualified threshold, and the image is output if all meet the standards; if the standards are not met, the calibration step corresponding to the failed indicator is triggered to perform parameter backtracking adjustment, and the whole process from initial image calibration to image fusion is repeated until the verification indicators meet the standards.

Images