A machine vision-based precision component size online measurement method

Abstract

This invention relates to the field of dimensional measurement technology and discloses an online measurement method for precision parts based on machine vision, comprising the following steps: Step S1, acquiring a background pattern reference image; Step S2, acquiring a contour image of the workpiece; Step S3, establishing a normal sampling zone; Step S4, calculating the liquid layer optical load; Step S5, calculating the thermo-optical load; Step S6, reconstructing the solid boundary by inferring the shoulder pseudo-boundary from the normal migration along the normal vector; Step S7, outputting the dimensional value. This invention can convert the coupling interference between the liquid layer and the thermal field into quantifiable optical parameters and complete boundary correction, allowing the measurement process to adapt to the structural characteristics of the shoulder chamfer composite boundary. The measurement process requires no additional hardware, and the output dimensional results closely match the true geometric state of the workpiece, improving the matching degree between the measurement results and the actual dimensions of the workpiece.

Description

Technical Field

[0001] This invention relates to the field of dimensional measurement technology, and more specifically, to a method for online measurement of the dimensions of precision parts based on machine vision. Background Technology

[0002] Precision shaft components are core foundational parts in high-end equipment manufacturing, and their dimensional accuracy directly determines the assembly quality and operational reliability of the equipment. Online machine vision measurement, with its advantages of being non-contact, efficient, and stable, has become the mainstream technology for dimensional inspection of shaft components. Optical dimensional measurement, as a core technology in this field, is widely used for the axial cross-sectional projection contour inspection of rotating components. Dual telecentric imaging systems, with their low distortion and constant magnification characteristics, have become the mainstream imaging solution for online measurement of key dimensions such as the outer diameter and shoulder distance of multi-segment shaft components, including stepped shafts.

[0003] Current online vision measurement technologies focus only on geometric error processing such as geometric edge extraction and lens distortion correction, without addressing the structural characteristics of components like shoulder chamfers and multi-scale composite boundaries. In actual online production, residual oil films and coolant on the workpiece surface can create unilateral retention at the boundary roots; the uneven air refractive index caused by residual heat from the workpiece can create an asymmetric optical path deflection field near the boundary. Both types of interference are distributed along the boundary normal and exhibit unilateral bias characteristics, making them difficult to eliminate using conventional image filtering or brightness adjustment methods.

[0004] Existing measurement methods struggle to quantify the peaking characteristics of the unilateral normal refraction moment formed by the coupling of the liquid layer and the thermal field, resulting in the vision system consistently locking onto a misaligned, pseudo-boundary. This issue directly causes measurement results for critical dimensional benchmarks such as the root of the shoulder chamfer to continuously deviate from their true values, and the measurement error cannot be compensated for by existing correction methods. Consequently, it is difficult to meet the accuracy requirements of online full inspection of precision shaft components, severely restricting the efficiency of quality control and the reliability of inspection in the manufacturing process of precision shaft parts. Summary of the Invention

[0005] This invention provides a method for online measurement of the dimensions of precision parts based on machine vision, which solves the technical problems mentioned in the background.

[0006] This invention provides a method for online measurement of the dimensions of precision parts based on machine vision, comprising the following steps: Step S1: Obtain the geometric magnification and the infrared reference image in the liquid-free state and the background pattern reference image in the thermal-free state; Step S2: Simultaneously acquire the workpiece's contour image, infrared image, and background pattern image; Step S3: Extract the shoulder pseudo-boundary and its normal vector from the contour map, and extend the normal vector to both sides to establish a normal sampling zone; Step S4: Using the infrared reference image and infrared image in the liquid-free state, calculate the equivalent liquid layer thickness in the normal sampling band and convert it into liquid layer optical load. Step S5: Obtain the apparent displacement using the background pattern reference image and the background pattern image under the condition of no thermal disturbance; extract the normal displacement component along the normal vector of the apparent displacement and deduce the normal deflection angle; integrate the normal deflection angle to obtain the thermo-optical load. Step S6: Combine the liquid layer optical load and the thermo-optical load into a composite optical load and integrate to obtain the zero-order load. Calculate the one-sided normal refraction moment of the composite optical load. Divide the one-sided normal refraction moment by the zero-order load to obtain the normal migration amount. Use the normal migration amount to inversely deduce the shoulder pseudo-boundary along the normal vector to restore the solid boundary. Step S7: Combine geometric magnification to transform the entity boundary to physical coordinates and perform piecewise straight line fitting, and calculate the output dimension value based on the fitting result.

[0007] The beneficial effects of this invention are as follows: By establishing three types of reference standards—geometry, liquid layer, and thermal field—it provides a unified basis for comparison during the measurement process. Employing a multi-channel synchronous acquisition method, it ensures that the three types of data—contour, liquid layer, and thermal field—correspond to the same workpiece state. By relying on the normal sampling band, the analysis range is limited to the boundary normal region, simplifying the analysis dimension of multi-physics interference. The optical loads formed by the liquid layer and thermal field are quantified separately, and the solid boundary of the workpiece is reconstructed through the calculation of the composite optical load and the unilateral normal refraction moment. After converting the solid boundary into physical coordinates, piecewise fitting and dimensional calculation are performed, ensuring that the measurement results are based on the true geometric contour of the workpiece. This invention can transform the coupling interference of the liquid layer and thermal field into quantifiable optical parameters and complete boundary correction, allowing the measurement process to adapt to the structural characteristics of the composite boundary of the shoulder chamfer. The measurement process requires no additional hardware and can be directly integrated into existing online vision measurement systems. The final output dimensional results closely match the true geometric state of the workpiece, meeting the metrological requirements for online dimensional measurement of precision shaft components and improving the matching degree between the measurement results and the actual dimensions of the workpiece. Attached Figure Description

[0008] Figure 1 This is a flowchart of a machine vision-based online measurement method for precision parts dimensions according to the present invention; Figure 2 This is a schematic diagram illustrating an application scenario of the online measurement method for precision parts dimensions based on machine vision, according to the present invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0011] like Figures 1-2 As shown, a method for online measurement of precision parts dimensions based on machine vision includes the following steps: Step S1: Obtain the geometric magnification and the infrared reference image in the liquid-free state and the background pattern reference image in the thermal-free state; Step S2: Simultaneously acquire the workpiece's contour image, infrared image, and background pattern image; Step S3: Extract the shoulder pseudo-boundary and its normal vector from the contour map, and extend the normal vector to both sides to establish a normal sampling zone; Step S4: Using the infrared reference image and infrared image in the liquid-free state, calculate the equivalent liquid layer thickness in the normal sampling band and convert it into liquid layer optical load. Step S5: Obtain the apparent displacement using the background pattern reference image and the background pattern image under the condition of no thermal disturbance; extract the normal displacement component along the normal vector of the apparent displacement and deduce the normal deflection angle; integrate the normal deflection angle to obtain the thermo-optical load. Step S6: Combine the liquid layer optical load and the thermo-optical load into a composite optical load and integrate to obtain the zero-order load. Calculate the one-sided normal refraction moment of the composite optical load. Divide the one-sided normal refraction moment by the zero-order load to obtain the normal migration amount. Use the normal migration amount to inversely deduce the shoulder pseudo-boundary along the normal vector to restore the solid boundary. Step S7: Combine geometric magnification to transform the entity boundary to physical coordinates and perform piecewise straight line fitting, and calculate the output dimension value based on the fitting result.

[0012] In one embodiment of the present invention, obtaining geometric magnification and an infrared reference image in a liquid-free state and a background pattern reference image in a thermally undisturbed state includes: Will have known physical dimensions The standard part is placed at the measurement position, and its contour image is obtained. Known physical dimensions are then extracted from the contour image of the standard part. Corresponding pixel span According to the formula Calculate the geometric magnification ,in Geometric magnification, For pixel span, Given physical dimensions; When the measurement area is in a liquid-free state, an infrared reference image in this liquid-free state is acquired using the infrared imaging channel. ,in This is an infrared reference image in a liquid-free state. Image coordinates; When the measurement area is in a state of no thermal disturbance, a background pattern reference image under this state is acquired using the background pattern imaging channel. ,in This is a reference image of the background pattern under conditions without thermal disturbance. This represents the image coordinates.

[0013] It should be noted that the measurement position is the unified spatial placement of the standard part and the workpiece under test in the measurement system. The preferred value is the position completely consistent with the position during the actual online measurement of the workpiece. The geometric magnification must share the same imaging ratio reference with subsequent physical conversion of the physical boundary. The standard part contour image is the boundary observation image formed by the standard part in the contour imaging channel, which can be acquired using a dual telecentric contour camera, a stable backlight, and a synchronously triggered acquisition module. The known physical dimension is the true length reference on the standard part with metrological traceability. The pixel span is the pixel distance of the known physical dimension in the standard part contour image. The geometric magnification is the proportional coefficient between the image pixel length and the true physical length. The measurement area is the effective working area jointly covered by contour imaging, infrared imaging, and background pattern imaging. The preferred value is the area that completely covers the shoulder pseudo-boundary and the normal disturbance areas on both sides. The liquid layer optical load and thermo-optical load must be solved correspondingly within the same spatial area. The liquid-free infrared reference image is the infrared reference image obtained when there is no residual liquid film or cooling droplets in the measurement area. It can be obtained by continuously acquiring multiple frames with an infrared camera after surface cleaning and drying and averaging them. The background pattern reference image under thermal disturbance-free conditions is a baseline image of the background pattern obtained when there is no thermal boundary layer disturbance in the measurement area. It can be obtained by continuously acquiring multiple frames and averaging them after the workpiece temperature has stabilized and the airflow field has returned to a stable state using a background pattern imaging camera. The horizontal coordinate of the image is the pixel position index arranged horizontally in the image plane. The vertical coordinate of the image is the pixel position index arranged vertically in the image plane.

[0014] It's important to note that geometric magnification is not an abstract camera intrinsic parameter, but rather a length ratio reference directly used when converting the entity boundary from pixel coordinates to physical coordinates. Therefore, the standard part must be placed at the same measurement position as the actual workpiece measurement. The magnification must be directly established using the pixel span and actual dimensions at that position to avoid proportional errors introduced by pose changes, depth of field changes, and working distance changes. Infrared images simultaneously contain the inherent radiation characteristics, surface reflection characteristics, and liquid absorption characteristics of the measured surface. Therefore, only by first acquiring an infrared reference image in a liquid-free state can the basic response of the exposed metal surface be fixed, allowing subsequent intensity ratios to primarily reflect additional absorption by the liquid layer rather than material background differences. Furthermore, the original position of the background pattern in the camera is itself affected by lens distortion, installation errors, and the geometric distribution of the background pattern. Therefore, a background pattern reference image in a thermally undisturbed state must be obtained first to interpret subsequent pattern displacement as additional apparent displacement caused by thermal boundary layer refraction, rather than static imaging deviation.

[0015] It should be noted that the standard parts should be block gauges, step gauges, or cylindrical step standard parts with metrological verification certificates or calibration certificates. The selected dimension should be consistent with the order of magnitude of the dimension to be measured. Standard parts with similar geometry to the shoulder area of ​​the workpiece should be preferred. When arranging the standard parts, the calibration boundary of the standard parts should be located at the center of the formal measurement area, and its posture, working distance, and optical path obstruction conditions should be consistent with the workpiece to be measured. The method to ensure consistency between the measurement position and the actual measurement posture is to set up a mechanical positioning reference, limit block, or precision fixture at the workstation, and to form a unified workstation coordinate system with the camera optical axis, backlight center, and background pattern center. Standard part calibration and workpiece measurement are both completed under the same fixture reference, and the fixture repeatability error is preferably controlled within one-tenth of the pixel equivalent. The method for determining the liquid-free state is to first clean, dry, and allow the measurement area to stand still, and then continuously acquire five to twenty frames of infrared images, calculating the average gray level and gray level variance of the target sampling area. When the average gray level change rate is less than one percent and the variance is stable within a preset threshold, it is determined to be in a liquid-free state, and the average result of multiple frames is used as the infrared reference image. The method for determining the thermal disturbance-free state is to continuously acquire five to twenty frames of background pattern images after the workpiece has not entered the measurement area or the workpiece temperature has reached a stable equilibrium with the ambient temperature, and calculate the average apparent displacement between adjacent frames. When the average displacement is less than 0.05 pixels and is maintained for a set time, it is determined to be in a thermal disturbance-free state, and the average result of multiple frames is used as the background pattern reference image. In addition, the exposure time, gain, lens aperture and light source power should be fixed during the acquisition of the reference image and kept consistent with the actual measurement. After acquisition, dark field subtraction and flat field correction should be performed first, and then subpixel registration and mean fusion should be performed on multiple frames of images to finally output the reference image.

[0016] In one embodiment of the present invention, the simultaneous acquisition of the workpiece's contour image, infrared image, and background pattern image includes: Place the workpiece within the measurement area and set the current synchronous acquisition time to [value]. and make the acquisition time of the contour image Infrared image acquisition time and the acquisition time of the background pattern image Satisfying Relationship: ; in The time of acquisition of the contour image. The moment the infrared image was acquired. The time of acquisition of the background pattern image. This is the current synchronous data collection time; At the current synchronous acquisition time Below, the contour image of the workpiece is acquired using the contour imaging channel and recorded as... ; in This is the contour image obtained during the current synchronous acquisition. For at any time The intensity distribution of the contour image, The image coordinates are the location of the image in the contour image; At the current synchronous acquisition time Below, infrared images of the workpiece are acquired using the infrared imaging channel and recorded as follows: ; in This is the infrared image currently being acquired synchronously. For at any time Infrared image intensity distribution, The coordinates of the image in the infrared image; At the current synchronous acquisition time Below, the background pattern image corresponding to the workpiece is acquired using the background pattern imaging channel and recorded as... ; in This is the background pattern image currently being acquired synchronously. For at any time Background pattern image intensity distribution The coordinates of the image within the background pattern image.

[0017] It should be noted that the current synchronous acquisition time is a unified time reference shared by the three-channel images. The preferred value is the unique sampling time defined by the same hardware trigger pulse. The three-channel observations must correspond to the same physical state to ensure that liquid layer effects and thermal effects can be superimposed at the same sampling point. The contour image acquisition time is the moment when the contour imaging channel completes effective exposure. The infrared image acquisition time is the moment when the infrared imaging channel completes effective exposure. The background pattern image acquisition time is the moment when the background pattern imaging channel completes effective exposure. The currently synchronously acquired contour image is the workpiece boundary image obtained by the contour imaging channel at the current synchronous sampling time, which can be obtained through dual telecentric contour cameras in conjunction with unified hardware triggering. The contour image intensity distribution is the spatial distribution of grayscale or brightness at each pixel position in the contour image, which can be obtained by reading the original image of the contour imaging channel pixel by pixel. The currently synchronously acquired infrared image is the infrared observation image of the workpiece obtained by the infrared imaging channel at the current synchronous sampling time, which can be acquired by an infrared camera under unified triggering. The infrared image intensity distribution is the spatial distribution of infrared intensity at each pixel position in the infrared image, which can be obtained by reading the original image of the infrared imaging channel pixel by pixel. The currently synchronously acquired background pattern image is the observed image of the workpiece background pattern obtained by the background pattern imaging channel at the current synchronous sampling moment. It can be acquired by the background pattern camera under unified triggering. The background pattern image intensity distribution is the brightness distribution of each pixel position in the background pattern image, which can be obtained by reading the original image of the background pattern imaging channel pixel by pixel. The synchronous acquisition sequence number is the serial number that identifies the current sampling batch and is used to associate the three-channel images under the same synchronous sampling.

[0018] It should be noted that the pseudo-boundary is the result of the combined effect of the residual liquid layer and the thermal boundary layer at the same physical moment. Therefore, the three-channel observations must be locked at the same sampling moment. Only in this way can there be a one-to-one correspondence between the boundary position in the contour image, the liquid layer distribution in the infrared image, and the thermally induced deflection in the background pattern image. Furthermore, a single contour image can only provide the visual boundary after disturbance and cannot distinguish between the contributions of the liquid layer and the thermal field. Therefore, the contour image, infrared image, and background pattern image must be used as inputs simultaneously to obtain the liquid layer optical load and thermo-optical load separately, and further reconstruct the solid boundary. A unified hardware trigger controller is used to simultaneously send trigger pulses to the contour camera, infrared camera, and background pattern camera, and the trigger timestamp is recorded using the same clock source. When the workpiece moves continuously, the encoder position signal can be further superimposed as a trigger condition to ensure synchronization between spatial and temporal positions. The three-channel spatial registration method involves placing a composite calibration target containing contour features, infrared response features, and background pattern features at the same workstation, and calculating the mapping relationship from the three channels to the coordinates of the unified workstation. Before formal measurement, the three-channel images are resampled to the same reference plane, and bilinear or cubic interpolation is used to ensure the consistency of corresponding pixels. The exposure control and gain consistency method during synchronous acquisition involves presetting the dynamic range of each of the three channels before formal measurement, and locking the exposure time, gain, and light source power within a range that does not saturate and has sufficient contrast. During continuous measurement, automatic exposure and automatic gain changes are not allowed unless a recalibration process is initiated. The generation and cycle control method for the current synchronous acquisition moment are as follows: when the workpiece enters the measurement window, the upper controller generates the current synchronous acquisition moment based on the encoder position, workstation sensor signal, or time cycle, and writes this moment into the three-channel data packet header. When the workpiece is running at a constant speed, equal displacement triggering is preferred over simple equal time triggering.

[0019] In one embodiment of the present invention, the process of extracting the shoulder pseudo-boundary and its normal vector from the contour map, and extending the normal vector to both sides to establish a normal sampling band includes: Based on contour images Extract the boundary points corresponding to the neighborhood of the workpiece shoulder, and represent the shoulder pseudo-boundary as a parametric curve. ,in For the boundary point on the pseudo-boundary of the shoulder, Here is the arc length parameter along the extension direction. The horizontal axis is... The vertical axis; From the pseudo-boundary of the shoulder Extract the set of straight boundary points corresponding to the workpiece generatrix. According to the following formula: And the constraints are satisfied. Perform a linear fit, where For the set of boundary points of a straight line, These are the boundary points in the set of boundary points for a straight line. , , To fit the parameters of the straight line, the constraints are... Used to normalize the normal parameters of the fitted line; The fitted straight line of the busbar is obtained based on the straight line fitting results. And determine the normal vector of the pseudo-boundary of the shoulder. ,in Fit a straight line to the generatrix. It is the normal vector; Boundary points on the pseudo-boundary of the shoulder Using the base point as an example, along the normal vector The sampling point locations are obtained by extending to both sides. ,in The location of the sampling point. For the pseudo-boundary relative to the shoulder The normal offset; Based on the location of the sampling point Constructing normal sampling bands ,in For normal sampling band, and This constitutes the sampling arc length range. and Construct along the normal vector Normal sampling range on both sides The normal sampling half-width.

[0020] It should be noted that the shoulder neighborhood is a local analysis region selected around the shoulder transition boundary. The preferred value is a region covering 20 to 100 pixels before and after the shoulder transition center, simultaneously including the stable generatrix segment and the root transition segment. That is, normal vector fitting requires the straight generatrix segment, and normal sampling needs to cover the disturbance area near the transition. The shoulder pseudo-boundary is the boundary curve in the image located within the shoulder neighborhood, initially locked by the contour algorithm, but not necessarily equal to the real metal boundary. The arc length parameter is the curve position parameter defined along the extension direction of the shoulder pseudo-boundary. The lateral coordinate function is the position function of the shoulder pseudo-boundary in the horizontal direction of the image. The longitudinal coordinate function is the position function of the shoulder pseudo-boundary in the vertical direction of the image. The straight boundary point set is the set of boundary points selected from the shoulder pseudo-boundary to represent the straight edge of the workpiece generatrix. A boundary point is a single image point on the shoulder pseudo-boundary or generatrix boundary. The generatrix fitting line is a reference straight line representing the direction of the workpiece generatrix, obtained by fitting the straight boundary point set. The normal vector is a direction vector perpendicular to the generatrix fitting line and used to establish the boundary normal sampling direction. The sampling point location is the spatial position obtained by offsetting the base point on the shoulder pseudo-boundary along the normal direction. The normal offset is the distance of the sampling point relative to the shoulder pseudo-boundary along the normal vector direction. The normal sampling zone is a strip-shaped sampling area jointly formed by the arc length direction along the shoulder pseudo-boundary and the normal directions on both sides. The starting point of the sampling arc length is the starting position of the normal sampling zone along the shoulder pseudo-boundary direction, preferably a position covering the arc length corresponding to one to three times the nominal shoulder height in front of the shoulder turning center, i.e., it must cover the outer open side disturbance initiation area. The ending point of the sampling arc length is the ending position of the normal sampling zone along the shoulder pseudo-boundary direction, preferably a position covering the arc length corresponding to one to three times the nominal shoulder height behind the shoulder turning center, i.e., it must cover the root inner side single-sided liquid accumulation and main thermal disturbance area. The normal sampling half-width is the half-width of the normal sampling band extending from the pseudo-boundary to both sides. It is preferably set to five to thirty pixels, or converted to two to five times the expected maximum composite optical disturbance thickness. It should fully cover the residual liquid and thermal boundary layer, while avoiding excessive bandwidth that introduces irrelevant background.

[0021] It should be noted that the shoulders, chamfer roots, and edge transition areas in the online workstation are most prone to unilateral liquid accumulation and thermally induced refractive shifts. Therefore, the position initially locked by the edge algorithm in the contour map is not necessarily equal to the real metal boundary. It should be defined as the pseudo-boundary of the shoulder to provide a clear physical interpretation space for subsequent boundary correction and solid reconstruction. The straight edge of the workpiece generatrix is ​​usually the most stable directional element in the local area of ​​the shoulder, closest to the real geometric reference. Therefore, the reference straight line is first fitted with the generatrix boundary point set, and then its normalized normal parameter is taken as the pseudo-boundary normal vector, which ensures that the subsequent normal sampling direction is consistent with the real dimension definition direction. The influence of residual liquid layer and thermal boundary layer on the visual boundary is mainly manifested as an optical distribution offset on both sides of the boundary normal, rather than a uniform change along the boundary tangential direction. Therefore, it is necessary to establish a normal sampling band around the pseudo-boundary to both sides of the normal to capture the unilateral load distribution that truly causes boundary migration. Furthermore, the pseudo-boundary of the shoulder is usually not a fixed straight line that is strictly horizontal or vertical, but may have local curvature, tilt and turn. Therefore, the position along the boundary is described by the arc length parameter, and the position on both sides of the boundary is described by the normal offset, so as to uniformly express the positional relationship of local sampling points under different boundary morphologies.

[0022] It should be noted that the shoulder neighborhood extraction method involves first performing global edge extraction in the contour image, then using line segment detection to obtain candidate segments for the busbar and turning points. Centered on the turning point, a set length is extended along both the busbar and shoulder directions to form a local analysis window containing line segments and turning points, which is defined as the shoulder neighborhood. The selection method for the straight boundary point set involves first calculating the local curvature or local direction change rate of each boundary point within the shoulder neighborhood, then retaining boundary points with curvature below a threshold and stable direction as candidate busbar points. Subsequently, random consistency filtering or iterative residual elimination is used to remove burrs and isolated points, resulting in the final straight boundary point set. The determination of the positive and negative directions of the normal vector involves first fixing the positive direction of the normal vector to point towards the root of the shoulder or the inner side of the chamfer, and the negative direction to point towards the open air. During system initialization, the positive direction can be automatically determined based on the topological relationship of the shoulder structure in the image, or it can be manually specified and written into the configuration parameters by the implementer during the calibration phase. The sampling arc length range and half-width setting method for the normal sampling band is as follows: first, estimate the maximum diffusion range of the single-sided liquid-holding area and thermal disturbance area based on the shoulder inflection point position; then, make the arc length range cover one to three times the feature size before and after the inflection point, and make the normal half-width cover two to five times the expected disturbance thickness; if the disturbance range is unknown, coarse sampling can be performed first, and then adaptively shrunk according to the load attenuation result. The discrete resampling and interpolation method for the sampling point position is to uniformly discretize along the arc length direction with a step size of one to three pixels, and uniformly discretize along the normal direction with a step size of 0.25 pixels to one pixel. For sampling points falling at non-integer pixel positions, bilinear interpolation or cubic convolution interpolation is used to read the image intensity.

[0023] In one embodiment of the present invention, the equivalent liquid layer thickness within the normal sampling band is calculated and converted into a liquid layer optical load using an infrared reference image and an infrared image in a liquid-free state, including: In the normal sampling band Sampling points within Read the corresponding infrared reference intensity in the liquid-free state and current infrared intensity : ; ; in For the infrared reference intensity in the liquid-free state, Given the current infrared intensity, Normal sampling band Sampling points within, For the pseudo boundary along the shoulder of the platform The arc length parameter, For the pseudo-boundary relative to the shoulder The normal offset; Based on infrared reference intensity in the liquid-free state and current infrared intensity According to the following formula: Calculate the normal sampling band Equivalent liquid layer thickness at each sampling point ,in The equivalent liquid layer thickness, The absorption coefficient of the liquid medium in the infrared imaging band. For natural logarithm operations; Based on the equivalent liquid layer thickness According to the following formula: Equivalent liquid layer thickness Converted into liquid layer optical payload ,in For liquid layer optical payload, The refractive index of the liquid medium, is the refractive index of air.

[0024] It should be noted that the infrared reference intensity in the liquid-free state is the infrared intensity value of each sampling point within the normal sampling band on the liquid-free state reference image, which can be obtained by reading the corresponding pixel of the infrared reference image on the normal sampling band coordinates. The current infrared intensity is the infrared intensity value of each sampling point within the normal sampling band on the current infrared image, which can be obtained by reading the corresponding pixel of the current infrared image on the normal sampling band coordinates. The equivalent liquid layer thickness is the effective thickness of the liquid layer obtained by recalculating the residual liquid film or cooling droplets along the current line of sight. The absorption coefficient is a parameter representing the ability of the liquid medium to attenuate infrared energy in the selected infrared imaging band. The liquid layer optical load is a liquid optical path influence quantity characterized by the refractive index difference of the liquid medium and the equivalent liquid layer thickness. The refractive index of the liquid medium is the refractive index of the medium corresponding to the residual liquid film or coolant. The air refractive index is the refractive index of the air medium in the current measurement environment.

[0025] It should be noted that the attenuation of infrared energy in a liquid medium within a selected infrared band typically follows an exponential absorption relationship. Therefore, by taking the natural logarithm of the ratio of the reference intensity to the current intensity, the exponential attenuation relationship can be transformed into a linear inversion relationship of thickness, thus quantitatively converting the image intensity difference into liquid layer thickness. Under reflective infrared observation conditions, the imaging optical path usually needs to pass through the liquid layer to reach the workpiece surface, and then, after reflection, pass through the liquid layer a second time to return to the detector. Therefore, the actual effective absorption path is twice that of a single-path path, and using twice the absorption coefficient can reflect this two-way absorption fact. The influence of the liquid layer on boundary imaging depends not only on the thickness of the liquid layer itself, but also on the refractive index difference between the liquid medium and air. Therefore, only by combining the thickness and the refractive index difference can a truly meaningful liquid layer optical load be obtained.

[0026] It should be noted that the selection of infrared imaging bands and the calibration method for absorption coefficients prioritize narrow or mid-band infrared bands where the liquid medium exhibits a significant absorption peak and the workpiece substrate has a stable reflection response. The absorption coefficient should be calibrated using standard liquid layer samples of known thickness, and obtained by fitting the intensity-thickness relationship using multi-thickness samples. The preprocessing method for infrared reference intensity and current infrared intensity involves performing bad pixel replacement, dark field subtraction, flat field correction, and mild noise reduction before reading the intensity of the normal sampling point. Then, the reference image and the current image are resampled to the normal sampling band coordinates under the same spatial registration result, and finally, the intensity ratio is calculated. The handling method for abnormal intensity ratios is as follows: when the current infrared intensity is close to zero, the pixel is saturated, or the intensity ratio is less than zero, the point should be marked as invalid. Isolated invalid points can be repaired using neighborhood interpolation; continuous invalid areas should be directly removed from the current arc length position to avoid distortion in the natural logarithm calculation. The method for obtaining the refractive index of the liquid medium and correcting for temperature is to first look up the basic refractive index according to the type of liquid medium, and then establish a temperature correction curve in combination with the on-site temperature. When the liquid temperature changes significantly, the measured liquid temperature should be input into the correction model before participating in the calculation of the optical load of the liquid layer.

[0027] In one embodiment of the present invention, apparent displacement is obtained using a background pattern reference image and a background pattern image under no thermal disturbance. The normal displacement component along the normal vector of the apparent displacement is extracted and the normal deflection angle is derived. The thermo-optical load is obtained by integrating the normal deflection angle, including: In the normal sampling band Sampling points within Read the reference intensity of the background pattern in the corresponding thermal disturbance-free state. and the intensity of the current background pattern : ; ; in The background pattern reference intensity is for a state without thermal disturbance. The intensity of the current background pattern. Normal sampling band Sampling points within, For the pseudo boundary along the shoulder of the platform The arc length parameter, For the pseudo-boundary relative to the shoulder The normal offset; Obtain sampling points Apparent displacement at point ,in For apparent displacement, This represents the displacement component in the horizontal direction of the image. This represents the displacement component in the vertical direction of the image; According to the following formula: Extract sampling points Normal displacement component at the location ,in For the normal displacement component, This refers to the dot product operation of vectors. It is the normal vector; According to the following formula:

[0028] Calculate the normal deflection angle ,in The normal deflection angle, The equivalent focal length for the contour imaging channel; According to the following formula: Obtain thermo-optical load ,in For thermo-optical loads, The refractive index of air, For the arc length parameter The position of the normal direction is Normal deflection angle at time For integration variables, Normal sampling band Along the normal vector The normal sampling range.

[0029] It should be noted that the reference intensity of the background pattern in the absence of thermal disturbance is the brightness value of each sampling point within the normal sampling band on the reference image of the background pattern in the absence of thermal disturbance, which can be obtained by reading the corresponding pixel of the background pattern reference image on the normal sampling band coordinates. The current background pattern intensity is the brightness value of each sampling point within the normal sampling band on the current background pattern image, which can be obtained by reading the corresponding pixel of the current background pattern image on the normal sampling band coordinates. Apparent displacement is the apparent position offset of the background pattern in the current image relative to the reference image in the absence of thermal disturbance. The lateral displacement component is the component of the apparent displacement in the horizontal direction of the image. The longitudinal displacement component is the component of the apparent displacement in the vertical direction of the image. The normal displacement component is the projection component of the apparent displacement in the direction of the shoulder pseudo-boundary. The equivalent focal length is the equivalent optical parameter of the imaging channel that converts the apparent displacement of the background pattern into the normal deflection angle of the light rays. The normal deflection angle is the local deflection angle of the light rays in the direction of the normal of the shoulder pseudo-boundary. Thermo-optical load is the amount of thermal field optical influence obtained by accumulating the normal deflection of light rays caused by the thermal boundary layer along the normal direction. The integral variable is the intermediate position variable used when performing continuous integration along the normal direction.

[0030] It's important to note that the thermal boundary layer causes the deflection of light rays after passing through the air, and doesn't directly provide a temperature value. Therefore, a more direct observation is the change in the apparent position of the background pattern relative to a thermally undisturbed reference image. By matching the local intensity distribution to obtain the apparent displacement, the refraction effect can be transformed into a calculable image displacement. Furthermore, not all displacement directions change the normal position of the dimensional reference boundary. The component of displacement along the normal direction truly affects the boundary coordinates. Therefore, it's necessary to project the two-dimensional apparent displacement onto the normal vector, retaining only the portion that drives the boundary normal migration. Under small-angle imaging conditions, there is an approximately linear relationship between image displacement and the incident ray deflection angle. Therefore, by dividing the normal displacement component by the equivalent focal length, the displacement of the image plane can be converted back into the more physically meaningful quantity of the ray normal deflection angle. The thermal boundary layer doesn't act at a single point but forms a continuously distributed refractive index gradient along the normal direction. Therefore, it's necessary to accumulate the local deflection angles at each normal position along the normal direction to obtain a superimposed thermo-optical load similar to the liquid layer optical load.

[0031] It should be noted that the background pattern should preferably use a high-contrast random speckle pattern or pseudo-random dot matrix pattern, with the pattern unit size preferably three to eight times the imaging pixel size. The background pattern should be placed on a stable plane behind the workpiece, ensuring that it covers the entire projection area of ​​the normal sampling band in the field of view. The specific algorithm for obtaining the apparent displacement by matching the local intensity distribution involves establishing a local matching window centered on each normal sampling point, and using normalized cross-correlation, phase correlation, or least squares optical flow methods to calculate the displacement between the reference image and the current image. To achieve sub-pixel accuracy, the correlation peaks should be parabolically or Gaussianly refined. The equivalent focal length is calibrated by placing a calibration target of known size in the contour imaging channel, calibrating the equivalent focal length through the relationship between the known displacement and the imaging displacement, or by converting it based on the camera pixel size, lens magnification, and working distance. After calibration, it should be verified at the actual workstation position. The discrete implementation and boundary conditions of the normal deflection angle integral are achieved by using trapezoidal integral or Simpson integral to integrate the normal deflection angle at discrete sampling points along the normal direction. The lower boundary of the integral is fixed at the position of the open side boundary of the normal direction, and the upper boundary is the current normal offset position. When there are no valid sampling points at the boundary, the nearest valid point is used to extrapolate and fill the edge.

[0032] In one embodiment of the present invention, the liquid layer optical load and the thermo-optical load are combined into a composite optical load and integrated to obtain the zero-order load. The one-sided normal refraction moment of the composite optical load is calculated, and the one-sided normal refraction moment is divided by the zero-order load to obtain the normal migration amount. The shoulder pseudo-boundary is inferred from the normal migration amount along the normal vector to restore the solid boundary, including: For the same arc length parameter and the same normal offset Liquid layer optical load With thermo-optical load The composite optical payload is obtained by superposition: ; in For composite optical payloads, For liquid layer optical payload, For thermo-optical loads, For the pseudo boundary along the shoulder of the platform The arc length parameter, For the pseudo-boundary relative to the shoulder The normal offset; According to composite optical load Sampling range along the normal direction Integrating to obtain the zero-order load : ; in For zero-order loads, Normal vector Normal sampling range in the direction; According to composite optical load With normal offset The sampling range along the normal direction is the weight. Integrating, we obtain the one-sided normal refraction moment. : ;in It is the normal refraction moment on one side; Based on the one-sided normal refraction moment and zero-order load According to the following formula: Calculate normal migration ,in Normal migration quantity; Based on normal migration Shoulder pseudo-boundary and normal vector According to the following formula: Reverse the pseudo-boundary of the shoulder and restore the solid boundary. ,in For entity boundary points, For the pseudo-boundary point of the platform shoulder, It is the normal vector.

[0033] It should be noted that the composite optical load is the total optical influence of the liquid layer optical load and the thermo-optical load superimposed at the same sampling point. The zero-order load is the total amount obtained by integrating the composite optical load over the normal sampling range. The one-sided normal refraction moment is the first-order normal moment obtained by integrating the composite optical load with the normal offset as a weight. The normal migration is the normal offset from the pseudo-boundary to the solid boundary, characterized by the ratio of the one-sided normal refraction moment to the zero-order load. The solid boundary is the real metal boundary curve obtained by extrapolating the shoulder pseudo-boundary based on the normal migration. In the actual workstation, the pseudo-boundary is often affected by both the residual liquid layer and the thermal boundary layer. Therefore, if the two are treated separately, it is difficult to describe the combined effect of the final boundary migration. Unifying the two types of loads into a composite optical load is necessary to describe the total optical disturbance with a single distribution. Furthermore, the most direct method to evaluate the overall optical disturbance along a normal line is to integrate the composite optical load over the entire normal sampling range. Therefore, the zero-order load is essentially the total normal load, used to measure the overall disturbance intensity. Boundary migration depends not only on the total load magnitude but also on the side and distance of the load relative to the boundary zero point. Therefore, a first-order integral must be performed with the normal offset as the weight to simultaneously encode the unilateral offset direction and intensity into the unilateral normal refraction moment. The unilateral normal refraction moment reflects the directional offset, while the zero-order load reflects the total amount. Therefore, dividing the former by the latter is equivalent to obtaining the equivalent offset center position of the composite optical load in the normal direction. This position is the normal migration of the pseudo-boundary relative to the solid boundary. The reason the pseudo-boundary deviates from the solid boundary is precisely because the composite optical load in the normal direction has undergone unilateral offset. Therefore, by subtracting the obtained normal migration along the opposite direction of the normal vector, the pseudo-boundary in the image can be pushed back to a position closer to the real metal boundary.

[0034] It should be noted that the registration method for liquid layer optical loads and thermo-optical loads at the same sampling point is as follows: first, the infrared imaging channel and the background pattern imaging channel are mapped to the contour channel coordinates through unified station calibration; then, the two types of load values ​​are read on the same normal sampling zone coordinates. Load superposition is only allowed when the arc length position and the normal offset position are simultaneously consistent. The one-sided definition of the unilateral normal refraction moment and its corresponding integration interval are as follows: first, the positive normal is fixed to point to the root side of the shoulder, and the negative normal is fixed to point to the outward open side; then, the sign of the normal offset is retained during the full interval integration, so that the load contribution on the root side is the positive moment, and the load contribution on the open side is the negative moment, thus expressing the unilateral offset with a signed first-order moment. The solution method for the normal migration when the zero-order load is close to zero is as follows: when the absolute value of the zero-order load is less than 0.5% of the peak value of the composite optical load in the current arc length neighborhood, a division operation is not performed directly. Instead, the normal migration at that point is set to zero, and smooth interpolation is performed using the effective migration at adjacent arc length positions. The smoothing constraints and outlier handling methods after backtracking the entity boundary are as follows: after obtaining the arc-length normal migration amount, median filtering and low-order smoothing are first performed on the migration amount sequence, and then outliers that deviate from the local mean by more than three times the standard deviation are removed; after backtracking to obtain the entity boundary, the boundary continuity and normal monotonicity should also be checked, and points that do not meet the conditions need to be re-interpolated and repaired.

[0035] In one embodiment of the present invention, the solid boundary is transformed to physical coordinates by combining geometric magnification and piecewise straight line fitting is performed. Based on the fitting result, the output dimension value is calculated, including: The entity boundary is represented as ,in For entity boundary points, The horizontal image coordinates are... The vertical image coordinates are... For arc length parameters; According to the following formula: entity boundary Convert to physical coordinates of entity boundary ,in For the physical coordinates of the entity boundary, Geometric magnification; Based on the straight boundary segment corresponding to the output dimension value, from the physical coordinates of the solid boundary Extract two sets of physical boundary points and ,in and For the physical boundary point set, and Physical coordinates of entity boundary The arc length range corresponds to different straight line boundary segments; For physical boundary point set and Perform line fitting on each line to obtain the first fitted line. The second fitted straight line And satisfy the constraints. ,in and For common normal parameters, The constant term parameters of the first fitted line, The constant term parameters of the second fitted line, and Physical coordinates; According to the following formula: Calculate and output dimensional values ,in This is for outputting dimensional values.

[0036] It should be noted that the horizontal image coordinate function of the entity boundary is the position function of the entity boundary in the horizontal direction of the image. The vertical image coordinate function of the entity boundary is the position function of the entity boundary in the vertical direction of the image. The physical coordinates of the entity boundary are the true physical coordinates obtained by converting the entity boundary image coordinates according to the geometric magnification. The first physical boundary point set is the first set of target straight line boundary points extracted from the physical coordinates of the entity boundary. The second physical boundary point set is the second set of target straight line boundary points extracted from the physical coordinates of the entity boundary. The first arc length range is the arc length segment corresponding to the first set of physical boundary points on the entity boundary. It is preferably selected as a stable straight line segment after avoiding chamfers, fillets, burrs, and end transitions, and it is preferably guaranteed to have no less than twenty effective boundary points, that is, it is necessary to ensure the stability of the fitting and the reliability of the common normal parameters. The second arc length range is the arc length segment corresponding to the second set of physical boundary points on the entity boundary. It is preferably selected as a stable straight line segment corresponding to the reference boundary on the other side of the target size, and it is preferably guaranteed to have no less than twenty effective boundary points, that is, the output size value is determined by the relative position of the two fitted lines. The first fitted line is the straight line representing the first size reference boundary obtained by fitting the first physical boundary point set. The second fitted line is a straight line representing the second dimensional reference boundary, obtained by fitting the second set of physical boundary points. The output dimension value is the final dimension result calculated from the normal distance between the two target fitted lines.

[0037] Specifically, this invention is deployed using an integrated vision inspection station, which includes three types of hardware devices: a dual telecentric contour imaging channel, an infrared imaging channel, and a background pattern imaging channel. The device is assembled with a standard part calibration module, a workpiece positioning fixture, and a synchronous trigger controller. Data acquisition is divided into a calibration phase and an online measurement phase. In the calibration phase, the standard part is placed in the positioning fixture, and images are acquired to calculate the geometric magnification, while reference images are acquired in a liquid-free and heat-free state. In the online measurement phase, the workpiece is transported from the production line to the positioning fixture for fixation, and the synchronous trigger controller simultaneously drives the three imaging channels to acquire images. The system sequentially performs pseudo-boundary extraction, normal sampling band construction, optical load calculation, solid boundary restoration, and dimensional value calculation, all without manual intervention. Furthermore, this invention is applicable to online inspection scenarios on machining production lines for precision shaft components and can be deployed at the blanking inspection station after grinding and turning processes of shaft parts. Applicable workpieces include rotating precision parts with shoulder chamfers and undercut grooves, such as automotive gearbox stepped shafts, motor spindles, hydraulic valve cores, and precision transmission shafts. It can be adapted to the continuous operation environment of metal cutting production lines.

[0038] Specifically, for example, when measuring the stepped shaft of an automotive gearbox, the output shoulder distance is 15.023 mm, the shaft outer diameter is 20.011 mm, and the relief groove width is 3.005 mm. Simultaneously, the corresponding measurement data records for each dimension are output, and the data can be directly used for workpiece qualification judgment. This invention enables real-time online measurement, with the complete measurement process for a single workpiece controlled within 200 milliseconds. The measurement process is non-contact, and the workpiece can be continuously transported with the production line without stopping for inspection. The measurement cycle can match the workpiece transport speed of conventional machining production lines, meeting the operational requirements of online full inspection. It adapts to the inspection needs of automated production of precision parts, replacing traditional manual sampling methods and achieving full-process online inspection of workpiece dimensions. The inspection data can be stably uploaded to the production control system, providing data reference for adjusting the parameters of processing equipment and reducing the number of defective workpieces produced. It can be directly integrated into existing production lines without significant equipment modifications, reducing the inspection and modification costs of the production line. It meets the control requirements for the dimensional accuracy of precision parts in the high-end equipment manufacturing field, contributing to the stable control of parts production quality.

[0039] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0040] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A method for online measurement of the dimensions of precision parts based on machine vision, characterized in that, Includes the following steps: Step S1: Obtain the geometric magnification and the infrared reference image in the liquid-free state and the background pattern reference image in the thermal-free state; Step S2: Simultaneously acquire the workpiece's contour image, infrared image, and background pattern image; Step S3: Extract the shoulder pseudo-boundary and its normal vector from the contour map, and extend the normal vector to both sides to establish a normal sampling zone; Step S4: Using the infrared reference image and infrared image in the liquid-free state, calculate the equivalent liquid layer thickness in the normal sampling band and convert it into liquid layer optical load. Step S5: Obtain the apparent displacement using the background pattern reference image and the background pattern image under the condition of no thermal disturbance, extract the normal displacement component along the normal vector of the apparent displacement and deduce the normal deflection angle, and integrate the normal deflection angle to obtain the thermo-optical load. Step S6: Combine the liquid layer optical load and the thermo-optical load into a composite optical load and integrate to obtain the zero-order load. Calculate the one-sided normal refraction moment of the composite optical load. Divide the one-sided normal refraction moment by the zero-order load to obtain the normal migration amount. Use the normal migration amount to inversely deduce the shoulder pseudo-boundary along the normal vector to restore the solid boundary. Step S7: Combine geometric magnification to transform the entity boundary to physical coordinates and perform piecewise straight line fitting, and calculate the output dimension value based on the fitting result.

2. The method for online measurement of precision parts dimensions based on machine vision according to claim 1, characterized in that, Place a standard part with known physical dimensions at the measurement position, obtain the contour image of the standard part, extract the pixel span corresponding to the known physical dimensions from the contour image of the standard part, and calculate the geometric magnification by dividing the pixel span by the known physical dimensions. When the measurement area is in a liquid-free state, an infrared reference image in the liquid-free state is acquired using the infrared imaging channel; When the measurement area is in a state of no thermal disturbance, a background pattern reference image is acquired using the background pattern imaging channel under the state of no thermal disturbance.

3. The online measurement method for precision parts dimensions based on machine vision according to claim 1, characterized in that, Place the workpiece within the measurement area, set the current synchronous acquisition time, and ensure that the acquisition times of the contour image, infrared image, and background pattern image are equal, and all are equal to the current synchronous acquisition time. At the current synchronous acquisition moment, the contour image of the workpiece is acquired using the contour imaging channel, and the intensity distribution of the contour image corresponding to the current synchronous acquisition moment is extracted. At the current synchronous acquisition time, infrared images of the workpiece are acquired using the infrared imaging channel, and the intensity distribution of the infrared image corresponding to the current synchronous acquisition time is extracted. At the current synchronous acquisition time, the background pattern image corresponding to the workpiece is acquired using the background pattern imaging channel, and the intensity distribution of the background pattern image corresponding to the current synchronous acquisition time is extracted.

4. The online measurement method for precision parts dimensions based on machine vision according to claim 1, characterized in that, Based on the contour image, the boundary points corresponding to the neighborhood of the workpiece shoulder are extracted, and the pseudo boundary of the shoulder is represented as a curve composed of horizontal and vertical coordinates and represented by the arc length parameter along the extension direction. Extract the set of straight boundary points corresponding to the workpiece generatrix from the shoulder pseudo-boundary, perform straight line fitting on the set of straight boundary points, and determine the normal parameters of the fitted straight line as the normal parameters of the generatrix fitted straight line. The normal vector of the shoulder pseudo-boundary is determined by the normal parameters of the fitted straight line. Using the boundary point on the pseudo-boundary of the shoulder as the base point, the sampling point position is obtained by extending along the normal vector in the positive and negative directions. The sampling point position is synthesized by the base point position and the normal offset along the normal vector direction. The normal sampling band is constructed based on the sampling point location, the sampling arc length range, and the normal sampling half-width, where the normal sampling band defines the normal sampling range along both sides of the normal vector.

5. The method for online measurement of precision parts dimensions based on machine vision according to claim 1, characterized in that, Read the infrared reference intensity and current infrared intensity of each sampling point in the liquid-free state within the normal sampling band; The intensity ratio is obtained by dividing the infrared reference intensity in the liquid-free state by the current infrared intensity. The natural logarithm of the intensity ratio is then performed, and the result of the natural logarithm is divided by twice the absorption coefficient of the liquid medium in the infrared imaging band to obtain the equivalent liquid layer thickness at each sampling point in the normal sampling band. The refractive index difference is obtained by subtracting the refractive index of air from the refractive index of the liquid medium. The refractive index difference is then multiplied by the equivalent liquid layer thickness, and the equivalent liquid layer thickness is converted into liquid layer optical load.

6. The method for online measurement of precision parts dimensions based on machine vision according to claim 1, characterized in that, Read the reference intensity of the background pattern in the thermally undisturbed state and the current background pattern intensity corresponding to the sampling point within the normal sampling band; The apparent displacement at the sampling point is obtained by matching the local intensity distribution of the background pattern reference image and the background pattern image under the condition of no thermal disturbance. The apparent displacement includes the displacement component in the horizontal direction of the image and the displacement component in the vertical direction of the image.

7. The online measurement method for precision parts dimensions based on machine vision according to claim 6, characterized in that, By performing a vector dot product operation between the apparent displacement and the normal vector, the normal displacement component at the sampling point is extracted. The normal deflection angle at the normal vector sampling point is calculated by dividing the normal displacement component by the equivalent focal length of the contour imaging channel; The thermo-optical load within the normal sampling band is obtained by multiplying the refractive index of air by the normal deflection angle and integrating the result along the normal offset direction within the normal sampling range.

8. The method for online measurement of precision parts dimensions based on machine vision according to claim 1, characterized in that, A composite optical load is obtained by superimposing the liquid layer optical load and the thermo-optical load at the same arc length parameter and the same normal offset. The zero-order load is obtained by integrating the composite optical load along the normal sampling range. Using the normal offset as a weight, the composite optical load within the normal sampling range is integrated to obtain the one-sided normal refractive moment; The normal migration is obtained by dividing the one-sided normal refraction moment by the zero-order load. By subtracting the product of the normal migration and the normal vector from the shoulder pseudo-boundary, the shoulder pseudo-boundary is inferred from the normal migration along the normal vector to restore the solid boundary.

9. The online measurement method for precision parts dimensions based on machine vision according to claim 1, characterized in that, Using horizontal and vertical image coordinates, the entity boundary is represented as a curve along the extension direction using the arc length parameter. The entity boundary is then converted into physical coordinates by dividing the entity boundary by the geometric magnification. Based on the straight line boundary segment corresponding to the output dimension value, extract two sets of physical boundary points corresponding to different arc length ranges from the physical coordinates of the entity boundary.

10. The method for online measurement of precision parts dimensions based on machine vision according to claim 9, characterized in that, Straight lines are fitted to two sets of physical boundary points respectively, and the normal parameters of the fitted lines are normalized to obtain the first fitted line and the second fitted line, wherein the first fitted line and the second fitted line have a common normal parameter. The constant term difference is obtained by subtracting the constant term parameters of the second fitted line from the constant term parameters of the first fitted line. The output dimension value is calculated by dividing the absolute value of the constant term difference by the arithmetic square root of the sum of squares of the common normal parameters.

Images