A method and system for evaluating screw fastening quality based on image features

Abstract

This invention belongs to the field of image processing technology, specifically relating to a method and system for evaluating screw tightening quality based on image features. The method includes: acquiring a screw tightening image; obtaining a complex response image through a Log-Gabor filter; obtaining a phase consistency image based on each complex response image; obtaining the gradient direction of each pixel in the phase consistency image; constructing an accumulator image based on the gradient direction and offset points within a preset radius; determining the screw positioning point by fitting a two-dimensional quadratic polynomial function based on salient points and their local signal-to-noise ratio in the accumulator image; and obtaining the screw torque value using a screw positioning point driving torque measuring device to determine whether the screw is tightened. This invention overcomes the interference of specular reflection and extreme shadows on screw positioning by utilizing phase consistency in the frequency domain, improving the accuracy of screw positioning and thus assisting in screw tightening quality evaluation.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for evaluating screw fastening quality based on image features. Background Technology

[0002] Circuit breakers protect circuits and electrical equipment, preventing electrical fires and equipment damage caused by overloads, short circuits, and other faults. In the automated production and assembly of circuit breakers, screw tightening is a crucial process for ensuring product quality and structural stability. To achieve automated quality control, machine vision systems are typically used to precisely locate the screws, enabling subsequent robotic arms or torque measuring devices to accurately perform tightening or re-inspection operations.

[0003] Currently, the automated assessment of screw fastening quality typically relies on the combination of machine vision systems and physical measurement devices. This process involves: capturing images of the assembled workpiece using an industrial camera; analyzing the images using machine vision algorithms to identify and locate the screws; and finally guiding a torque measuring device mounted on a robotic arm to that location to measure the actual torque value and compare it with a standard range to determine whether the screws are properly fastened.

[0004] In the visual positioning stage, methods such as template matching based on grayscale values ​​and Hough circle transform are commonly used to extract the contour features of screws. However, when existing visual positioning technologies are applied to screws with metallic surfaces, the metal surface is prone to strong specular reflection and extreme shadows under industrial lighting conditions. This non-uniformity of illumination severely interferes with the screw image features in the spatial domain, resulting in the loss of features in highlight areas and the fragmentation of edge information in shadow areas. Consequently, traditional operators based on spatial domain grayscale transitions cannot reliably extract complete screw edges and structural features, often leading to edge fragmentation and feature loss, ultimately causing screw positioning failure and preventing subsequent torque measurement and quality assessment. Summary of the Invention

[0005] To address the technical problem that traditional grayscale-based operators are sensitive to brightness non-uniformity such as specular reflection and extreme shadows on the surface of metal screws, resulting in screw positioning failure and inability to measure screw torque, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a screw fastening quality assessment method based on image features, comprising: obtaining complex response maps of screw fastening images through different Log-Gabor filters; each pixel in the complex response map has a response vector composed of a real part and an imaginary part; obtaining the phase consistency of each pixel based on the sum of the real parts and the sum of the imaginary parts of pixels at the same position in each complex response map, and the sum of the magnitudes of the response vectors of pixels at the same position in each complex response map, to obtain a phase consistency map; taking any pixel in the phase consistency map as the target point, obtaining the offset point of the target point in the gradient direction of the target point and in the opposite direction of the gradient direction; initializing the accumulator map, and setting the offset point in the accumulator map... The pixel values ​​of the corresponding pixels are summed with the pixel value of the target point to obtain an accumulator image. The pixel with the maximum value in the accumulator image is designated as the salient point. The local signal-to-noise ratio of the salient point is obtained based on its pixel value and the mean and standard deviation of the pixel values ​​in its neighborhood. If the salient point satisfies the loop condition, a polynomial function is fitted to the pixels in the neighborhood of the salient point. The position of the maximum value in the fitted function in the accumulator image is designated as the screw positioning point. The pixel values ​​of the pixels in the neighborhood of the salient point in the accumulator image are set to 0. A new salient point is determined again, and the loop iteration continues until the salient point satisfies the loop termination condition. The screw fastening quality is evaluated using a screw positioning point control torque measuring device.

[0007] This invention employs a Log-Gabor filter to extract features in the frequency domain and calculates a phase consistency map based on the accumulation of real and imaginary parts and amplitudes, achieving feature extraction insensitive to illumination changes. This effectively overcomes the problems of edge fragmentation and feature loss caused by specular reflection and extreme shadows on the surface of metal screws. Furthermore, this invention utilizes gradient direction voting in the accumulator map to fully leverage the screw's ring structure characteristics to respond to the features of the screw's center point. By using local signal-to-noise ratio threshold judgment and polynomial function fitting, sub-pixel accuracy screw center positioning is achieved while suppressing noise interference. Finally, through iterative loops and a neighborhood zeroing mechanism, multiple screws in the image are accurately positioned, providing precise position input for subsequent torque measurement devices, thereby achieving automated and high-precision evaluation of screw fastening quality.

[0008] Preferably, the phase consistency satisfies the following relationship: In the formula, For the screw fastening diagram, the first one Phase consistency of each pixel The first part in the real part diagram The value of each pixel. The first in the imaginary part diagram The value of each pixel. The first in the total amplitude graph The value of each pixel. The scalar noise threshold. To prevent division by zero errors, It is a function for maximizing the value.

[0009] This invention uses the ratio between the vector sum and magnitude of the response vector and the scalar sum and magnitude to measure the phase consistency of features, effectively eliminating the influence of illumination intensity and shadows on feature extraction. At the same time, by introducing a scalar noise threshold, the scalar noise threshold is subtracted when calculating both the vector sum and magnitude and the scalar sum and magnitude, effectively suppressing spurious responses caused by image sensor noise in flat areas and improving the reliability of the phase consistency map.

[0010] Preferably, obtaining the offset point of the target point in the gradient direction and the opposite direction of the gradient direction includes: a preset radius range, wherein the offset point of the target point is a pixel whose distance from the target point in the gradient direction and the opposite direction of the gradient direction is within the radius range.

[0011] This invention determines the offset point by setting a preset radius range, limiting the accumulation operation to a distance interval in the gradient direction and the opposite direction of the target point. It can simultaneously respond to the ring structure within a certain radius range, enhancing the algorithm's adaptability to screw size changes under different specifications or shooting distances. At the same time, limiting the radius range also constrains the voting space, reduces interference from irrelevant pixels, and improves the significance of peaks in the accumulator graph.

[0012] Preferably, the method for obtaining the gradient direction of the target point includes: obtaining the horizontal gradient and vertical gradient of the target point through the Sobel operator, and obtaining the gradient direction of the target point based on the horizontal gradient and vertical gradient of the target point.

[0013] Preferably, the local signal-to-noise ratio satisfies the following relationship: In the formula, For the local signal-to-noise ratio of the salient point, The pixel values ​​of significant points in the accumulator graph. and These represent the mean and standard deviation of pixel values ​​within the neighborhood of a significant point in the accumulator graph, respectively. Parameters to prevent division by zero errors.

[0014] This invention measures the prominence of a salient point as a peak by the ratio of the difference between its pixel value and the mean of its neighborhood to the standard deviation of the neighborhood. It can effectively distinguish between strong peaks formed by the convergence of the screw center and weak local peaks caused by random noise or image clutter, and can effectively eliminate false salient points, thereby improving the accuracy and robustness of screw positioning.

[0015] Preferably, the salient point satisfies the loop condition, including: a preset absolute threshold and a sharpness threshold, wherein the loop condition is satisfied when the pixel value of the salient point is greater than the absolute threshold and the local signal-to-noise ratio of the salient point is greater than the sharpness threshold.

[0016] This invention achieves dual screening of salient points by setting both an absolute threshold and a sharpness threshold as loop conditions. The absolute threshold ensures that only strong response points with sufficient accumulated energy are considered, effectively filtering out low-energy background noise in the accumulator plot. The sharpness threshold, on the other hand, uses the local signal-to-noise ratio to ensure that the point is a sharp peak rather than a smooth local maximum, effectively eliminating spurious peaks caused by noise and clutter, and ensuring that the screw center candidate points entering the subsequent fitting steps are all high-quality.

[0017] Preferably, the step of fitting a polynomial function to the pixels within the neighborhood of the salient point includes: fitting a two-dimensional quadratic polynomial function to the pixels within the neighborhood of the salient point.

[0018] Preferably, the salient point satisfies the loop termination condition, including: the pixel value of the salient point is less than or equal to the absolute threshold.

[0019] Preferably, the screw fastening quality assessment via the screw positioning point control torque measuring device includes: sending the screw positioning point to the robotic arm control system; controlling the torque measuring device to move directly above the screw positioning point and engage with the screw; obtaining the screw torque value via the torque measuring device; and determining that the screw is not tightened if the torque value exceeds the range required by the process standard.

[0020] Secondly, the present invention provides a screw fastening quality assessment system based on image features, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned screw fastening quality assessment method based on image features is implemented.

[0021] By adopting the above technical solution, a computer program for evaluating screw fastening quality based on image features is generated and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, facilitating its use.

[0022] The beneficial effects of this invention are as follows: Addressing the problem of traditional spatial domain image feature extraction failure caused by metal reflection and shadows on screw surfaces, this invention performs feature analysis in the frequency domain using a Log-Gabor filter to obtain phase consistency that is insensitive to changes in illumination energy. This allows for stable extraction of the screw's ring-shaped structural features even in images with extreme lighting conditions. Furthermore, this invention utilizes gradient information from the phase consistency map to perform voting in the accumulator space to locate the screw center, fully leveraging the ring-shaped characteristics of the screw component. In addition, this invention eliminates spurious peaks generated by random noise through local signal-to-noise ratio and absolute thresholding, and achieves sub-pixel precision center positioning through polynomial function fitting, ensuring the robustness and high accuracy of the positioning results. Finally, an iterative search mechanism ensures that all screws in the image can be accurately located, and high-precision coordinate information is transmitted to the torque measurement device, providing accurate positioning support for screw fastening quality assessment. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating an image feature-based screw fastening quality assessment method according to the present invention;

[0024] Figure 2 This is a schematic diagram illustrating the screw fastening of the present invention;

[0025] Figure 3 This is a schematic diagram of the accumulator;

[0026] Figure 4 This is a schematic diagram illustrating the positioning effect of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] This invention discloses a screw fastening quality assessment method based on image features, referring to... Figure 1 This includes steps S1-S5:

[0030] S1. Collect screw tightening diagrams and obtain complex response diagrams through Log-Gabor filters.

[0031] It should be noted that the main part of the screw fastening diagram has a metallic texture, and its features are severely affected by specular reflection and extreme shadows in the spatial domain. Traditional operators based on spatial domain gray-level transitions are extremely sensitive to such brightness non-uniformity, leading to edge fragmentation, feature loss, and thus positioning failure. To overcome this defect, this invention transforms the image from the spatial domain to the frequency domain and obtains a complex response map through a Log-Gabor filter.

[0032] Specifically, a screw fastening image is acquired, converted into a grayscale image, and then converted to the frequency domain using a Fast Fourier Transform (FFT) to obtain a spectrum. A set of scale parameters and a set of orientation parameters are preset. Any parameter from the scale parameter set is combined with any parameter from the orientation parameter set to form a filter parameter set. Log-Gabor filters are constructed using each filter parameter set, and the spectrum is filtered using each Log-Gabor filter to obtain each filtered spectrum. An Inverse Fast Fourier Transform (IFFT) is then performed on each filtered spectrum to obtain its complex response.

[0033] For example, the scale parameter set is The direction parameter set is There are 4 scale parameters in the scale parameter group and 6 direction parameters in the direction parameter group, resulting in a total of 24 filter parameter groups and 24 Log-Gabor filters. Implementers can determine the scale parameter group and direction parameter group according to the actual situation.

[0034] For example, Figure 2 The diagram shows the screw fastening process of this invention. It can be seen from the diagram that the surface of the screw with a metallic texture will have specular reflection and extreme shadows, which will interfere with the positioning of the screw center point, causing the torque measuring device to be unable to move to the screw position, and thus affecting the evaluation of the screw fastening quality.

[0035] It should be further explained that performing an inverse fast Fourier transform on the filtered spectrogram obtained through the Log-Gabor filter yields a complex number for each pixel in the image. This complex number consists of a real part and an imaginary part, which together form the pixel's response vector. For example, the complex value of a pixel in the complex response map is... Then the response vector of that pixel is In other words, if there are 24 sets of filtering parameters, then each pixel in the screw fastening diagram corresponds to 24 response vectors.

[0036] S2. Obtain the phase consistency diagram based on the complex response diagram of each filter spectrum diagram.

[0037] It should be noted that the complex response maps of each filtered spectrogram reflect the local structural information of the screw fastening image at multiple scales and in multiple directions; however, this information is encoded in complex form, and its amplitude is directly related to the signal energy. Therefore, in images with both extreme reflections and shadows, the amplitude of the highlight edges is extremely high, while the amplitude of the shadow edges is extremely low. Directly using this information, which is heavily affected by illumination, will cause the positioning algorithm to fail. In order to reduce the influence of illumination on screw positioning and thus better evaluate the screw fastening quality, this invention obtains a phase consistency map based on the complex response maps of each filtered spectrogram.

[0038] Specifically, the real parts of pixels at the same position in the complex response maps of each filtered spectrum are summed to obtain a real part map, and the imaginary parts of pixels at the same position in the complex response maps of each filtered spectrum are summed to obtain an imaginary part map. The magnitude of the response vector of each pixel in the complex response maps of each filtered spectrum is obtained, and the magnitudes of the response vectors of pixels at the same position in the complex response maps of each filtered spectrum are summed to obtain a total magnitude map. The phase consistency of each pixel is obtained based on the values ​​of each pixel in the real part map and the imaginary part map, as well as the value of the corresponding pixel in the total magnitude map. Then, the phase consistency of each pixel is constructed into a phase consistency map.

[0039] Specifically, phase consistency satisfies the following relationship:

[0040] ;

[0041] In the formula, For the screw fastening diagram, the first one Phase consistency of each pixel The first part in the real part diagram The value of each pixel. The first in the imaginary part diagram The value of each pixel. The first in the total amplitude graph The value of each pixel. The scalar noise threshold. It is a function with maximum value. To prevent division by zero errors in parameters, this embodiment... The value is 0.01, and the implementers can adjust it according to the actual situation. The value of .

[0042] in, The first screw in the screw fastening diagram The vector and magnitude of the response vector of each pixel. The first screw in the screw fastening diagram The scalar and magnitude of the response vector of each pixel; at feature points with high phase consistency, the response vectors tend to be in the same direction, and the magnitude of their vector sum will be close to their scalar sum, causing phase consistency to approach 1; in non-feature regions with phase disorder on flat surfaces or in noisy areas, the vector sum approaches 0 due to the cancellation of vectors, causing phase consistency to approach 0. Therefore, phase consistency is obtained by the ratio between the magnitude of the vector sum and the magnitude of the scalar sum.

[0043] Ideally, completely flat or featureless regions in an image should have a value of 0 in the spatial domain, resulting in corresponding values ​​of 0 in both the real and imaginary parts of the image. However, due to sensor noise, even completely flat regions in the image may have some values, thus interfering with the accuracy of screw positioning. Subtract the scalar noise threshold, and take the maximum value between the subtraction and 0, such that only when... A phase-consistent response is only achieved when the scalar noise threshold is exceeded, in order to mitigate the effects of noise.

[0044] S3. Obtain the gradient direction of the pixels in the phase consistency map, obtain the offset point based on the gradient direction of the pixels in the phase consistency map, and construct the accumulator map based on the pixel value of the offset point.

[0045] It should be noted that the screw in the screw fastening diagram has a certain ring structure. The normal vector of any point on the ring structure of the screw will point to the same center point. Therefore, the pixel at the center point of the screw in the image is pointed to by the normal vectors of more pixels. Thus, this invention uses the ring characteristics of the screw to construct an accumulator diagram for screw positioning.

[0046] Specifically, the gradient direction of each pixel in the phase consistency map is obtained using the Sobel operator; an accumulator map with the same size as the screw fastening map is constructed, with the initial value of each pixel in the accumulator map being 0. Taking any pixel in the phase consistency map as the target point, when the value of the target point is greater than the scalar noise threshold, the unit vector of the gradient direction of the target point is obtained, and the offset point of the target point is obtained in the unit vector direction of the target point. The pixel value of the target point is accumulated by adding the pixel value of the target point to the pixel value of the offset point in the accumulator map; at the same time, the offset point of the target point is obtained in the opposite direction of the unit vector of the target point, and the pixel value of the target point is accumulated by adding the pixel value of the target point to the pixel value of the offset point in the accumulator map. The accumulator map is updated by accumulating the pixel value of the target point; obtaining the offset point of the target point in the unit vector direction of the target point includes: a preset radius range, with the pixel value of the target point whose distance from the target point in the unit vector direction is within the radius range as the offset point of the target point; obtaining the offset point of the target point in the opposite direction of the unit vector is similar and will not be elaborated here.

[0047] For example, the preset radius range is 60 to 100, and the implementer can adjust the radius range according to the actual situation.

[0048] For example, Figure 3 As shown in the accumulator diagram of this invention, it can be seen that the screw center point has a strong response in the accumulator diagram. Therefore, this invention obtains the screw positioning point based on the response of the screw center point in the accumulator diagram.

[0049] S4. Determine the screw positioning points according to the accumulator diagram.

[0050] Specifically, the pixel with the maximum value in the accumulator image is taken as the salient point. If the pixel value of the salient point is less than or equal to the absolute threshold, it means that there is no screw in the image. If the pixel value of the salient point is greater than the absolute threshold, the local signal-to-noise ratio of the salient point is obtained based on the pixel value of the salient point and the mean and standard deviation of the pixel values ​​in the neighborhood of the salient point. If the local signal-to-noise ratio is greater than the sharpness threshold, a two-dimensional quadratic polynomial function is fitted to the pixels in the neighborhood of the salient point to obtain the fitting function. The pixel corresponding to the maximum value in the fitting function in the accumulator image is taken as the screw positioning point.

[0051] Furthermore, the process involves iterative iteration, setting the values ​​of pixels within the neighborhood of salient points in the accumulator graph to zero, updating the accumulator graph, and searching for new salient points again. If the pixel value of a salient point is less than or equal to the absolute threshold, the iteration stops. If the pixel value of a salient point is greater than the absolute threshold and the local signal-to-noise ratio of the salient point is less than or equal to the sharpness threshold, the next iteration begins.

[0052] For example, the neighborhood of a pixel is the range centered on the pixel. The range of pixels is defined by an absolute threshold of 100 and a sharpness threshold of 2. Implementers can determine the neighborhood range, as well as the absolute and sharpness thresholds, based on the actual situation.

[0053] Specifically, the local signal-to-noise ratio satisfies the following relationship:

[0054] ;

[0055] In the formula, For the local signal-to-noise ratio of the salient point, The pixel values ​​of significant points in the accumulator graph. and These represent the mean and standard deviation of pixel values ​​within the neighborhood of a significant point in the accumulator graph, respectively. To prevent division by zero errors in parameters, this embodiment... The value is 0.01, and the implementers can adjust it according to the actual situation.

[0056] Among them, salient points are local peak points in the accumulator graph. Since there is a certain amount of random noise in the screw fastening graph, some noise points may be regarded as salient points, thus interfering with the determination of the screw positioning points. Therefore, the local signal-to-noise ratio is used to measure whether salient points are invalid points caused by random noise. The larger the value, the more likely the significant point is to be generated by the effective peak corresponding to the screw center, and the more necessary it is to determine the screw positioning point in the subsequent process. The smaller the value, the more likely the significant point is generated by invalid peaks corresponding to random noise, and the less necessary it is to determine the screw positioning point later.

[0057] It should be further explained that the coordinates of the salient point are discrete integer coordinates, while the actual screw center point may be between pixels. For example, the coordinates of the salient point may be (100, 100), but the actual position of the screw center point may be (100.5, 100.5). In order to make the obtained screw center point more accurate, this invention fits the pixels in the neighborhood of the salient point with a two-dimensional quadratic polynomial function, and takes the position of the maximum value in the fitting function in the accumulator graph as the screw positioning point.

[0058] For example, Figure 4 This is a positioning effect diagram of the present invention. As can be seen from the diagram, even if different screws in the screw fastening diagram have both specular reflection and extreme shadow features, the present invention can still extract the center point of the screw.

[0059] S5. Evaluate the screw fastening quality based on the screw positioning points.

[0060] Specifically, the screw positioning point is sent to the robotic arm control system, the torque measuring device carried by the robotic arm is moved directly above the screw positioning point, the torque measuring device is engaged with the screw, and the torque value of the screw is obtained through the torque measuring device. If the torque value exceeds the range required by the process standard, the screw is not tightened.

[0061] This invention also discloses an image feature-based screw fastening quality assessment system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an image feature-based screw fastening quality assessment method according to the present invention.

[0062] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for evaluating screw fastening quality based on image features, characterized in that, include: The complex response maps of the screw fastening diagram are obtained by using different Log-Gabor filters; each pixel of the complex response map has a response vector containing real and imaginary parts; the real part, imaginary part and response vector magnitude of the corresponding pixels of each complex response map are summed to obtain the real part map, imaginary part map and total magnitude map. The phase consistency of each pixel is obtained from the real part map, imaginary part map, and total magnitude map, resulting in a phase consistency map, specifically: ； In the formula, For the screw fastening diagram, the first one Phase consistency of each pixel The first part in the real part diagram The value of each pixel. The first in the imaginary part diagram The value of each pixel. The first in the total amplitude graph The value of each pixel. The scalar noise threshold. To prevent division by zero errors, It is a function for maximizing the value; Taking any pixel in the phase consistency map as the target point, the offset point is obtained in its gradient direction and the opposite direction. Initialize the accumulator image by summing the pixel values ​​of the target point to the pixel values ​​of the offset points in the accumulator image. The pixel with the maximum value in the accumulator image is designated as the salient point. The local signal-to-noise ratio (SNR) of the salient point is obtained based on its pixel value and the mean and standard deviation of the pixel values ​​within its neighborhood. If the salient point satisfies the loop condition, a polynomial function is fitted to the pixels within its neighborhood. The position of the maximum value in the fitted function within the accumulator image is used as the screw positioning point. The pixel values ​​of the pixels within the neighborhood of the salient point in the accumulator image are set to 0. A new salient point is determined, and the loop iteration continues until the salient point satisfies the loop termination condition. The quality of screw fastening is assessed by using a screw positioning point control torque measuring device.

2. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The step of obtaining the offset point of the target point in the gradient direction and the opposite direction of the gradient direction includes: a preset radius range, wherein the offset point of the target point is a pixel whose distance from the target point in the gradient direction and the opposite direction of the gradient direction is within the radius range.

3. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The method for obtaining the gradient direction of the target point includes: obtaining the horizontal and vertical gradients of the target point through the Sobel operator, and obtaining the gradient direction of the target point based on the horizontal and vertical gradients of the target point.

4. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The local signal-to-noise ratio satisfies the following relationship: ； In the formula, For the local signal-to-noise ratio of the salient point, The pixel values ​​of significant points in the accumulator graph. and These represent the mean and standard deviation of pixel values ​​within the neighborhood of a significant point in the accumulator graph, respectively. Parameters to prevent division by zero errors.

5. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The salient point satisfies the loop condition, which includes: a preset absolute threshold and a sharpness threshold. The loop condition is satisfied when the pixel value of the salient point is greater than the absolute threshold and the local signal-to-noise ratio of the salient point is greater than the sharpness threshold.

6. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The step of fitting a polynomial function to the pixels within the neighborhood of the salient point includes: fitting a two-dimensional quadratic polynomial function to the pixels within the neighborhood of the salient point.

7. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The salient point satisfies the loop termination condition, including: the pixel value of the salient point is less than or equal to the absolute threshold.

8. The method for evaluating screw fastening quality based on image features according to claim 1, characterized in that, The method of evaluating screw fastening quality by controlling the torque measuring device at the screw positioning point includes: sending the screw positioning point to the robotic arm control system; controlling the torque measuring device to move directly above the screw positioning point and engage with the screw; obtaining the torque value of the screw through the torque measuring device; and determining that the screw is not fastened if the torque value exceeds the range required by the process standard.

9. A screw fastening quality assessment system based on image features, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a screw fastening quality assessment method based on image features according to any one of claims 1-8.

Images