A Machine Vision-Based Method and System for Quality Inspection of Battery Pack Frame Shape

Abstract

The application discloses a battery pack frame appearance quality inspection method and system based on machine vision, and relates to the technical field of battery pack detection.The application solves the problem that the traditional quality inspection method cannot adapt to the needs of industrial development.After the error image and the standard image are calibrated, the center point and the direction of movement are analyzed to sort the pixel value sequence and the ratio column, so that the abnormal section and the abnormal area can be accurately identified.The method discards the limitations of traditional single point comparison, starts from the overall feature change, not only improves the accuracy of abnormal point determination, but also comprehensively displays the abnormal area distribution, provides intuitive and detailed abnormal information for external personnel, helps to quickly locate the problem source, greatly improves the quality inspection efficiency and problem solving speed, effectively guarantees the comprehensiveness, accuracy and efficiency of the battery pack frame appearance quality inspection, and has important significance for improving the overall quality of the battery pack.

Description

Technical Field

[0001] This invention relates to the field of battery pack inspection technology, specifically to a method and system for quality inspection of the shape of a battery pack frame based on machine vision. Background Technology

[0002] In fields such as new energy vehicles and energy storage systems, the performance and safety of the battery pack are crucial as the core energy storage component; while the battery pack frame, as a key structure that protects the battery modules and supports electrical connectors and cooling systems, directly affects the overall performance and reliability of the battery pack in terms of its shape and quality.

[0003] Traditional battery pack frame shape quality inspection methods mainly rely on manual visual inspection and simple measuring tools. Manual visual inspection is limited by the inspector's experience, fatigue level and subjective judgment, making it difficult to guarantee the consistency and accuracy of the inspection results. It is easy to miss some minor defects, such as tiny scratches and cracks. Moreover, for complex shapes and large-area frame shapes, manual inspection is extremely inefficient and cannot meet the needs of large-scale production.

[0004] While simple measuring tools can obtain some dimensional information, they can only measure specific key dimensions and cannot fully and in real time reflect the overall quality of the frame shape. Traditional measuring tool methods are powerless to deal with some surface flatness, subtle changes in edge contours, and potential defects in internal structures.

[0005] With the rapid development of the new energy industry, the production of battery packs has increased significantly, and the requirements for the quality of battery pack frame shapes are also rising. Traditional quality inspection methods can no longer meet the needs of industrial development, and there is an urgent need for an efficient, accurate, and comprehensive quality inspection method to ensure that the shape of the battery pack frame meets design standards and improves the overall quality and safety of the battery pack. Therefore, a machine vision-based quality inspection method for battery pack frame shapes has emerged, which can overcome the limitations of traditional methods and provide strong support for the development of the battery pack industry. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for quality inspection of battery pack frame shape based on machine vision, solving the problem that traditional quality inspection methods can no longer meet the needs of industrial development.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a machine vision-based method for quality inspection of the shape of a battery pack frame, comprising the following steps:

[0008] Step 1: Acquire the frame outline image of the battery pack using machine vision equipment, perform edge processing on the acquired frame outline image, confirm the edge contour using the Sobel algorithm, and divide the frame outline image into multiple verification images. Specific sub-steps are as follows:

[0009] The acquired frame outline image is converted to grayscale to confirm the associated grayscale image.

[0010] The pixel values ​​associated with different pixels within a grayscale image are labeled as X. i , where i represents different pixels. The Sobe l algorithm is then used to confirm the comprehensive gradient associated with the corresponding pixel. Pixels that satisfy the condition: comprehensive gradient > Y1 are labeled as gradient pixels, where Y1 is a preset value. If the condition is not met, no labeling is performed.

[0011] Connect several consecutive gradient pixels to confirm the gradient contour, and label the different regions contained in the gradient contour as different region contours, and confirm the image area of ​​different region contours. The region contour that satisfies: image area ∈ preset interval is recorded as feature contour, and the part of the image associated with the feature contour is labeled as the verification image, and the endpoint values ​​of its preset interval are all preset values.

[0012] Step 2: Perform feature verification on the confirmed multiple verification images. Identify the center point of the verification image from within the image, and then identify the deviation pixels outwards from this center point. This identifies the verification feature lines belonging to this verification image. Based on these feature lines, assess whether the verification image contains errors and label it as an error image. The specific method is as follows:

[0013] P1. Based on the edge contour of this verification image, combined with the two-dimensional coordinate system, the center point of this verification image is determined.

[0014] P2. Record the pixel value associated with the center point as the initial pixel value, and record the pixel values ​​associated with other pixels in the surrounding circle of this center point as other pixel values. Use: |initial pixel value - other pixel value| = pixel difference to identify the different pixel differences associated with other different pixels. Select the maximum value from the different pixel differences, and take the other pixels associated with the maximum value as the selected point. If there are multiple sets of selected points, randomly select one set of selected points from the multiple sets of selected points.

[0015] P3. Then, taking the surrounding ring as the inner ring, confirm its adjacent outer ring, and perform difference processing on the pixel values ​​between other pixels in the outer ring and the selected point to confirm the corresponding pixel difference. Select the maximum value from the confirmed multiple sets of pixel differences to determine the selected point. If there are multiple sets of selected points, confirm the point closest to the inner ring selected point from the multiple sets of selected points as the determined selected point.

[0016] P4. Then, using the outer ring as the inner ring, reconfirm the outer ring and repeat P3. Confirm the different selected points associated with the subsequent outer ring in sequence until the confirmed selected point is located on the edge contour of the verification image.

[0017] P5. Starting from the center point, connect the selected points in sequence to confirm the verification feature lines belonging to this verification image.

[0018] P6. Apply the same processing method as P1-P5 to the preset standard image. Identify the verification feature line from the standard image and record it as the standard line. Align the verification feature line with the standard line so that the starting points of the two sets of lines coincide. Then, rotate the verification feature line to identify whether the verification feature line and the standard line coincide during the rotation process. If not, record this verification image as an error image. If it does, no processing is required.

[0019] Step 3: Verify and analyze the calibrated error image against the standard image to confirm the variation characteristics of the same region in both images. Based on the differences in these variation characteristics, identify anomalous segments and record the areas covered by several anomalous segments as anomalous regions. Display the identified anomalous regions as follows:

[0020] The error image and the standard image are calibrated so that the error image coincides with the standard image;

[0021] Starting from the center point marked in the error image and the calibration image, a set of directions is randomly determined as the path direction. Other pixels associated with the path direction are recorded as path pixels. Starting from the starting point, the path pixels associated with each point are sorted sequentially, and the pixel values ​​associated with the corresponding points are sorted to confirm the pixel value sorting sequence. The pixel value sorting sequence associated in the error image is recorded as the error sequence, and the pixel value sorting sequence associated in the calibration image is recorded as the calibration sequence.

[0022] The absolute value of the pixel difference associated with adjacent pixel values ​​in the error sequence is determined, and the determined absolute values ​​are sorted according to the sorting method of the error sequence. According to the sorting method, the ratio of several groups of absolute values ​​is processed to confirm a set of error ratio columns. The calibration sequence is processed in the same way to confirm a set of calibration ratio columns.

[0023] The system identifies whether the error ratio column and the calibration ratio column are completely identical. If they are not identical, the first ratio of the error sequence and the calibration sequence are adjusted to the same value. The adjusted error sequence is then compared with the calibration sequence for verification. If they are completely identical, no processing is required. The system then processes the error sequences and calibration sequences associated with other path directions.

[0024] Identify the different ratios associated with the same sorting position within the system, record such ratios as error ratios, record the pixel differences associated with the error ratios as error differences, record the two sets of pixels associated with this error difference as error points, and record the value segments associated between the two sets of error points as outlier segments.

[0025] The abnormal segments associated with several directional lines in the error image are determined sequentially, and the areas covered by adjacent abnormal segments are recorded as abnormal patches, which are then marked in the error image.

[0026] Preferably, a machine vision-based battery pack frame shape inspection system includes:

[0027] On the machine vision side, the frame shape image of the battery pack is acquired, and the acquired frame shape image is processed for edge processing. The Sobe I algorithm is used to confirm the edge contour and the frame shape image is divided into multiple verification images.

[0028] The error image calibration end performs feature verification on multiple verified images. It identifies the center point of the image from within the verified image and then identifies the deviation pixels outward from the center point, thereby locking the verification feature line belonging to the verified image. Based on the verification feature line, it evaluates whether the verified image has an error and calibrates it as an error image.

[0029] The abnormal area display terminal verifies and analyzes the labeled error image and the standard image to confirm the change characteristics of the same area in the error image and the standard image. Based on the feature differences between the change characteristics, it locks out the abnormal segments and records the area covered by several abnormal segments as abnormal areas, and displays the confirmed abnormal areas.

[0030] This invention provides a machine vision-based method and system for quality inspection of battery pack frame shape. Compared with existing technologies, it has the following advantages:

[0031] This invention determines the center point of the verification image and gradually selects points based on pixel value differences to finally construct verification feature lines. This method can keenly capture the distribution of points and changes in numerical characteristics within the image. By comparing the verification feature lines with those of the standard image, it can accurately determine whether there are errors in the verification image, quickly lock the erroneous image, and achieve efficient detection of subtle differences in local areas of the battery pack frame shape. This significantly improves the accuracy of quality inspection and avoids errors caused by subjective human judgment.

[0032] After calibrating the error image with the standard image, the pixel value sorting sequence and ratio column are analyzed based on the center point and the direction of the path, which can accurately identify abnormal segments and abnormal areas. This method abandons the limitations of traditional single-point comparison and starts from the perspective of overall feature changes. It not only improves the accuracy of abnormal point location determination, but also comprehensively displays the distribution of abnormal areas, providing external personnel with intuitive and detailed abnormal information, helping to quickly locate the root cause of the problem, greatly improving the efficiency of quality inspection and the speed of problem solving, and effectively ensuring the comprehensiveness, accuracy and efficiency of battery pack frame shape quality inspection, which is of great significance to improving the overall quality of battery packs. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0034] Figure 2 This is a schematic diagram of the principle framework of the present invention. Detailed Implementation

[0035] 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 embodiments of the present invention, and not all embodiments. 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.

[0036] First Embodiment

[0037] Please see Figure 1 This application provides a machine vision-based method for quality inspection of the shape of a battery pack frame, including the following steps:

[0038] Step 1: Acquire the frame shape image of the battery pack using machine vision equipment, and perform edge processing on the acquired frame shape image. Use the Sobe I algorithm to confirm the edge contour and divide the frame shape image into multiple verification images. Specifically, the Sobe I algorithm can effectively detect the contour edges of the corresponding image. By confirming the gradient pixels associated with the edges, the gradient edge contour associated with the corresponding partition is confirmed. Based on the different partitions associated with different gradient edge contours, multiple verification images are locked. There are multiple rectangular partition areas with the same area on the outer end face of the battery pack. After each partition area is verified, it is divided into the corresponding verification image, which facilitates the subsequent overall quality inspection of the shape.

[0039] The specific sub-steps for segmenting the verification image are as follows:

[0040] The acquired frame outline image is converted to grayscale to confirm the associated grayscale image (each point in the image is associated with a corresponding RGB value, and different parameters in the RGB value are assigned different weight factors. Based on the weight factors and the associated specific values, the grayscale value associated with the corresponding point is confirmed. Then, the grayscale image is reconstructed based on the grayscale values ​​associated with different points to determine the corresponding grayscale image. This type of processing method is common in the existing technology, so it will not be elaborated on here).

[0041] The pixel values ​​associated with different pixels within a grayscale image are labeled as X. i Where i represents different pixels, the Sobe l algorithm is used to confirm the comprehensive gradient associated with the corresponding pixel (confirmed by combining the vertical gradient and the vertical gradient). Pixels that satisfy the condition: comprehensive gradient > Y1 are marked as gradient pixels, where Y1 is a preset value, and its specific value is determined by the operator based on experience. Pixels that do not meet the corresponding conditions are not marked.

[0042] Connect several consecutive gradient pixels (that is, gradient pixels that appear in sequence and are adjacent to each other) to confirm the gradient contour (the gradient contour is a bounding contour, and the area contained within the gradient contour is the corresponding contour area). Mark the different areas contained in the gradient contour as different region contours and confirm the image area of ​​different region contours. The region contour that satisfies: the image area ∈ the preset interval is recorded as the feature contour. The part of the image associated with the feature contour is marked as the verification image. The endpoint values ​​of the preset interval are all preset values. The preset interval is determined in advance by the operator according to the actual area size of the verification area, that is, multiple different rectangular areas on the surface of the battery pack.

[0043] Step 2: Perform feature verification on the confirmed multiple verification images. Identify the center point of the verification image, and then identify the deviation pixels outwards from this center point. This identifies the verification feature lines belonging to this verification image. Based on these feature lines, assess whether the verification image contains errors and label it as an error image. The specific labeling method is as follows:

[0044] P1. Based on the edge contour of this verification image, combined with the two-dimensional coordinate system, the center point of this verification image is confirmed (by combining the edge contour with the two-dimensional coordinate system, the different two-dimensional coordinates associated with different contour points are confirmed, and then the average value of several sets of two-dimensional coordinates is processed to confirm the average coordinates. Based on the location of the average coordinates, they are marked in the verification image to confirm the center point of this verification image).

[0045] P2. Record the pixel value associated with the center point as the initial pixel value, and record the pixel values ​​associated with other pixels in the surrounding circle of this center point as other pixel values. Use: |initial pixel value - other pixel value| = pixel difference to identify the different pixel differences associated with other different pixels. Select the maximum value from the different pixel differences, and take the other pixels associated with the maximum value as the selected point. If there are multiple sets of selected points, randomly select one set of selected points from the multiple sets of selected points.

[0046] P3. Then, taking the surrounding ring as the inner ring, confirm its adjacent outer ring, and perform difference processing on the pixel values ​​between other pixels in the outer ring and the selected point to confirm the corresponding pixel difference. Select the maximum value from the confirmed multiple sets of pixel differences to determine the selected point. If there are multiple sets of selected points, confirm the point closest to the inner ring selected point from the multiple sets of selected points as the determined selected point.

[0047] P4. Then, using the outer ring as the inner ring, reconfirm the outer ring and repeat P3. Confirm the different selected points associated with the subsequent outer ring in sequence until the confirmed selected point is located on the edge contour of the verification image.

[0048] P5. Starting from the center point, connect the selected points in sequence (that is, connect the selected points from the center point outwards step by step to confirm a set of connecting lines) to confirm the verification feature lines belonging to this verification image.

[0049] P6. Apply the same processing method as P1-P5 to the preset standard image. Identify the verification feature line within the standard image and record it as the standard line. Align the verification feature line with the standard line so that the starting points of the two sets of lines coincide. Then, rotate the verification feature line (based on the position of the starting point) to identify whether there is an overlap between the verification feature line and the standard line during the rotation process. If there is, no processing is required. If not, record this verification image as an error image.

[0050] Specifically, the existence of an error image indicates that there are differences between the arrangement of its internal points and the numerical characteristics of the image and the standard image. When these differences exist, the difference between the corresponding pixels will vary significantly. In the case of a large variation in the difference, it is necessary to confirm the error, locate the corresponding error area, and perform subsequent verification. Based on the changes between the pixels in the corresponding image and the changes in the standard image, it can be quickly determined whether there is a large error in this verification image, and a comprehensive determination can be made to complete the calibration process of the error image.

[0051] Step 3: Verify and analyze the calibrated error image against the standard image to confirm the variation characteristics of the same region in both images. Based on the differences in these variation characteristics, identify anomalous segments and record the areas covered by several anomalous segments as anomalous regions. Display the identified anomalous regions for external viewing. The specific method for identifying anomalous regions is as follows:

[0052] The error image and the standard image are calibrated to make the error image coincide with the standard image (based on the image pose associated with the image acquisition process and the associated direction marked in the standard image, the angles of the corresponding image edges are checked for coincidence during the overlap to make them consistent).

[0053] Starting from the center point marked in both the error image and the calibration image, a set of directions is randomly determined as the path direction. Other pixels associated with each path direction are recorded as path pixels. Starting from the starting point, the associated path pixels are sequentially sorted, and the pixel values ​​associated with each corresponding point are also sorted. The pixel value sorting sequence is then confirmed. The pixel value sorting sequence associated with the error image is recorded as the error sequence, and the pixel value sorting sequence associated with the calibration image is recorded as the calibration sequence (the error sequence and calibration sequence are associated with the same path direction). For example, the corresponding error image... The calibration image contains a designated center point. Taking due north as the direction of travel, starting from the center point, several pixels located in the same direction of travel are sorted to confirm that several pixels belong to the same direction of travel. Each different pixel is associated with a different pixel value. The difference between the pixel values ​​of adjacent pixels can be confirmed and the corresponding pixel difference can be locked. Then, several pixel differences are processed into ratios to lock the corresponding ratio sorting sequence. Subsequently, through the comparison and verification process with the calibration image, the different ratios existing within it are confirmed and the error ratio is locked.

[0054] The absolute value of the pixel difference associated with adjacent pixel values ​​in the error sequence is determined, and the determined absolute values ​​are sorted according to the sorting method of the error sequence. According to the sorting method, the ratio of several groups of absolute values ​​is processed to confirm a set of error ratio columns. The calibration sequence is processed in the same way to confirm a set of calibration ratio columns.

[0055] If the error ratio column and the calibration ratio column are exactly the same, no processing is required. Process the error sequences and calibration sequences associated with other path directions. If they are different, adjust the first ratio of the error sequence and the calibration sequence to the same value (after adjustment, other ratios associated with the corresponding ratio column will also change accordingly). Compare and verify the adjusted error sequence with the calibration sequence.

[0056] Identify the different ratios associated with the same sorting position within the system, record such ratios as error ratios, record the pixel differences associated with the error ratios as error differences, record the two sets of pixels associated with this error difference as error points, and record the value segments associated between the two sets of error points as outlier segments.

[0057] The abnormal segments associated with several directional lines in the error image are identified sequentially, and the areas covered by adjacent abnormal segments are recorded as abnormal regions. The abnormal regions are then marked in the error image, and the marked error image is displayed for external personnel to view.

[0058] Based on this process, abnormal points with anomalies in the corresponding path direction can be identified. By analyzing the changes in the difference characteristics, points with characteristic changes can be effectively confirmed. Compared with the original method of comparing single points, it has higher accuracy and can achieve better identification results.

[0059] Second Embodiment

[0060] Combination Figure 2 A machine vision-based battery pack frame shape inspection system includes:

[0061] On the machine vision side, the frame shape image of the battery pack is acquired, and the acquired frame shape image is processed for edge processing. The Sobe I algorithm is used to confirm the edge contour and the frame shape image is divided into multiple verification images.

[0062] The error image calibration end performs feature verification on multiple verified images. It identifies the center point of the image from within the verified image and then identifies the deviation pixels outward from the center point, thereby locking the verification feature line belonging to the verified image. Based on the verification feature line, it evaluates whether the verified image has an error and calibrates it as an error image.

[0063] The abnormal area display terminal verifies and analyzes the labeled error image and the standard image to confirm the change characteristics of the same area in the error image and the standard image. Based on the feature differences between the change characteristics, it locks out the abnormal segments and records the area covered by several abnormal segments as abnormal areas, and displays the confirmed abnormal areas.

[0064] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0065] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A battery pack frame contour quality inspection method based on machine vision, characterized in that, Comprise the following steps: Step one, based on the machine vision equipment on the frame of the battery pack appearance image acquisition, and the edge of the frame appearance image is handled, using Sobel algorithm to confirm the edge contour, the frame appearance image is divided into a plurality of check image; Step two, the confirmed a plurality of check image is carried out feature check, from the check image in the confirmation of this image center point, and then according to the center point outward deviation pixel point confirmation, so as to lock the check feature quantity line belongs to this check image, based on this check feature quantity line to assess whether there is error in this check image, and calibration for error image, the specific way is: P1, based on the edge contour of the check image, combined with two-dimensional coordinate system, confirm the center point of the check image; P2, with the center point associated with the pixel value as the initial pixel value, the other pixel points associated with the pixel value in the surrounding ring of the center point is recorded as other pixel value, using: | initial pixel value- other pixel value|= pixel difference value, confirm the different pixel difference value associated with different other pixel points, select the maximum value from different pixel difference value, the other pixel point associated with the maximum value is selected point, if there are multiple groups of selected points, then randomly confirm a group of selected points from the multiple groups of selected points; P3, again take the surrounding ring as the inner ring, confirm its adjacent surrounding outer ring, and the pixel value between the other pixel points in the surrounding outer ring and the selected point is processed, the corresponding pixel difference value is confirmed, and the maximum value is selected from the confirmed multiple groups of pixel difference value, to determine the selected point, if there are multiple groups of selected points, then confirm the point closest to the inner ring selected point from the multiple groups of selected points as the determined selected point; P4, again take the surrounding outer ring as the inner ring, and reconfirm the surrounding outer ring, repeat P3, and the different selected points associated with the subsequent surrounding outer ring are confirmed in turn, until the determined selected point is located on the edge contour of the check image; P5, with the center point as the starting point, the selected points determined in turn are connected, and the check feature quantity line belonging to the check image is confirmed; P6, the same processing method of P1-P5 is used for the preset standard image, the check feature quantity line is confirmed from the standard image and recorded as the standard line, the check feature quantity line and the standard line are compared, the starting points of the two groups of lines are overlapped, and then the check feature quantity line is rotated to identify whether there is overlap between the check feature quantity line and the standard line in the rotation process. If not, the check image is recorded as an error image; Step three, the calibrated error image and the standard image are checked and analyzed, the change characteristics of the same region in the error image and the standard image are confirmed, the abnormal segments are locked based on the feature difference between the change characteristics, the regions covered by a plurality of abnormal segments are recorded as abnormal areas, and the confirmed abnormal areas are displayed.

2. The machine vision-based battery pack frame contour quality inspection method of claim 1, wherein, In the step one, the specific substep for dividing the check image is: The acquired frame appearance image is processed by grayscale, and the grayscale image associated with the image is confirmed; The pixel values associated with different pixel points in the grayscale image are marked as X i Wherein i represents different pixel points, and the comprehensive gradient associated with the corresponding pixel points is confirmed by using a Sobel algorithm, and the pixel points satisfying comprehensive gradient>Y1 are marked as gradient pixel points, and Y1 is a preset value. A plurality of continuous gradient pixels are connected, the gradient contour is confirmed, and different regions contained in the gradient contour are marked as different region contours, and the image area of the different region contours is confirmed. The region contour meeting the image area∈ preset interval is recorded as a feature contour, and the part of the image associated with the feature contour is marked as a verification image, and the end point values of the preset interval are both preset values.

3. The machine vision-based battery pack frame contour quality inspection method according to claim 2, wherein, The pixel point not meeting the comprehensive gradient>Y1 is not marked.

4. The machine vision-based battery pack frame contour inspection method of claim 1, wherein, In the step P6, if the verification feature quantity line and the standard line coincide in the rotation process, no processing is required.

5. The machine vision-based battery pack frame contour inspection method of claim 1, wherein, In the step three, the specific way of confirming the abnormal area is: The error image and the standard image are calibrated to make the error image coincide with the standard image; The center point marked in the error image and the calibration image is taken as the starting point, a group of directions are randomly determined as the marching direction, the other pixel points associated with the marching direction are recorded as the marching pixel points, the marching pixel points associated with the starting point are sequentially sorted, the pixel values associated with the corresponding points are sorted, the pixel value sorting sequence is confirmed, the pixel value sorting sequence associated with the error image is recorded as the error sequence, and the pixel value sorting sequence associated with the calibration image is recorded as the calibration sequence; The absolute values of the pixel difference values associated with the adjacent pixel values in the error sequence are determined, the determined absolute values are sorted according to the sorting mode of the error sequence, and the same way is adopted for the calibration sequence to determine a group of calibration ratio columns; Whether the error ratio column and the calibration ratio column are completely same is identified, if not, the first ratio of the error sequence and the calibration sequence is adjusted to be the same ratio, and the error sequence and the calibration sequence after the adjustment are compared and verified: The different ratios associated with the same sorting position are confirmed, such ratios are recorded as error ratios, the pixel difference values associated with the error ratios are recorded as error difference values, the two groups of pixel points associated with the error difference values are recorded as error points, and the value segment between the two groups of error points is recorded as an abnormal segment. The abnormal segments associated with a plurality of marching directions in the error image are sequentially determined, the area covered by adjacent abnormal segments is recorded as an abnormal area, and the abnormal area is marked in the error image.

6. The machine vision-based battery pack frame contour quality inspection method according to claim 5, wherein, If the error ratio column and the calibration ratio column are completely same, no processing is required, and the error sequence and the calibration sequence associated with other marching directions are processed.

7. A battery pack frame contour quality inspection system based on machine vision, which operates according to the machine vision-based battery pack frame contour quality inspection method according to any one of claims 1-6, characterized in that, It includes: The machine vision end acquires the frame shape image of the battery pack, performs edge processing on the acquired frame shape image, confirms the edge contour by adopting the Sobel algorithm, and divides the frame shape image into a plurality of verification images; The error image marking end performs feature verification on the confirmed plurality of verification images, confirms the center point of the image from the verification image, and then confirms the deviation pixel point outward based on the center point, so as to lock the verification feature quantity line belonging to the verification image, and evaluate whether the verification image has error based on the verification feature quantity line, and mark as an error image; The abnormal area display end checks and analyzes the calibrated error image and the standard image, confirms the change characteristics of the same area in the error image and the standard image, and locks the abnormal section based on the feature difference between the change characteristics. The area covered by a plurality of abnormal sections is recorded as an abnormal area, and the confirmed abnormal area is displayed.

Images