Visual system positioning method and system, electronic equipment and computer program product

Abstract

The invention discloses a visual system positioning method and system, electronic equipment, a computer program product and a battery piece recognition device applied to a conveying belt, the conveying belt is used for conveying battery pieces, and the method comprises the steps that firstly, at least two interested areas on a battery piece image are obtained, the projection of the region of interest in the conveyor belt direction is not on the conveyor belt; secondly, acquiring the boundary of each region of interest; and finally, angular point positioning information in each region of interest is obtained through intersection point positioning information formed by intersecting every two target straight lines obtained on the boundary of each region of interest, and when the belt on the conveying belt is dirty, the position of the battery piece can still be accurately positioned through the algorithm.

Description

Technical Field

[0001] The present application relates to the field of computer algorithms, and in particular to a visual system positioning method, system, electronic equipment and computer program product. Background Art

[0002] In the manufacturing process of photovoltaic string welding machine components, the automatic string welding technology of battery cells is very critical. The visual positioning system plays an important role in the battery cell orientation correction, battery cell string transportation and other links. However, with the continuous operation of the equipment, the belt becomes dirty, causing positioning errors of the battery cells, which in turn causes unsatisfactory detection of defective battery cells, affecting the actual equipment output and causing product degradation.

[0003] In the related art, the edge of the battery cell is located by calculating the gradient of the area boundary, but this positioning method is only applicable when the belt is relatively clean. When the belt is dirty, gradient calculation errors will occur, which will cause deviations in edge positioning, resulting in inaccurate battery cell positioning. Summary of the invention

[0004] The embodiments of the present application disclose a visual system positioning method, system, electronic device and computer program product, which can improve the positioning accuracy of the battery cell when the conveyor belt of the photovoltaic string welding machine assembly is dirty.

[0005] A first aspect of an embodiment of the present application discloses a visual system positioning method, which is applied to a battery cell identification device on a conveyor belt, wherein the conveyor belt is used to transmit the battery cell, and comprises:

[0006] Acquire at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt;

[0007] Obtaining the boundary of the region of interest;

[0008] The corner point positioning information in each of the regions of interest is obtained, and the corner point positioning information in each of the regions of interest is obtained through the intersection positioning information in each of the regions of interest, and the intersection positioning information in the regions of interest is the positioning information of the intersection formed by the intersection of the target straight line segments on the boundary of the region of interest.

[0009] As an optional implementation manner, in the first aspect of this embodiment, obtaining the boundary of the region of interest includes:

[0010] Inputting the obtained regions of interest into a pre-trained segmentation model respectively to obtain a mask image corresponding to each region of interest;

[0011] Obtaining the largest edge region of each mask image;

[0012] Get all the straight line segments of the boundary of each edge area.

[0013] As an optional implementation manner, in the first aspect of this embodiment, obtaining the edge region with the largest area in each mask image includes:

[0014] For each of the mask images, a connected domain analysis method is used on each of the mask images to identify continuous regions in each of the mask images, and multiple continuous regions in each of the mask images are obtained. By comparing the area sizes of the multiple continuous regions, the continuous region with the largest area is obtained as the boundary of the region of interest.

[0015] As an optional implementation manner, in the first aspect of this embodiment, obtaining corner point positioning information in each of the regions of interest, wherein each corner point in the region of interest is obtained through intersection positioning information in each of the regions of interest, includes:

[0016] Filter out target straight line segments whose angles between each other are a preset angle from all straight line segments corresponding to each region of interest;

[0017] For each of the regions of interest, respectively, in each of the regions of interest, when the positioning information of at least two intersection points where the target straight line segments intersect each other, the obtained intersection positioning information is weighted averaged to obtain the corner point positioning information of the region of interest.

[0018] As an optional implementation, in the first aspect of this embodiment, when the target straight line segments in each region of interest intersect each other and have only one intersection point, the positioning information of the intersection point in the region of interest is the positioning information of the corner point in the region of interest.

[0019] As an optional implementation manner, in the first aspect of this embodiment, the acquiring at least two regions of interest on the cell includes:

[0020] At least two regions of interest are obtained from the cell image using the priori position information of the cell and pre-specified region size information.

[0021] As an optional implementation, in the first aspect of this embodiment, the method further includes verifying whether the number of corner points acquired in the region of interest is equal to the number of the regions of interest, wherein:

[0022] If the number of corner points obtained in at least two of the regions of interest is equal to the number of the regions of interest obtained, it is checked whether the relationship between the corner points in each of the regions of interest obtained satisfies the boundary angle characteristic of the battery cell.

[0023] If the number of corner points obtained in at least two of the regions of interest is one less than the number of the regions of interest, a positioning relationship between the corner points in each of the regions of interest is established through the prior information of the battery cell, so as to obtain the missing corner point positioning information in the region of interest;

[0024] If the number of corner points acquired in at least two of the regions of interest is at least two less than the number of the regions of interest, an error is reported.

[0025] As an optional implementation manner, in the first aspect of this embodiment, if the number of corner points obtained in at least two of the regions of interest is equal to the number of the regions of interest obtained, then verifying whether the relationship between the corner points in each of the regions of interest obtained satisfies the boundary angle characteristics of the battery cell includes:

[0026] Obtaining a positioning offset of the region of interest relative to the cell image, obtaining positioning information of corner points in the region of interest on the cell image through the positioning offset, and judging whether an angle between a straight line formed by the corner point on the cell image and an adjacent corner point satisfies a boundary angle characteristic of the cell through the obtained corner point positioning information on the cell image;

[0027] If the boundary angle characteristic of the battery cell is satisfied, the positioning information of the corner point on the battery cell image is the verified corner point positioning information;

[0028] If the boundary angle characteristic of the battery cell is not met, an error is reported.

[0029] As an optional implementation manner, in the first aspect of this embodiment, the method further includes:

[0030] The geometric relationship between the battery cell image and the actual positioning of the battery cell is obtained, and the verified corner positioning information on the battery cell image is mapped to the corner positioning information of the battery cell on the conveyor belt through the geometric relationship, thereby obtaining the corner positioning information of the battery cell on the conveyor belt.

[0031] A second aspect of an embodiment of the present application discloses a visual positioning system, comprising:

[0032] An area of ​​interest acquisition module is used to acquire at least two areas of interest on the cell image, wherein the projections of the areas of interest in the direction of the conveyor belt are not on the conveyor belt;

[0033] Image segmentation module: obtaining the boundary of the region of interest;

[0034] Corner point acquisition module: obtains the corner point positioning information in each of the regions of interest, each of the corner points in the region of interest is obtained through the intersection positioning information in each of the regions of interest, and the intersection positioning information in the region of interest is the positioning information of the intersection formed by the intersection of the target straight line segments on the boundary of the region of interest.

[0035] A third aspect of an embodiment of the present application discloses an electronic device, comprising: a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, any one of the above-mentioned visual positioning system methods is implemented.

[0036] A fourth aspect of an embodiment of the present application discloses a computer program product, including a computer program, which implements any of the above-mentioned visual positioning system methods when executed by a processor.

[0037] Compared with the related art, the embodiments of the present application include at least the following beneficial effects:

[0038] The visual system positioning method is applied to the battery cell recognition device on the conveyor belt. First, at least two regions of interest on the battery cell image are obtained, wherein the projection of the regions of interest in the conveyor belt direction is not on the conveyor belt; secondly, the boundary of each region of interest is obtained; finally, the corner point positioning information in each region of interest is obtained through the intersection positioning information formed by the intersection of two target straight lines obtained on the boundary of each region of interest. Through this algorithm, the position of the battery cell can still be accurately located when the conveyor belt is dirty. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flowchart of a first visual system positioning method provided in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of obtaining an area of ​​interest using a visual system positioning method provided in an embodiment of the present application;

[0042] Figure 3 A flowchart of a second visual system positioning method is provided for an embodiment of the present application;

[0043] Figure 4 A flowchart of a third visual system positioning method is provided for an embodiment of the present application;

[0044] Figure 5 A flowchart of a fourth visual system positioning method is provided for an embodiment of the present application;

[0045] Figure 6 A flowchart of a fifth visual system positioning method is provided for an embodiment of the present application;

[0046] Figure 7 A flowchart of a sixth visual system positioning method is provided for an embodiment of the present application;

[0047] Figure 8 A flowchart of a seventh visual system positioning method is provided for an embodiment of the present application;

[0048] Fig. 9 The statistical result of the deviation between the predicted edge and the real edge coordinates of a visual system method disclosed in an embodiment of the present application;

[0049] Fig.10 The statistical result of the angle deviation between the predicted edge and the real edge of a visual system method disclosed in an embodiment of the present application; DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present application are used to distinguish similar or different objects, and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0052] It should be noted that the terms "including" and "having" and any variations thereof in the embodiments of the present application and the accompanying drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0053] The photovoltaic string welding machine is the core equipment in the automated production line of solar photovoltaic modules. Its main function is to weld the positive and negative poles of the battery cells in series through welding ribbons. With the growth of demand for photovoltaic module manufacturing, the market size of photovoltaic string welding machines is also expanding rapidly.

[0054] In the process of photovoltaic module manufacturing, the automatic string welding technology of cells is crucial, and the visual positioning system plays a key role in the alignment and string handling of cells. However, long-term operation of the equipment may cause the belt to become dirty, thereby reducing the belt's adsorption force and causing the cell to drift. This drift will lead to cell positioning errors, which will affect the detection of defective cells, interfere with subsequent processes, affect the actual output of the equipment, and may lead to a decline in product quality. Therefore, optimizing and improving the positioning function of the visual positioning system is an urgent problem to be solved.

[0055] Current technology usually locates the edge of the battery cell by calculating the gradient of the area boundary. This method can achieve high positioning accuracy when the belt is clean, but when the belt is dirty, it will cause gradient calculation errors, which will cause deviations in edge positioning, resulting in inaccurate positioning of the battery cell.

[0056] The embodiments of the present application disclose a visual system positioning method, system, electronic device and computer program product, so that when the conveyor belt of the photovoltaic string welding machine assembly is dirty, the position of the solar cell can still be accurately located. The following are detailed descriptions:

[0057] See also Figure 1 , Figure 1 A flowchart of a first visual system positioning method provided in an embodiment of the present application, wherein the algorithm is applied to a cell on a conveyor belt, wherein the conveyor belt is a component of a photovoltaic stringer, and is used to transport the cell through a welding component in the photovoltaic stringer, and the flowchart includes at least the following steps S101-S103:

[0058] Step S101: acquiring at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt;

[0059] The number of regions of interest on the cell is selected according to the shape of the cell. For example, if the cell is a rectangle with negligible width, the cell can be positioned as a whole by only acquiring two regions of interest.

[0060] Step S102: obtaining the boundary of the region of interest;

[0061] Step S103: obtaining corner point positioning information in each of the regions of interest, wherein each corner point in the region of interest is obtained through intersection positioning information in each of the regions of interest, and the intersection positioning information in the region of interest is the positioning information of the intersection formed by the intersection of target straight line segments on the boundary of the region of interest.

[0062] For a clearer explanation of the above algorithm flow, see Figure 2 , Figure 2 A schematic diagram of obtaining a region of interest using a visual positioning algorithm provided in an embodiment of the present application includes a conveyor belt 21, a battery cell 22, and a region of interest 23.

[0063] In step S101 , at least two regions of interest 23 on the image of the battery cell 22 are acquired, wherein the projection of the regions of interest 23 in the direction of the conveyor belt 21 is not on the conveyor belt 21 .

[0064] Among them, the term Region of Interest (ROI) has different meanings in different fields, but generally refers to a part of the area that users or systems pay special attention to in fields including but not limited to image processing, computer vision, medical imaging, data analysis, etc. For example, in image processing and computer vision scenarios, when analyzing pictures or videos, ROI refers to a part of the image that contains information that needs to be further analyzed or processed, such as feature extraction, object recognition, etc. In the field of medical imaging, such as X-rays, CT scans, MRI, etc., ROI refers to an anatomical area that requires special attention, which may require special attention due to suspected lesions or other medical reasons. In the field of data analysis, ROI may refer to a specific part of a data set that is of special significance for research problems or business decisions. In the field of machine learning and artificial intelligence, when training a machine learning model, ROI may be specified as the focus of model training so that the model can better learn and identify the characteristics of these areas. In the field of video surveillance, ROI refers to key areas in the surveillance picture, such as entrances and exits, valuables storage areas, etc., and the dynamics of these areas will receive special attention and analysis.

[0065] There are many methods for obtaining the region of interest and determining the ROI, which can be selected according to actual needs, such as manual selection, automatic segmentation algorithm, machine learning and deep learning, feature matching, template matching, statistical analysis, expert system, interactive tools, model-based methods, etc., which are not specifically limited here. Among them, manual selection mainly involves users directly selecting areas on the image through a graphical user interface, usually using a mouse or touch operation to draw a rectangular, circular or other shaped area; automatic segmentation algorithms mainly use image processing techniques, such as threshold segmentation, edge detection, region growing, etc., to automatically identify specific features or objects in the image; machine learning and deep learning mainly involve training machine learning models or deep learning networks to identify and locate ROIs in images. Common deep learning methods include but are not limited to object detection and semantic segmentation. Common object detection algorithms include but are not limited to YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), Faster R-CNN (Fast Region-based Convolutional Network), R-CNN (Region-based Convolutional Network), and common semantic segmentation algorithms include but are not limited to FCN (Fully Convolutional Networks), U-type networks, MSG (Multi-Scale Combinatorial Grouping), Gated-SCNN (Gated Spectral CNN), etc.; feature matching mainly involves finding the corresponding areas between two images through feature matching technology in image registration or object recognition tasks. Common feature matching technologies include but are not limited to SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), ORB (Oriented FAST and Rotated BRIEF), etc. Template matching mainly uses predefined template images to find matching areas in large images, which is useful when looking for specific objects or landmarks. Statistical analysis mainly uses statistical tests or cluster analysis to determine specific subsets or patterns in the data; expert systems are mainly used in fields such as medical imaging, combining expert knowledge to define ROIs, for example, selecting areas based on anatomical knowledge or pathological features; interactive tools mainly combine automatic algorithms and user interaction, allowing users to make adjustments based on the areas recommended by the algorithm; model-based methods are mainly in some applications, and may define ROIs based on physical models or prior knowledge, for example, in remote sensing, areas are selected based on terrain models.

[0066] Once the ROIs are identified, various analytical tools and techniques can be applied to process these areas to extract useful information or make further decisions.

[0067] After the region of interest of the cell is obtained in step S101, step S102 is performed, that is, obtaining the boundary of the region of interest according to the obtained region of interest.

[0068] Different methods can be used to obtain the boundary of the region of interest in different situations. You can choose according to actual needs. It can be introduced in four parts.

[0069] The first part, that is, in the image processing method, the commonly used methods for obtaining the boundary of the region of interest include but are not limited to threshold segmentation, edge detection and contour detection. Among them, threshold segmentation mainly divides the pixels in the image into foreground and background according to the grayscale value. First, a threshold is selected to divide the pixels into two categories, which can be a global threshold or a local threshold. Secondly, the threshold is used to segment the image into a binary image, the foreground image is white, that is, the pixel value is 255, and the background image is black, that is, the pixel value is 0; edge detection is usually used to identify places in the image where the pixel intensity changes significantly, these places are usually the edges of objects, and commonly used edge detection algorithms include but are not limited to Canny edge detection: using a dual threshold method to determine the edge, with good noise suppression ability; Sobel Operator operator: calculating the intensity of the gradient in the image to detect the edge of the image; contour detection usually refers to extracting the boundary of an object from a binary image. Common methods include but are not limited to using the findContours function: finding contours in a binary image and returning the point set of the contour.

[0070] In the second part, when extracting the boundaries of the area of ​​interest, the data type faced is vector data or raster data. For vector data, that is, polygons, you can use geographic information system software or libraries to extract and process these boundaries. Commonly used geographic information system software includes but is not limited to QGIS (Quantum GIS), ArcGIS (Arc Geographic Information System), etc. Commonly used libraries include but are not limited to GDAL (Geospatial Data Abstraction Library), Shapely (Shapely Geometry Library), etc.; for the processing of raster data, such as remote sensing images, the boundaries of the area of ​​interest can be obtained through classification, segmentation and vectorization steps.

[0071] In the third part, in machine learning and deep learning methods, target detection or segmentation models are usually used to obtain the boundaries of the region of interest. Target detection mainly detects and locates the region of interest by using the target detection model. The model can output a bounding box or polygon to represent the region of interest; the segmentation model mainly uses the image segmentation model to segment the image and extract the precise boundaries of the region of interest.

[0072] In the fourth part, in the interactive method, the boundaries of the region of interest can be manually annotated. For example, in some applications, tools including but not limited to LabelImg, Labelbox, etc. can be used to interactively annotate the boundaries of the region of interest, and then the annotated boundaries can be extracted.

[0073] After the boundary of the region of interest is obtained in step S102, step S103 is performed, that is, obtaining the corner point positioning information of each region of interest according to the positioning information of the intersection points formed by the intersection of the target straight line segments on the boundary.

[0074] There are many methods for obtaining the straight line segments on the boundary according to the obtained boundary of the area of ​​interest, which can be selected according to the actual situation. Common methods include but are not limited to using image processing technology, using geographic information system tools, and using functions in the python library. In the image processing technology, it is necessary to use edge detection algorithms to detect edges on the image before performing line detection. Common edge detection algorithms include but are not limited to Canny edge detection and Sobel Operator operators, etc., and then apply line detection algorithms to identify straight line segments in the edge. Common line detection algorithms include but are not limited to Hough Transform, Gradient-based Methods, and Least Squares Line Fitting. Among them, Hough Transform mainly identifies straight lines by finding the maximum value of accumulation in parameter space; the cumulative gradient method mainly uses image gradient information to detect straight lines; the least squares line fitting method mainly estimates the parameters of the straight line by minimizing the sum of the squares of the distances from data points to the fitted straight line. When using the tool QGIS to extract straight line segments using geographic information system tools, you can use the "Extract Straight Line" plug-in or tool to extract straight line segments from vector data. When using ArcGIS, you can use the "Feature to Line" tool to convert polyline data into straight line segments. When using Python libraries to extract straight line segments of the boundaries of the area of ​​interest, commonly used libraries include Shapely, OpenCV, and scikit-image. Shapely can be used for geometric operations, that is, to create straight line segments and perform geometric analysis; OpenCV is a popular computer vision library that contains functions such as Hough transform to extract straight line segments in images; scikit-image provides a variety of image processing functions, including edge detection and line extraction.

[0075] After obtaining the straight line segments on the boundary of the region of interest, the corner points of the region of interest are obtained based on the intersection points formed by the intersection of the straight line segments. Since the projection of the region of interest in the direction of the conveyor belt is not on the conveyor belt, the corner points obtained based on the region of interest are not on the conveyor belt. Contamination of the conveyor belt will not affect the corner point positioning information obtained in the region of interest. Therefore, the problem that when the conveyor belt is dirty, it will cause gradient calculation errors, and then cause edge positioning deviations, resulting in inaccurate positioning of the battery cell is solved.

[0076] With respect to the step S102 of obtaining the boundary of the region of interest in the flowchart of the first visual system positioning method provided in the embodiment of the present application, the embodiment of the present application provides a flowchart of the second visual system positioning method, see Figure 3, the flow chart at least includes the following steps S301-S303:

[0077] Step S301: acquiring at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt;

[0078] Step S302: inputting the obtained regions of interest into a pre-trained segmentation model respectively to obtain a mask image corresponding to each region of interest;

[0079] Obtaining the largest edge region of each mask image;

[0080] Get all the straight line segments of the boundary of each edge area.

[0081] Step S303: Obtain corner point positioning information in each of the regions of interest, where each corner point in the region of interest is obtained through intersection positioning information in each of the regions of interest, where the intersection positioning information in the region of interest is the positioning information of intersections formed by the intersection of target straight line segments on the boundary of the region of interest.

[0082] Wherein in step S302, the segmentation model is a deep learning model for image processing, which is widely used in various computer vision tasks. It can identify and mark various parts or objects in the image mainly by pixel-level classification of the image. The pre-trained segmentation model is usually pre-trained on an open data set, which provides a good foundation and can be fine-tuned in a specific task to meet the needs of its own image segmentation task. Commonly used pre-trained segmentation models include but are not limited to U-Net, DeepLab series, Mask Region-based Convolutional Neural Network (Mask R-CNN), Segmentation Network SegNet, Pyramid Scene Parsing Network (Pyramid Scene Parsing Network, PSPNet), Bilateral Segmentation Network (Bilateral Segmentation Network, BiSeNet), High-Resolution Network (High-Resolution Network, HRNet) and Fast Semantic Segmentation Convolutional Neural Network (Fast Semantic Segmentation Convolutional Neural Network, Fast-SCNN), etc. Among them, U-Net is a convolutional neural network architecture consisting of an encoder and a decoder, which can capture detailed information and achieve high-precision segmentation; for the DeepLab series models, DeepLabv3+ is an improved version of the DeepLab series, which combines hollow convolution and encoder-decoder networks to obtain more refined segmentation results. It can not only handle objects of different scales, but also provide accurate boundary information; the Mask R-CNN model is based on Faster R-CNN's instance segmentation method generates an accurate segmentation mask for each detected object, which can perform object detection and instance segmentation and is suitable for complex scenes; the SegNet model is a deep convolutional encoding-decoding network for pixel-level segmentation; the PSPNet model captures information of different scales through a pyramid pooling module to enhance scene parsing capabilities, which can not only handle panoramic scene segmentation tasks, but also provide more fine-grained segmentation results; the BiSeNet network model combines spatial information and semantic information to improve segmentation accuracy, with fast inference speed and high accuracy; the HRNet network model maintains high-resolution features throughout the network process and improves segmentation accuracy through multi-scale feature fusion; the Fast-SCNN network model is a network designed for real-time semantic segmentation, with efficient inference speed and low computational complexity, and is very suitable for applications on embedded and mobile devices.

[0083] Please participate Figure 3In step S302, after the obtained region of interest is input into a pre-trained segmentation model, a mask image corresponding to each region of interest is obtained.

[0084] Among them, mask images are mainly used to define areas, select specific parts or mask out unnecessary areas. It can help limit image processing to specific areas, reduce the amount of calculation and improve processing efficiency. The mask image itself is an image with the same dimensions as the original image, in which each pixel value represents the processing status of the corresponding pixel in the original image. Mask images are usually of two types, binary masks and multi-value masks. The binary mask is a binary image, and the image has only two colors or grayscale values, which represent the areas to be processed and ignored, respectively. For example, in a black and white image, the black part, that is, the part with a pixel value of 0, represents the area to be ignored, and the white part, that is, the part with a pixel value of 255, represents the area to be processed. Multi-value mask: The pixel values ​​of the mask image are not limited to two values, but represent different areas or processing intensities. For example, grayscale values ​​can represent different weights or areas of different categories.

[0085] In some embodiments, after obtaining the mask image, a connected domain analysis method is generally used to obtain the edge region with the largest area in the mask image. For each of the mask images, a connected domain analysis method is used on each of the mask images to identify continuous regions in the image, and multiple continuous regions in the image are obtained. By comparing the area sizes of the multiple continuous regions, the continuous region with the largest area is obtained as the boundary of the region of interest.

[0086] Among them, Connected Component Analysis (CCA) is a technology used for image segmentation and analysis, mainly used to identify all connected regions in an image. These connected regions are regions composed of pixels with the same label. Connected domain analysis includes but is not limited to the following two steps: the first step is connected domain labeling, and the second step is regional attribute calculation.

[0087] The first step of connected domain labeling usually uses an algorithm to label the obtained mask image, mainly calculating all connected areas in the image and assigning a unique label to each area. Before that, it is necessary to select connectivity. Commonly used connectivity includes but is not limited to 4-connectivity and 8-connectivity. 4-connectivity means that each pixel is connected to its four adjacent pixels above, below, left, and right. 8-connectivity means that each pixel is connected to its eight adjacent pixels above, below, left, and right and on the diagonal. Commonly used connected domain labeling algorithms include but are not limited to depth-first search, breadth-first search, and connectivity labeling algorithms. Among them, depth-first search and breadth-first search mainly label each connected area through recursion or queues.

[0088] The second step is to calculate the regional attributes, including but not limited to the calculation of area, bounding box or shape characteristics. Calculating the area usually refers to calculating the number of pixels in each connected area; calculating the bounding box refers to calculating the bounding rectangle of each connected area; calculating the shape feature refers to calculating other shape features of the area, including but not limited to the perimeter, center point, etc.

[0089] Please read further Figure 1 In step S103, corner point positioning information in each of the regions of interest is obtained, and the corner points are obtained through intersection positioning information, and the intersection positioning information is the positioning information of the intersection formed by the intersection of target straight line segments on the boundary of the region of interest.

[0090] That is, after obtaining the region of interest, respectively, from all the straight line segments corresponding to each of the regions of interest, select the target straight line segments that satisfy the preset angles between them, and then obtain the positioning information of at least two intersection points where the target straight line segments intersect with each other in each of the regions of interest, and obtain the corner point positioning information of the region of interest by weighted averaging the obtained intersection positioning information.

[0091] In some embodiments, when the shape characteristic of the battery cell is a rectangular shape, the preset angle is an angle that deviates from a right angle by a preset value.

[0092] That is, after obtaining the region of interest, the target straight line segments that satisfy the approximately right angles between each other are screened out from all the straight line segments corresponding to each of the regions of interest, and then the positioning information of at least two intersection points where the target straight line segments intersect each other in each of the regions of interest is obtained, and the intersection positioning information obtained is weighted averaged to obtain the corner point positioning information of the region of interest.

[0093] In some embodiments, when it is obtained that in each region of interest, when the target straight line segments intersect each other with only one intersection point, the positioning information of the intersection point is the positioning information of the corner point.

[0094] It is understandable that when the target straight line segments intersect each other at only one intersection point, it indicates that the edge of the region of interest is relatively vertical, and therefore the positioning information of the single intersection point obtained by the intersection of the target straight line segments is considered to be the positioning information of the corner point.

[0095] For example, please refer to Figure 2, the cell is rectangular in shape. First, obtain at least two regions of interest on the cell image, wherein the projection of the regions of interest in the direction of the conveyor belt is not on the conveyor belt; second, input the regions of interest into the pre-trained segmentation model to obtain the mask image corresponding to each region of interest; obtain the edge region with the largest area of ​​each mask image by the connected domain analysis method; obtain all the straight line segments of the boundary of each edge region by the straight line detection algorithm. Finally, obtain the corner point positioning information in each region of interest, the corner point is obtained by the intersection positioning information, and the intersection positioning information is the positioning information of the intersection formed by the intersection of the target straight line segments on the boundary of the region of interest.

[0096] Optionally, the segmentation model includes but is not limited to U-Net, DeepLab series, Mask Region-based Convolutional Neural Network (Mask R-CNN), segmentation network SegNet, Pyramid Scene Parsing Network (Pyramid Scene Parsing Network, PSPNet), Bilateral Segmentation Network (BiSeNet), High-Resolution Network (HRNet) and Fast Semantic Segmentation Convolutional Neural Network (Fast-SCNN).

[0097] Optionally, the line detection algorithm includes but is not limited to Hough Transform, Gradient-based Methods, and Least Squares Line Fitting.

[0098] In some embodiments, the region of interest in the cell is obtained mainly by obtaining at least two regions of interest from the cell image through prior position information and pre-specified region size information. Figure 4 , Figure 4 A flowchart of a third visual system positioning method is provided for an embodiment of the present application, and the flowchart at least includes the following steps S401-S403:

[0099] Step S401: obtaining at least two regions of interest from the cell image by using the priori position information of the cell and pre-specified region size information;

[0100] Step S402: obtaining the boundary of the region of interest;

[0101] Step S403: obtaining corner point positioning information in each of the regions of interest, wherein each corner point in the region of interest is obtained through intersection positioning information in each of the regions of interest, and the intersection positioning information in the region of interest is the positioning information of the intersection formed by the intersection of target straight line segments on the boundary of the region of interest.

[0102] In step S401, the prior position information refers to the estimation of the approximate position of the region of interest in the cell based on existing knowledge or data, such as the structural features of the cell or certain specific marking positions, which are not specifically limited here. The pre-specified region size information refers to the size or shape of the region that has been set when the region is extracted. Here, the actual size of the target region in the cell can be pre-set according to the shape of the cell. The shape of the cell is not limited to rectangular, circular or other shapes, which are not specifically limited here.

[0103] By using prior position information and pre-specified area size information, at least two regions of interest are obtained from the battery cell image. That is, by combining the above information, these specific areas are located and extracted from the battery cell image. These areas are usually the key parts of the image for subsequent analysis and processing, including but not limited to defects, marks or other important features of the battery cell.

[0104] In some embodiments, based on the prior information of the shape or side length of the battery cell, when the battery cell is rectangular, two corner points in the region of interest are obtained according to the side length and shape of the battery cell, such as two corner points on the diagonal of the rectangle, or two corner points on the side of the rectangle. The positioning information of the entire battery cell can be obtained according to these two points and the shape and side length of the rectangle. Please refer to Figure 2 , the battery cell 22 is rectangular, and the battery cell 22 is located on the conveyor belt 21. According to the prior information of the side length and shape of the battery cell, the following can be obtained: Figure 2 The area marked 1 and 2 in the region of interest shown in the figure or Figure 2 The regions marked 1 and 3 in the region of interest are shown, and then the boundaries of the two regions of interest are obtained; finally, the target straight line segments of the boundaries of the region of interest are obtained according to the boundaries of the region of interest, and the corner point positioning information of the 4 regions of interest is obtained according to the intersection positioning information of the angles between the two target straight lines.

[0105] In some embodiments, see Figure 2The cell 22 is a rectangle and is located on the conveyor belt 21. The projection of the cell's region of interest in the direction of the conveyor belt is not on the conveyor belt. Therefore, the cell's region of interest 23 is set to be the region where the four corners of the cell are located, that is, four regions of interest are obtained.

[0106] For example, please refer to Figure 2 , the cell is rectangular in shape. First, the region of interest of the rectangular cell is specified as the four corners, as follows: Figure 2 The 1, 2, 3, and 4 marked in the region of interest 23 are obtained from the cell image. Figure 2 The image areas marked 1, 2, 3, and 4 in the region of interest 23 are shown, wherein the projection of the region of interest in the direction of the conveyor belt is not on the conveyor belt; secondly, the boundaries of the four regions of interest are obtained; finally, the target straight line segments of the boundaries of the regions of interest are obtained according to the boundaries of the regions of interest, and the corner point positioning information of the four regions of interest is obtained according to the intersection positioning information of the angles between the two target straight lines.

[0107] After obtaining the corner points of the region of interest, you need to determine whether the obtained corner points are reasonable. Figure 5 , Figure 5 A flowchart of a fourth visual system positioning method is provided for an embodiment of the present application, and the flowchart includes at least the following steps S501-S504.

[0108] Step S501: acquiring at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt;

[0109] Step S502: obtaining the boundary of the region of interest;

[0110] Step S503: obtaining corner point positioning information in each region of interest.

[0111] Step S504: Verify whether the number of corner points obtained in the region of interest is equal to the number of regions of interest. If so;

[0112] If the number of corner points obtained in at least two of the regions of interest is equal to the number of the regions of interest obtained, proceed to step S505: verifying whether the relationship between the corner points in each of the regions of interest obtained satisfies the boundary angle characteristic of the cell.

[0113] If the number of corner points obtained in at least two of the regions of interest is one less than the number of the regions of interest, proceed to step S506: establishing a positioning relationship between the corner points in each of the regions of interest through the prior information of the battery cell, thereby obtaining the missing corner point positioning information in the region of interest;

[0114] If the number of corner points acquired in at least two of the regions of interest is at least two less than the number of the regions of interest, the process proceeds to step S507: reporting an error.

[0115] In some embodiments, an algorithm is used to determine whether the number of corner points obtained is equal to the number of regions of interest obtained, wherein the algorithm includes but is not limited to the following four steps for comparison: step one: using a corner detection algorithm to identify corner points in an image; step two: determining the ROI in the image according to specific criteria or user input, wherein the ROI may be manually selected or automatically determined by an image segmentation algorithm, which is not specifically limited herein; step three: counting the detected corner points; step four: determining the number of ROIs in the image, which is divided into two cases: if the ROI is predefined or user-specified, it can be determined directly; if the ROI is determined by an image segmentation algorithm, it may be necessary to analyze the segmentation results to determine the number of independent regions; step four: comparing the number of corner points with the number of ROIs.

[0116] If the number of corner points obtained is equal to the number of regions of interest obtained, then check whether the relationship between the corner points obtained satisfies the boundary angle characteristics of the battery cell, and obtain the real positioning information of the battery cell through the verified corner point positioning information obtained, refer to Figure 6 , Figure 6 A flowchart of a fifth visual system positioning method is provided for an embodiment of the present application, and the flowchart includes at least the following steps S601-S604.

[0117] Step S601: acquiring at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt;

[0118] Step S602: obtaining the boundary of the region of interest;

[0119] Step S603: obtaining corner point positioning information in each region of interest.

[0120] Step S604: verifying that the number of corner points obtained in the region of interest is equal to the number of regions of interest, and then checking whether the relationship between the corner points in each region of interest meets the boundary angle characteristics of the battery cell;

[0121] Obtaining a positioning offset of the region of interest relative to the cell image, obtaining positioning information of corner points in the region of interest on the cell image through the positioning offset, and judging whether an angle between a straight line formed by the corner point on the cell image and an adjacent corner point satisfies a boundary angle characteristic of the cell through the obtained corner point positioning information on the cell image;

[0122] If the boundary angle characteristic of the cell is satisfied, the process proceeds to step S605: the positioning information of the corner point on the cell image is the verified corner point positioning information;

[0123] If the boundary angle characteristic of the battery cell is not satisfied, the process proceeds to step S606: reporting an error.

[0124] In step S604, the positioning offset of the region of interest relative to the cell image is obtained, and the positioning information of the corner point on the cell image is obtained through the positioning offset, that is, the positioning information of the corner point on the original cell is obtained according to the obtained corner point positioning information in the region of interest. In some embodiments, its implementation method mainly includes but is not limited to the following three steps: the first step: obtaining the region of interest, the second step: obtaining the coordinates of the corner points, obtaining the coordinates of the corner points within the ROI, and the coordinates of these corner points are relative to the coordinate system of the ROI. The third step: calculating the offset, that is, mapping the corner point from the coordinate system of the ROI back to the coordinate system of the original image, and the position and size of the ROI are needed to calculate the offset. Assuming that the position of the ROI in the original image is (x_offset, y_offset), the coordinates of the corner point are relative to the coordinates of the ROI (x_roi, y_roi). Then, the coordinates of the corner point in the original image (x_img, y_img) can be calculated by the following formula:

[0125] x_img = x_roi + x_offset;

[0126] y_img = y_roi + y_offset;

[0127] (x_offset, y_offset) is the position of ROI in the original image.

[0128] (x_roi, y_roi) are the coordinates of the corner points detected within the ROI.

[0129] If the number of corner points in the obtained region of interest is equal to the number of the obtained regions of interest, and the relationship between the corner points in each region of interest obtained by verification satisfies the boundary angle characteristics of the cell, then it is necessary to obtain the corner point positioning information of the cell on the conveyor belt through the obtained verified corner point positioning information on the cell image, please refer to Figure 7 , Figure 7 A flowchart of a sixth visual system positioning method is provided for an embodiment of the present application, and the flowchart includes at least the following steps S701-S706.

[0130] Step S701: acquiring at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt;

[0131] Step S702: obtaining the boundary of the region of interest;

[0132] Step S703: obtaining corner point positioning information in each region of interest.

[0133] Step S704: verifying that the number of corner points obtained in the region of interest is equal to the number of regions of interest, and checking whether the relationship between the corner points in each region of interest meets the boundary angle characteristics of the battery cell.

[0134] Step S705: Acquire the geometric relationship between the battery cell image and the position of the battery cell on the conveyor belt, and map the verified corner point positioning information on the battery cell image to the corner point positioning information of the battery cell on the conveyor belt through the geometric relationship, thereby obtaining the corner point positioning information of the battery cell on the conveyor belt.

[0135] In some embodiments, the acquired corner point coordinates are coordinates of the image plane, and therefore need to be converted into spatial coordinates to obtain the actual spatial positioning information of the battery cell, that is, the corner point positioning information of the battery cell on the conveyor belt, which is mainly achieved through camera calibration.

[0136] Camera calibration is to determine the camera's intrinsic parameters, i.e. focal length, principal point, distortion coefficient, and extrinsic parameters, i.e. the position and posture of the camera relative to the world coordinate system, by photographing a calibration plate of known size, including but not limited to a chessboard. These parameters help establish the mathematical relationship between image coordinates and actual world coordinates.

[0137] In some embodiments, converting the corner point coordinates into spatial coordinates to obtain the actual spatial positioning information of the battery sheet, that is, the corner point positioning information of the battery sheet on the conveyor belt, includes but is not limited to the following two steps:

[0138] Step 1: Determine the mapping relationship between image coordinates and actual space coordinates based on camera calibration, which is mainly achieved through the intrinsic matrix, distortion coefficients and extrinsic matrix. The intrinsic matrix includes focal length and principal point, which describes how the camera projects 3D world coordinates onto the 2D image coordinate plane. The distortion coefficient can correct the effect of lens distortion on image coordinates. The extrinsic matrix includes the position and orientation of the camera in the world coordinate system, which helps convert image coordinates to actual world coordinates.

[0139] The second step is to perform coordinate conversion. First, convert the image coordinates to camera coordinates, that is, use the camera's intrinsic parameter matrix and distortion coefficient to convert the image coordinates to coordinates in the camera coordinate system. Second, convert the camera coordinates to world coordinates, that is, apply the extrinsic parameter matrix to convert the coordinates in the camera coordinate system to coordinates in the actual world coordinate system.

[0140] Through the above conversion steps, the corner points on the battery cell are accurately mapped from the position in the image to the position in the actual space, and the exact position of the battery cell in the actual space is obtained, that is, the corner point positioning information of the battery cell on the conveyor belt, providing a basis for subsequent analysis and operation.

[0141] Exemplarily, the shape of the battery cell is a rectangle. First, four regions of interest on the battery cell image are obtained, wherein the projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt. Secondly, the boundaries of the four regions of interest are obtained. Thirdly, target straight line segments of the boundaries of the regions of interest are obtained based on the boundaries of the regions of interest. Corner point positioning information of the four regions of interest is obtained based on the intersection positioning information of the angles between the target straight lines. Finally, it is determined whether the corner point positioning information of the regions of interest is reasonable.

[0142] The positioning information of the corner point on the original cell image is obtained according to the obtained corner point positioning information in the region of interest. Under the ideal condition that only one corner point can be obtained for each region of interest, an algorithm is used to determine whether the number of corner points obtained is equal to the number of regions of interest obtained. If the number of corner points is equal to the number of regions of interest, it is checked whether the angle between the obtained corner point and the straight line formed by the adjacent corner points satisfies the boundary angle characteristic of the cell.

[0143] If the number of corner points is 3, i.e., one less than the area of ​​interest, i.e., one corner point is missing, the missing corner point can be supplemented according to the fact that the shape of the cell is a rectangle, i.e., the line connecting the coordinates of each corner point is a rectangle, and then it is checked whether the angle between the obtained corner point and the straight line formed by the adjacent corner point meets the boundary angle characteristic of the cell;

[0144] If the number of corner points is 0, 1 or 2, that is, at least two less than the number of regions of interest, the system reports an error and directly exits the algorithm process, prompting the system interface to display information such as "size detection abnormality".

[0145] If the angle between the corner point and the adjacent corner point obtained by verification satisfies the characteristic that the boundary angle of the battery cell is a right angle, the obtained corner point positioning information is the verified corner point positioning information, and then according to the verified corner point positioning information, the relationship between the image coordinates and the actual space coordinates is determined by camera calibration, so as to map the obtained corner point positioning information to the actual space position information of the battery cell, so as to obtain the actual space positioning information of the battery cell, that is, the corner point positioning information of the battery cell on the conveyor belt;

[0146] If the angle between the corner point and the adjacent corner point obtained by verification does not satisfy the characteristic that the boundary angle of the battery cell is a right angle, the system reports an error and directly exits the algorithm process, prompting the system interface to display information such as "size detection abnormality";

[0147] According to the above-mentioned visual system positioning detection algorithm, the present application embodiment provides a visual positioning system, see Figure 8 , Figure 8 The present application embodiment discloses a visual positioning system 80, including:

[0148] The region of interest acquisition module 81 is used to obtain at least two regions of interest from the cell image by using prior position information and pre-specified region size information.

[0149] Image segmentation module 82: used for inputting the obtained regions of interest into a pre-trained segmentation model respectively to obtain a mask image corresponding to each region of interest;

[0150] Corner point acquisition module 83: used for obtaining the edge region with the largest area of ​​each mask image by a connected domain analysis method for each mask image; using a straight line detection algorithm to obtain all the straight line segments of the boundary of each edge region, and respectively screening out target straight lines satisfying a preset angle between each of the straight line segments corresponding to each region of interest; according to the positioning information of at least two intersection points where the target straight line segments intersect with each other, weighted average of the positioning information of at least two intersection points is obtained to obtain the corner point positioning information of the region of interest, wherein when the target straight line segments intersect with each other and have only one intersection point, the positioning information of the intersection point is the positioning information of the corner point.

[0151] According to the above-mentioned visual system positioning detection algorithm, an embodiment of the present application provides an electronic device, including: a memory and a processor, the memory stores a computer program that can be run on the processor, and the processor implements any of the above-mentioned visual system positioning detection algorithms when executing the computer program.

[0152] According to the above-mentioned visual system positioning detection algorithm, an embodiment of the present application provides a computer program product, including a computer program, which implements any of the above-mentioned visual system positioning detection algorithms when executed by a processor.

[0153] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a ROM, etc.

[0154] As used herein, any reference to memory, storage, database, or other medium may include nonvolatile and / or volatile memory. Suitable nonvolatile memory may include ROM, programmable ROM (Programmable ROM, PROM), erasable PROM (Erasable PROM, EPROM), electrically erasable PROM (Electrically Erasable PROM, EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which is used as an external cache memory. By way of illustration and not limitation, RAM may be in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus direct RAM (RDRAM) and direct memory bus dynamic RAM (DRDRAM).

[0155] In order to verify the performance of the positioning algorithm involved in this application, one of the visual positioning methods is selected as an example to detect the statistical results. The statistical results are detailed in Fig. 9 , Fig.10 .

[0156] See further Figure 2 , the cell is rectangular in shape. First, the region of interest of the rectangular cell is specified as the four corners, as follows: Figure 2 The 1, 2, 3, and 4 marked in the region of interest 23 are obtained from the cell image. Figure 2 The image areas marked 1, 2, 3, and 4 in the region of interest 23 are shown, wherein the projection of the region of interest in the direction of the conveyor belt is not on the conveyor belt; secondly, the four regions of interest obtained are respectively input into the pre-trained mask region convolutional neural network, i.e., the segmentation model, to obtain the black and white image corresponding to each region of interest, i.e., the binary image; the edge area with the largest area of ​​each black and white image is obtained by the connected domain analysis method; and all the straight line segments of the boundary of each edge area are obtained by the Hough line detection algorithm. Finally, the target straight line segments of the boundary of the region of interest are obtained according to the boundary of the region of interest, and the corner point positioning information of the four regions of interest is obtained according to the intersection positioning information of the angles between the two target straight lines. The statistical results are detailed in Fig. 9 , Fig.10 .

[0157] See also Fig. 9 , Fig. 9 The statistical results of the deviation between the predicted edge and the real edge coordinates of the above-mentioned visual positioning method involved in this application are as follows: 2300 samples were collected on site, and 500 samples were randomly selected from them. The real edge coordinates are the position information of the four corner edges of the rectangular battery cell obtained by human eye observation on the original image. The predicted edge coordinates are obtained by sending these 500 samples to the positioning algorithm involved in this application, and the predicted edge coordinates output by the positioning algorithm are calculated point by point. The predicted edge coordinates and the real edge coordinates of all edge points are calculated point by point, and the sum is accumulated and then the average is obtained as follows Fig. 9 The statistical results of the coordinate deviation between the predicted edge and the actual edge of the positioning algorithm involved in this application in four regions of interest are shown, wherein the point-to-point calculated position deviation includes but is not limited to calculating the Euclidean distance, Manhattan distance, Chebyshev distance, etc. between two coordinate points.

[0158] See also Fig.10 , Fig.10The statistical results of the angle deviation between the predicted edge and the real edge of the above-mentioned visual positioning method involved in this application are as follows: 2300 samples were collected on site, and 500 samples were randomly selected from them. The real edge coordinates are the information of the edge directions of the four corners of the rectangular battery cell obtained by human eye observation on the original image. The predicted edge direction is obtained by sending these 500 samples to the positioning algorithm involved in this application, and the predicted edge direction output by the positioning algorithm is used to calculate the angle deviation between the predicted edge directions and the real edge directions of all edge points point by point, and then the sum is accumulated and the average is obtained as follows: Fig.10 The statistical results of the angle deviation between the predicted edge and the actual edge of the visual positioning method involved in this application in four areas of interest are shown, wherein the point-to-point calculation direction deviation includes but is not limited to calculating the angle between two edge directions by using the difference method.

[0159] It can be seen from the statistical results that the position deviation of each edge positioning does not exceed 2 pixels, and the angle deviation basically does not exceed 0.15°. Therefore, the positioning algorithm of the present invention has good performance and effect.

[0160] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. Those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application. It should be noted that "multiple" in the present application includes "two or more".

[0161] In the various embodiments of the present application, it should be understood that the size of the serial numbers of the above-mentioned processes does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0162] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0163] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0164] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The above is a detailed introduction to a visual positioning method disclosed in the embodiment of the present application. The principle and implementation method of the present application are described in detail using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those of ordinary skill in the art, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A visual system positioning method, characterized in that: A cell identification device applied to a conveyor belt, wherein the conveyor belt is used to transmit the cell, comprising: Acquire at least two regions of interest on the cell image, wherein projections of the regions of interest in the direction of the conveyor belt are not on the conveyor belt; Obtaining the boundary of the region of interest; The corner point positioning information in each of the regions of interest is obtained, and the corner point in each of the regions of interest is obtained through the intersection positioning information in each of the regions of interest, and the intersection positioning information in the regions of interest is the positioning information of the intersection formed by the intersection of the target straight line segments on the boundary of the region of interest.

2. The visual system positioning method according to claim 1, characterized in that: The obtaining of the boundary of the region of interest comprises: Inputting the obtained regions of interest into a pre-trained segmentation model respectively to obtain a mask image corresponding to each region of interest; Obtaining the largest edge region of each mask image; Get all the straight line segments of the boundary of each edge area.

3. The visual system positioning method according to claim 2, characterized in that: The step of obtaining the largest edge region of each mask image includes: For each of the mask images, a connected domain analysis method is used on each of the mask images to identify continuous regions in each of the mask images, and multiple continuous regions in each of the mask images are obtained. By comparing the area sizes of the multiple continuous regions, the continuous region with the largest area is obtained as the boundary of the region of interest.

4. The visual system positioning method according to claim 1, characterized in that: Obtaining corner point positioning information in each of the regions of interest, wherein each corner point in the region of interest is obtained by using intersection positioning information in each of the regions of interest, including: Filter out target straight line segments whose angles between each other are a preset angle from all straight line segments corresponding to each region of interest; For each of the regions of interest, respectively, in each of the regions of interest, when the positioning information of at least two intersection points where the target straight line segments intersect each other, the obtained intersection positioning information is weighted averaged to obtain the corner point positioning information of the region of interest.

5. The visual system positioning method according to claim 4, characterized in that: include: When the target straight line segments in each region of interest intersect each other with only one intersection point, the positioning information of the intersection point in the region of interest is the positioning information of the corner point in the region of interest.

6. The visual system positioning method according to any one of claims 1 to 5, characterized in that: The obtaining of at least two regions of interest on the cell includes: At least two regions of interest are obtained from the cell image using the priori position information of the cell and pre-specified region size information.

7. The visual system positioning method according to any one of claims 1 to 5, characterized in that: The method further includes verifying whether the number of corner points acquired in the region of interest is equal to the number of regions of interest, wherein: If the number of corner points obtained in at least two of the regions of interest is equal to the number of the regions of interest obtained, it is checked whether the relationship between the corner points in each of the regions of interest obtained satisfies the boundary angle characteristic of the battery cell. If the number of corner points obtained in at least two of the regions of interest is one less than the number of the regions of interest, a positioning relationship between the corner points in each of the regions of interest is established through the prior information of the battery cell, so as to obtain the missing corner point positioning information in the region of interest; If the number of corner points acquired in at least two of the regions of interest is at least two less than the number of the regions of interest, an error is reported.

8. The visual system positioning method according to claim 7, characterized in that: If the number of corner points obtained in at least two of the regions of interest is equal to the number of the regions of interest obtained, then verifying whether the relationship between the corner points in each of the regions of interest obtained satisfies the boundary angle characteristic of the battery cell includes: Obtaining a positioning offset of the region of interest relative to the cell image, obtaining positioning information of corner points in the region of interest on the cell image through the positioning offset, and judging whether an angle between a straight line formed by the corner point on the cell image and an adjacent corner point satisfies a boundary angle characteristic of the cell through the obtained corner point positioning information on the cell image; If the boundary angle characteristic of the battery cell is satisfied, the positioning information of the corner point on the battery cell image is the verified corner point positioning information; If the boundary angle characteristic of the battery cell is not met, an error is reported.

9. The visual system positioning method according to claim 8, characterized in that: The method further comprises: The geometric relationship between the battery cell image and the actual positioning of the battery cell is obtained, and the verified corner positioning information on the battery cell image is mapped to the corner positioning information of the battery cell on the conveyor belt through the geometric relationship, thereby obtaining the corner positioning information of the battery cell on the conveyor belt.

10. A visual positioning system, characterized in that: include: An area of ​​interest acquisition module is used to acquire at least two areas of interest on the cell image, wherein the projections of the areas of interest in the direction of the conveyor belt are not on the conveyor belt; Image segmentation module: obtaining the boundary of the region of interest; Corner point acquisition module: obtains the corner point positioning information in each of the regions of interest, each of the corner points in the region of interest is obtained through the intersection positioning information in each of the regions of interest, and the intersection positioning information in the region of interest is the positioning information of the intersection formed by the intersection of the target straight line segments on the boundary of the region of interest.

11. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the visual system positioning method described in any one of claims 1 to 9 is implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the visual system positioning method according to any one of claims 1 to 9 is implemented.

Images