Automatic detection method and device for LED instrument

Abstract

The invention discloses an automatic detection method and device for an LED instrument. The automatic detection method comprises the following steps: shooting the LED instrument in a working state to obtain a UI image; identifying a feature image in the UI image to obtain feature data; for one feature image, comparing the feature data of the feature image with the calibration data of the calibration image to obtain a comparison result; obtaining a detection result of each feature image according to the comparison result; and outputting the detection result. The production efficiency and the product percent of pass can be remarkably improved, meanwhile, the influence on human eyes is reduced, and the method is suitable for various detection scenes of the LED display equipment and has wide application prospects.

Description

Technical Field

[0001] The invention relates to an automatic detection method and device for an LED instrument. Background Art

[0002] As one of the core applications of modern display technology, the development of LED (light-emitting diode) instruments is rooted in breakthroughs in semiconductor materials and optoelectronics. Early instruments relied on mechanical pointers or liquid crystal displays (LCDs), but these displays suffered from limited viewing angles, slow response times, and performance degradation at low temperatures. In the 1990s, with the invention of GaN (gallium nitride)-based blue LEDs and the maturity of white LED technology, LEDs, with their advantages of high brightness, low power consumption, and long life, gradually replaced traditional display solutions and became the preferred technology in the instrumentation field.

[0003] In the prior art, LED instrument detection has the disadvantages of low efficiency, high error rate, and significant negative visual impact on detection personnel. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the prior art of LED instruments, such as low detection efficiency, high error rate, and great negative visual impact on inspectors, and to provide an automated detection method and device for LED instruments that can significantly improve production efficiency and product qualification rate while reducing the impact on the human eye.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] An automated detection method for LED instruments, characterized in that the automated detection method comprises:

[0007] Photograph the LED instrument in working state to obtain the UI image;

[0008] Identify the feature image in the UI image and obtain feature data;

[0009] For a feature image, the feature data of the feature image is compared with the calibration data of the calibration image to obtain a comparison result;

[0010] Obtaining a detection result for each feature image according to the comparison result;

[0011] The detection result is output.

[0012] Preferably, obtaining the detection result of each feature image according to the comparison result includes:

[0013] For a feature image, the difference between the feature data and the calibration data is obtained according to the comparison result;

[0014] Determine whether the difference value meets the threshold, and if so, record abnormal information of the feature image, the abnormal information including the difference degree value and abnormal location data;

[0015] The abnormal information is used as the detection result.

[0016] Preferably, the automated detection method comprises:

[0017] Use feature data to obtain the size and angle changes of the UI image;

[0018] Adjusting the UI image to coincide with the calibration image using the transformation amount;

[0019] The overlapped images are used for comparison to obtain the comparison results.

[0020] Preferably, the step of obtaining the size and angle changes of the UI image using the feature data includes:

[0021] Acquire a target feature image, where the target feature image is an image close to an edge of the UI image;

[0022] Acquire a contrast feature image, where the contrast feature image is an image of a contour line having a preset length in the target feature image;

[0023] Obtaining a corresponding feature image in the calibration image corresponding to the contrast feature image;

[0024] Obtaining a contrast feature image and a plurality of straight contour lines in the corresponding feature image;

[0025] Obtaining the transformation amount using all straight contour lines;

[0026] Among them, the transformation amount is obtained by using all the straight contour lines: adjusting the length of the first straight contour line to overlap the calibration image with the first straight contour line of the UI image, using the two endpoints of the second straight contour line to perform proportional scaling and trapezoidal transformation of the UI image to make the second straight contour line overlap, and obtaining the adjustment amount of the first straight contour line, the proportional scaling and trapezoidal transformation amount of the second straight contour line as the transformation amount.

[0027] Preferably, the automated detection method comprises:

[0028] Photograph the LED instrument in normal state to obtain several instrument images;

[0029] Identify the characteristic images in all instrument images and obtain characteristic data;

[0030] The feature data of all instrument images are used to generate calibration images and calibration data.

[0031] Preferably, the automated detection method comprises:

[0032] Perform image preprocessing on UI images and instrument images;

[0033] Use feature extraction algorithm to identify feature images in UI images;

[0034] The comparison result is obtained by using a detection and recognition algorithm.

[0035] Preferably, the image preprocessing includes one or more of a grayscale algorithm, a filtering algorithm, and an image enhancement algorithm; the feature extraction algorithm includes one or more of an edge detection algorithm, a corner detection algorithm, and a shape feature extraction algorithm; and the detection and recognition algorithm includes one or more of a template matching algorithm, a feature-based target recognition algorithm, and a target tracking algorithm.

[0036] Preferably, the filtering algorithm includes:

[0037] Determine the size of the filter kernel;

[0038] For each pixel in the image, take the pixel as the center and take the pixel value within the range of the filter kernel size;

[0039] Calculate the average value of pixel values ​​and assign the average value to the current pixel to achieve smooth denoising of the image.

[0040] Preferably, the automated detection method comprises:

[0041] Obtain the ratio of the RGB values ​​of each feature image in a preset direction;

[0042] The rate of change of the ratio is obtained, and the rate of change is compared with the corresponding feature image in the calibration image to obtain a color detection result of the feature image, wherein the preset direction is the direction of the longest continuous pixel line segment of the feature image in the calibration image.

[0043] The present invention also provides an automated detection device for LED meters, comprising a shooting module, a computing module, and a fixing module. The shooting module is supported by the fixing module and mounted above the LED meter. The automated detection device is used to implement the automated detection method for LED meters as described above.

[0044] On the basis of conforming to the common sense in this field, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present invention.

[0045] The positive progress effect of the present invention is:

[0046] This invention uses image processing and feature recognition technology to automatically detect abnormalities in the LED instrument display of electric vehicles. This intelligent, efficient, and precise method significantly improves production efficiency and product qualification rates while minimizing visual impact. The method and device are applicable to a variety of LED display device detection scenarios and have broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of the automated detection method according to Example 1 of the present invention.

[0048] Figure 2 This is a schematic diagram of the effect of the automated detection device of Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0050] Example 1

[0051] See also Figure 1 , this embodiment provides an automatic detection device for an LED instrument of an electric vehicle, and the electric vehicle is an electric vehicle or an electric vehicle.

[0052] The electric vehicle described in this embodiment is an electric two-wheeled vehicle.

[0053] The automatic detection device includes a shooting module, a calculation module and a fixing module. The shooting module is supported by the fixing module and is mounted above the LED meter. The automatic detection device is used to implement the automatic detection method of the LED meter as described above.

[0054] The shooting module is used to shoot the LED instrument in working state to obtain the UI image;

[0055] The computing module is used to identify the characteristic image in the UI image and obtain characteristic data;

[0056] For a feature image, the operation module is used to compare the feature data of the feature image with the calibration data of the calibration image to obtain a comparison result;

[0057] The operation module is used to obtain the detection result of each feature image according to the comparison result and output the detection result.

[0058] Specifically, this embodiment uses a 2.3MP camera acquisition module to acquire the UI image displayed by the LED instrument in the target area during normal operation.

[0059] Detect the brightness, color and other characteristic images of each luminous area in the UI image, automatically analyze and process them, and identify the target object or feature point;

[0060] The target object or feature point clearly shows the normal and abnormal standards and ranges, as well as the corresponding accuracy requirements, for each small target detection area;

[0061] The target object or feature point is used to perform classification or recognition results, and corresponding thresholds or decision rules are set to determine the final detection results to judge whether the electric vehicle LED instrument display has display abnormalities. If so, the preset operation is executed, and the preset operation includes making abnormal prompts and / or recording abnormal locations, as well as abnormal alarm prompts.

[0062] The automated inspection device is implemented using an instrument inspection camera, which includes a camera module, an AI computing module, a mounting module, and a touchscreen display module. The camera module is used to capture images of the target area, initializes parameters such as frame rate, resolution, and exposure to ensure that the captured image quality meets subsequent processing requirements, and transmits the captured image data to the computing module in a suitable format. The mounting module is used to fix the relative position of the camera and the object being inspected, ensuring imaging stability.

[0063] Furthermore, the operation module is used to:

[0064] For a feature image, the difference between the feature data and the calibration data is obtained according to the comparison result;

[0065] Determine whether the difference value meets the threshold, and if so, record abnormal information of the feature image, the abnormal information including the difference degree value and abnormal location data;

[0066] The abnormal information is used as the detection result.

[0067] The operation module is also used for:

[0068] Use feature data to obtain the size and angle changes of the UI image;

[0069] Adjusting the UI image to coincide with the calibration image using the transformation amount;

[0070] The overlapped images are used for comparison to obtain the comparison results.

[0071] In the process of acquiring UI images, different shooting angles and distances will cause image deformation and misalignment of size. To address this, this application uses feature data to adjust the angle and size of the acquired UI image, that is, using the transformation amount to adjust the size and angle of the UI image to coincide with the calibration image.

[0072] The operation module is also used for:

[0073] Acquire a target feature image, where the target feature image is an image close to an edge of the UI image;

[0074] The images close to the edge of the UI image refer to images (icons) close to the edge (border) among several icons in the UI image, for example, icons closest to the top, bottom, left, and right sides.

[0075] Acquire a contrast feature image, where the contrast feature image is an image of a contour line having a preset length in the target feature image;

[0076] To compare feature images, you need to find images (icons) with contour lines of sufficient length. Such icons are more likely to represent the overall size and angle characteristics of the UI image.

[0077] Obtaining a corresponding feature image in the calibration image corresponding to the contrast feature image;

[0078] Obtaining a contrast feature image and a plurality of straight contour lines in the corresponding feature image;

[0079] Comparing the UI image and the calibration image using a straight contour line that is long enough can quickly obtain the transformation amount. For example, Figure 2 The bracket-shaped images on both sides of the image in the UI image are contrast feature images. The straight contour line in the middle can be used to quickly align the UI image and the calibration image to obtain the transformation amount.

[0080] Obtaining the transformation amount using all straight contour lines;

[0081] Among them, the transformation amount is obtained by using all the straight contour lines: adjusting the length of the first straight contour line to overlap the calibration image with the first straight contour line of the UI image, using the two endpoints of the second straight contour line to perform proportional scaling and trapezoidal transformation of the UI image to make the second straight contour line overlap, and obtaining the adjustment amount of the first straight contour line, the proportional scaling and trapezoidal transformation amount of the second straight contour line as the transformation amount.

[0082] The AI ​​operation module is used to extract effective features from the input image or video based on the collected target area image, and identify and detect the target object to determine whether the final detection result is normal or abnormal.

[0083] The touch screen display module is used for the user to intuitively display the collected target area image, and can operate the relevant settings of the detection camera and AI algorithm, such as parameter setting, image display, detection result display, alarm prompts, etc., to meet the interaction requirements of the human-computer interface.

[0084] The operation module is also used for:

[0085] Photograph the LED instrument in normal state to obtain several instrument images;

[0086] Identify the characteristic images in all instrument images and obtain characteristic data;

[0087] The feature data of all instrument images are used to generate calibration images and calibration data.

[0088] The operation module is also used for:

[0089] Perform image preprocessing on UI images and instrument images;

[0090] Use feature extraction algorithm to identify feature images in UI images;

[0091] The comparison result is obtained by using a detection and recognition algorithm.

[0092] The image preprocessing includes one or more of a grayscale algorithm, a filtering algorithm, and an image enhancement algorithm; the feature extraction algorithm includes one or more of an edge detection algorithm, a corner detection algorithm, and a shape feature extraction algorithm; the detection and recognition algorithm includes one or more of a template matching algorithm, a feature-based target recognition algorithm, and a target tracking algorithm.

[0093] The grayscale algorithm includes the following steps:

[0094] Read the input color image data and obtain the image width, height, and number of channels.

[0095] For each pixel in the image, its RGB color value is converted to a grayscale value according to the set color space conversion formula (such as `Y = 0.299R +0.587G + 0.114B`).

[0096] Assign the converted grayscale value to the pixel point at the corresponding position of the new grayscale image to obtain the grayscale image.

[0097] The filtering algorithm includes the following steps:

[0098] Determine the size of the filter kernel (such as 3x3, 5x5, etc.).

[0099] For each pixel in the image, take the pixel as the center and take the pixel value within the filter kernel size range.

[0100] Calculate the average value of these pixel values ​​and assign the average value to the current pixel to achieve smooth denoising of the image.

[0101] It can also be a median filter, including the steps:

[0102] Determine the filter kernel size.

[0103] For each pixel in the image, take the pixel value within the filter kernel range with the pixel as the center.

[0104] Sort these pixel values ​​and take the middle value as the new value of the current pixel, effectively removing isolated noise points such as salt and pepper noise.

[0105] It can also be Gaussian filtering, including the steps:

[0106] Determine the size and standard deviation of the Gaussian kernel.

[0107] Computes the Gaussian kernel coefficient matrix from a Gaussian function.

[0108] For each pixel in the image, a convolution operation is performed with the Gaussian kernel to obtain the filtered pixel value, thereby smoothing the image. At the same time, the distance relationship between the pixel and the surrounding pixels is taken into account, which better preserves the image details.

[0109] Image enhancement algorithms include:

[0110] Histogram equalization:

[0111] Calculate the grayscale histogram of the input image and count the number of pixels at each grayscale level.

[0112] Calculates the cumulative distribution function (CDF) based on the grayscale histogram.

[0113] The grayscale value of each pixel in the image is transformed according to CDF, and the original grayscale value is mapped to a new grayscale value range, making the grayscale distribution of the image more uniform and improving the contrast and visual effect of the image.

[0114] Feature extraction algorithms include:

[0115] Edge detection algorithm (Canny algorithm), algorithm steps:

[0116] Perform Gaussian filtering on the input image to remove noise interference and smooth the image. Use the cv2.GaussianBlur() function and set the appropriate Gaussian kernel size and standard deviation.

[0117] Calculate the gradient strength and direction of the image in the horizontal and vertical directions. You can use the Sobel operator (`cv2.Sobel()`) for gradient calculation.

[0118] Perform non-maximum suppression on the gradient intensity, retain the local maximum in the gradient direction, remove non-edge pixels, and refine the edges.

[0119] Through dual threshold detection, a high threshold and a low threshold are set, and pixels with gradient values ​​greater than the high threshold are determined as strong edges, and pixels with gradient values ​​less than the low threshold are determined as non-edges. Pixels between the high and low thresholds are determined to be edges based on their connectivity with the strong edge, and finally an edge image is obtained.

[0120] Corner detection algorithm, Harris corner detection, algorithm steps:

[0121] Compute the horizontal and vertical gradients of the image (e.g. using the Sobel operator).

[0122] Computes the product matrix of gradients.

[0123] The corner response value of each pixel is calculated according to the Harris corner response function, which takes into account the gradient changes of the pixel in the horizontal and vertical directions and their product relationship.

[0124] The response value of the corner point is thresholded and the pixel points with response values ​​greater than the set threshold are selected as the corner points. At the same time, non-maximum suppression can be performed to further filter the corner points and obtain the corner point features in the image.

[0125] Shi-Tomasi corner detection, algorithm steps:

[0126] Calculate the covariance matrix of the image, which reflects the grayscale changes of the image in different directions.

[0127] The corner response function value is calculated according to the eigenvalue of the covariance matrix. Pixels with larger eigenvalues ​​indicate larger grayscale changes in multiple directions, which are corner points.

[0128] The corner point response values ​​are thresholded and non-maximum suppressed to determine the final corner point position.

[0129] Shape feature extraction algorithms, including:

[0130] Area and perimeter calculations:

[0131] First, the contour of the target object is obtained through edge detection (`cv2.findContours()`).

[0132] For area calculation, use `cv2.contourArea()` function to calculate the area of ​​the region enclosed by the contour based on the coordinates of the contour points.

[0133] For perimeter calculation, use `cv2.arcLength()` function to calculate the perimeter of the contour based on the coordinates of the contour points, and you can choose whether to calculate the closed contour.

[0134] Circularity and rectangularity calculation:

[0135] Circularity calculation:

[0136] First calculate the area and perimeter of the target object.

[0137] The circularity value is calculated according to the circularity formula (e.g., `Circularity = 4 * pi * Area / Perimeter^2`). The closer the value is to 1, the closer the object is to a circle.

[0138] Rectangularity calculation:

[0139] Calculate the minimum enclosing rectangle of the target object (`cv2.minAreaRect()`) and get the length and width of the rectangle.

[0140] The rectangularity value is calculated according to the rectangularity formula (e.g., rectangularity = target object area / minimum bounding rectangle area). This value reflects the degree of similarity between the target object and the rectangle.

[0141] Hu moment calculation:

[0142] Calculate the geometric moments of image objects (`cv2.moments()`), including zero-order moments, first-order moments, and second-order moments.

[0143] According to the definition and calculation formula of Hu moment, Hu invariant moment is calculated from geometric moment. Hu moment has translation, rotation and scaling invariance, can effectively describe the shape information of target objects, and can be used for target recognition and image matching.

[0144] Target detection and recognition algorithms include:

[0145] Read the target template image and perform the same preprocessing operations as the input image (grayscale, filtering, etc.).

[0146] Use `cv2.matchTemplate()` function to perform template matching on the input image and select an appropriate matching method (such as `cv2.TM_CCOEFF_NORMED`, etc.).

[0147] The matching results are thresholded and the areas where the matching value is greater than the set threshold are selected as the possible locations of the target objects.

[0148] Based on the position and size information of the matching area, the position and posture information of the target object in the input image are determined, and a bounding box can be drawn or other marking operations can be performed.

[0149] Feature-based target recognition algorithm

[0150] Feature descriptor extraction (taking SIFT as an example):

[0151] Construct a scale space and detect extreme points in the image through the Gaussian Difference Pyramid (DoG), which may be potential feature points.

[0152] Accurately locate extreme points, remove unstable extreme points, and improve the accuracy of feature points.

[0153] A direction is assigned to each feature point to make it rotation invariant, and the main direction is determined based on the gradient direction distribution of the pixels around the feature point.

[0154] Generate a feature descriptor, take the feature point as the center, calculate the gradient amplitude and direction information of the pixel in a certain area, and combine this information into a feature vector to describe the characteristics of the feature point.

[0155] Feature matching and target recognition:

[0156] During the training phase, feature descriptors of multiple target objects are extracted and a feature library is established. Each target object corresponds to a set of feature descriptors, and its category information is recorded.

[0157] In the detection stage, feature descriptors are extracted from the input image.

[0158] Use a feature matching algorithm (such as `cv2.FlannBasedMatcher()`) to match the feature descriptors of the input image with the features in the feature library and calculate the matching score.

[0159] Based on the matching score and the set threshold, the category and location information of the target object in the input image are determined, and voting or other decision-making strategies can be used to determine the final recognition result.

[0160] The specific operation method is as follows:

[0161] Take 1-3 calibration images of the normal display of the LED instrument;

[0162] For testing the display UI area of ​​the LED instrument, calibrate the image collection of the luminous UI area under normal circumstances, and use it as a preset standard image to participate in the comparison and verification of subsequent tested images.

[0163] Use the actual detection area to frame the image, crop it, and reduce the image size;

[0164] Perform binarization calculation to extract the luminous area or color features in the image;

[0165] Based on the extracted features, classification and identification are performed and corresponding thresholds or decision rules are set to determine whether the final detection result is normal or abnormal.

[0166] The detection image is used to determine whether the matching threshold with the calibration sample is reasonable; if not, an abnormality is displayed, and the LED meter is judged as a failed prototype. The detection process is stopped and abnormal prompts and alarm prompts are given.

[0167] The steps of the entire automated detection process can be:

[0168] 1. Extract features from the target image and calibration image respectively to obtain key points and descriptors.

[0169] 2. Match the descriptors of the two images to obtain preliminary matching point pairs.

[0170] 3. Use iterative algorithms to eliminate incorrect matches and obtain reliable matching point pairs.

[0171] 4. Use these correctly matched point pairs to solve the homography matrix H by the least squares method.

[0172] 5. To align the target image to the calibration image, calculate H_inv = the inverse matrix of H.

[0173] 6. Use the inverse matrix H_inv to perform perspective transformation on the target image and map it to the coordinate system of the calibration image to achieve alignment.

[0174] After alignment, the absolute difference of the two images is calculated to determine the difference area of ​​the difference image, which is used for subsequent contour detection to locate abnormalities.

[0175] The operation module is also used for:

[0176] Determine the size of the filter kernel;

[0177] For each pixel in the image, take the pixel as the center and take the pixel value within the range of the filter kernel size;

[0178] Calculate the average value of pixel values ​​and assign the average value to the current pixel to achieve smooth denoising of the image.

[0179] The operation module is also used for:

[0180] Obtain the ratio of the RGB values ​​of each feature image in a preset direction;

[0181] The rate of change of the ratio is obtained, and the rate of change is compared with the corresponding feature image in the calibration image to obtain a color detection result of the feature image, wherein the preset direction is the direction of the longest continuous pixel line segment of the feature image in the calibration image.

[0182] Utilizing the above-mentioned automatic detection device for LED meters, this embodiment further provides an automatic detection method including:

[0183] Step 100: Photograph the LED instrument in working state to obtain a UI image;

[0184] Step 101: Identify the feature image in the UI image and obtain feature data;

[0185] Step 102: For a feature image, compare the feature data of the feature image with the calibration data of the calibration image to obtain a comparison result;

[0186] Step 103: Obtain the detection result of each feature image according to the comparison result;

[0187] Step 104: Output the detection result.

[0188] Wherein, the step 103 includes:

[0189] Step 1031: For a feature image, obtain a difference value between the feature data and the calibration data according to the comparison result;

[0190] Step 1032: Determine whether the difference value meets the threshold. If yes, execute step 1033; otherwise, execute step 1034.

[0191] Step 1033: record abnormal information of the characteristic image, wherein the abnormal information includes a difference degree value and abnormal position data, and then execute step 1035;

[0192] Step 1034 : record the difference value, and then execute step 1035 .

[0193] Step 1035: Use the abnormal information as the detection result.

[0194] Wherein, step 101 includes:

[0195] Step 1011: Identify the feature image in the UI image and obtain feature data.

[0196] Step 1012: Obtain the size and angle change of the UI image using the feature data.

[0197] Step 1013: Use the transformation amount to adjust the UI image to overlap with the calibration image.

[0198] Step 1014: perform comparison using the overlapped images to obtain a comparison result.

[0199] Step 1012 includes:

[0200] Acquire a target feature image, where the target feature image is an image close to an edge of the UI image;

[0201] Acquire a contrast feature image, where the contrast feature image is an image of a contour line having a preset length in the target feature image;

[0202] Obtaining a corresponding feature image in the calibration image corresponding to the contrast feature image;

[0203] Obtaining a contrast feature image and a plurality of straight contour lines in the corresponding feature image;

[0204] Then all straight contour lines are used to obtain the transformation amount;

[0205] Among them, the transformation amount is obtained by using all the straight contour lines: adjusting the length of the first straight contour line to overlap the calibration image with the first straight contour line of the UI image, using the two endpoints of the second straight contour line to perform proportional scaling and trapezoidal transformation of the UI image to make the second straight contour line overlap, and obtaining the adjustment amount of the first straight contour line, the proportional scaling and trapezoidal transformation amount of the second straight contour line as the transformation amount.

[0206] Step 100 includes:

[0207] Photograph the LED instrument in normal state to obtain several instrument images;

[0208] Identify the characteristic images in all instrument images and obtain characteristic data;

[0209] The feature data of all instrument images are used to generate calibration images and calibration data.

[0210] The automated detection method comprises:

[0211] Perform image preprocessing on UI images and instrument images;

[0212] Use feature extraction algorithm to identify feature images in UI images;

[0213] The comparison result is obtained by using a detection and recognition algorithm.

[0214] The image preprocessing includes one or more of a grayscale algorithm, a filtering algorithm, and an image enhancement algorithm; the feature extraction algorithm includes one or more of an edge detection algorithm, a corner detection algorithm, and a shape feature extraction algorithm; the detection and recognition algorithm includes one or more of a template matching algorithm, a feature-based target recognition algorithm, and a target tracking algorithm.

[0215] The filtering algorithm includes:

[0216] Determine the size of the filter kernel;

[0217] For each pixel in the image, take the pixel as the center and take the pixel value within the range of the filter kernel size;

[0218] Calculate the average value of pixel values ​​and assign the average value to the current pixel to achieve smooth denoising of the image.

[0219] The automated detection method comprises:

[0220] Obtain the ratio of the RGB values ​​of each feature image in a preset direction;

[0221] The rate of change of the ratio is obtained, and the rate of change is compared with the corresponding feature image in the calibration image to obtain a color detection result of the feature image, wherein the preset direction is the direction of the longest continuous pixel line segment of the feature image in the calibration image.

[0222] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these are merely illustrative and that the scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, and such changes and modifications are intended to fall within the scope of the present invention.

Claims

1. An automated detection method for LED instruments, characterized in that: The automated detection method comprises: Photograph the LED instrument in working state to obtain the UI image; Identify the feature image in the UI image and obtain feature data; For a feature image, the feature data of the feature image is compared with the calibration data of the calibration image to obtain a comparison result; Obtaining a detection result for each feature image according to the comparison result; outputting the detection result; The automated detection method comprises: Use feature data to obtain the size and angle changes of the UI image; Adjusting the UI image to coincide with the calibration image using the transformation amount; The overlapped images are used for comparison to obtain the comparison results.

2. The automated detection method for LED meters according to claim 1, wherein: The step of obtaining a detection result of each feature image according to the comparison result includes: For a feature image, the difference between the feature data and the calibration data is obtained according to the comparison result; Determine whether the difference value meets the threshold, and if so, record abnormal information of the feature image, the abnormal information including the difference degree value and abnormal location data; The abnormal information is used as the detection result.

3. The automated detection method for LED meters according to claim 1, wherein: The method of obtaining the size and angle change of the UI image by using the feature data includes: Acquire a target feature image, where the target feature image is an image close to an edge of the UI image; Acquire a contrast feature image, where the contrast feature image is an image of a contour line having a preset length in the target feature image; Obtaining a corresponding feature image in the calibration image corresponding to the contrast feature image; Obtaining a contrast feature image and a plurality of straight contour lines in the corresponding feature image; Adjust the length of the first straight contour line in the UI image to overlap the calibration image with the first straight contour line of the UI image, use the two endpoints of the second straight contour line to proportionally scale and trapezoidally transform the UI image to make the second straight contour line overlap, and obtain the adjustment amount of the first straight contour line, the proportional scaling and trapezoidal transformation amount of the second straight contour line as the transformation amount, where the first contour line and the second contour line are the straight contour lines with the farthest distance.

4. The automated detection method for LED meters according to claim 1, wherein: The automated detection method comprises: Photograph the LED instrument in normal state to obtain several instrument images; Identify the characteristic images in all instrument images and obtain characteristic data; The feature data of all instrument images are used to generate calibration images and calibration data.

5. The automated detection method for LED meters according to claim 4, wherein: The automated detection method comprises: Perform image preprocessing on UI images and instrument images; Use feature extraction algorithm to identify feature images in UI images; The comparison result is obtained by using a detection and recognition algorithm.

6. The automated detection method for LED meters according to claim 4, wherein: The image preprocessing includes one or more of a grayscale algorithm, a filtering algorithm, and an image enhancement algorithm; the feature extraction algorithm includes one or more of an edge detection algorithm, a corner detection algorithm, and a shape feature extraction algorithm; the detection and recognition algorithm includes one or more of a template matching algorithm, a feature-based target recognition algorithm, and a target tracking algorithm.

7. The automated detection method for LED meters according to claim 6, wherein: The filtering algorithm includes: Determine the size of the filter kernel; For each pixel in the image, take the pixel as the center and take the pixel value within the range of the filter kernel size; Calculate the average value of pixel values ​​and assign the average value to the current pixel to achieve smooth denoising of the image.

8. The automated detection method for LED meters according to claim 1, wherein: The automated detection method comprises: Obtain the ratio of the RGB values ​​of each feature image in a preset direction; The rate of change of the ratio is obtained, and the rate of change is compared with the corresponding feature image in the calibration image to obtain a color detection result of the feature image, wherein the preset direction is the direction of the longest continuous pixel line segment of the feature image in the calibration image.

9. An automatic detection device for LED instruments, characterized in that: The automated detection device includes a shooting module, a computing module, and a fixing module. The shooting module is supported by the fixing module and mounted above the LED meter. The automated detection device is used to implement the automated detection method for the LED meter according to any one of claims 1 to 8.

Images