A method for adjusting the mirror reflection of fill light at night based on big data analysis

Abstract

This invention provides a method for adjusting the specular reflection of supplementary lighting at night based on big data analysis, belonging to the field of gimbal telemetry technology. The method includes the following steps: constructing a database; performing nighttime identification of instrument data based on an instrument detection model, obtaining and storing the optimal parameters for camera identification of the instrument under different time periods and lighting conditions through nighttime testing; moving a gimbal with a supplementary light to the instrument to be imaged; adjusting the gimbal's camera parameters by calling the optimal camera parameters for that time period from the database, and acquiring an image of the instrument; obtaining the identification result and reflectivity of the instrument image based on the instrument detection model, and determining whether reflection has occurred based on the reflectivity; if reflection has occurred, adjusting the brightness of the supplementary light until accurate instrument data identification results are obtained. This method solves the problem of gimbal reflection during photography in practical applications, resulting in the instrument appearing completely white.

Description

Technical Field

[0001] This invention relates to the field of gimbal telemetry technology, specifically to a method for adjusting the nighttime specular reflection of a supplementary light based on big data analysis. Background Technology

[0002] With the development of communication technology, telemetry equipment, and other hardware and software technologies, as well as the continuous improvement of automation levels, PTZ telemetry is being widely used in more and more fields of people's production and daily life. Behind these PTZs replacing manual labor in corresponding tasks, there are specific algorithmic logics that support the PTZs in judging various relevant factors in the scene, so as to ultimately complete the designated tasks.

[0003] In the field of instrument telemetry and identification technology, due to the principle of specular reflection, when the gimbal takes pictures of the instrument dial at night, it is very easy to produce a white area, which greatly affects the identification effect and accuracy of the instrument dial data. Summary of the Invention

[0004] To address the aforementioned issues, the present invention aims to provide a nighttime specular reflection adjustment method for supplementary lighting based on big data analysis. This method solves the problem of gimbal reflection during photography and the resulting white-tinted display on instruments in practical applications.

[0005] To achieve the above objectives, the present invention provides the following technical solution.

[0006] A method for adjusting the specular reflection of supplementary lighting at night based on big data analysis includes the following steps:

[0007] Build a database;

[0008] The collected instrument data is used for nighttime identification to obtain the optimal parameters for the camera to identify the instruments under different time periods and lighting conditions, and these parameters are stored in the database. The parameters include time interval, ambient light intensity, camera zoom value, and zoom value.

[0009] Move the pan-tilt unit with the fill light to the instrument where the image to be captured;

[0010] The optimal parameters for that time period are retrieved from the database to adjust the camera parameters of the gimbal and to obtain instrument photos.

[0011] The instrument photograph is identified to obtain the identification result and reflective characteristics; the reflective characteristics include regional pixel gradient texture, regional pixel brightness, and target region contour shape; based on these reflective characteristics, it is determined whether reflection occurs;

[0012] If glare occurs, adjust the brightness of the supplementary light until accurate instrument data recognition results are obtained.

[0013] Preferably, the nighttime identification of the collected instrument data includes the following steps:

[0014] The target region in the instrument image is extracted by using a depth detection model, and the pointer region image and dial region image are obtained by using a depth semantic segmentation model.

[0015] Based on the pointer region image, a straight line representing the pointer position is obtained by skeleton extraction and distance transformation, and by line fitting.

[0016] Based on the dial area image, morphological filtering is used to calculate the transformation matrix using the dial outline ellipse, and the image points are corrected using the transformation matrix.

[0017] Preferably, the step of using morphological filtering to calculate the transformation matrix based on the dial area image, and then correcting image points using the transformation matrix, includes the following steps:

[0018] Morphological filtering is applied to the image of the dial area, and the outermost contour is extracted from the filtered image.

[0019] For a rectangular dial, straight lines are fitted to the four sides of the outline. The coordinates of the four intersection points are calculated using the four straight lines. The four points corresponding to the minimum bounding rectangle of the four points are found, and the transformation matrix is ​​calculated.

[0020] For a circular dial, ellipse fitting is performed using contour coordinate points. The fitted ellipse reflects the degree of dial deformation. The coordinate information of the four endpoints is calculated using the ellipse parameters obtained from the fitting. At the same time, the coordinates of the points corresponding to the four endpoints on the circle coinciding with the center of the ellipse are calculated. The transformation matrix M is calculated from these four pairs of coordinate points and used for the correction of points on the subsequent image.

[0021] Preferably, the depth detection model is a YOLOv4, YOLOv3, or SSD detection model.

[0022] Preferably, the deep semantic segmentation model is the DeepLabv3 model of MobileNetv3, or a deep segmentation model of SegNet, UET, or PSPNet.

[0023] Preferably, in the database, each record corresponds to the recognition result Fruit, ambient light intensity Intensity, camera zoom value Zoom, camera focus value Focus, and the degree of approximation to the true value Level;

[0024] Among them, the recognition result Fruit - the true value Value = the level of proximity.

[0025] The beneficial effects of this invention are:

[0026] This invention proposes a nighttime mirror reflection adjustment method for supplementary lighting based on big data analysis. This method leverages the strong adaptability and robustness of the depth detection model, achieving a target detection rate of over 95% for watch faces in natural working scenarios. This provides strong support for image correction. The overall method significantly enhances robustness against conditions such as label occlusion, watch face tilt and deformation, watch face dirt, and water vapor inside the watch face. Attached Figure Description

[0027] Figure 1 This is a data information flow diagram according to an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram of the minimized gimbal component assembly according to an embodiment of the present invention;

[0029] Figure 3 This is a diagram showing the actual test results of an embodiment of the present invention. Detailed Implementation

[0030] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0031] Example 1

[0032] This invention provides a method for adjusting the specular reflection of supplementary lighting at night based on big data analysis, such as... Figure 1-3 As shown.

[0033] S1: A method for adjusting the specular reflection of a fill light at night based on big data analysis, and establishing a gimbal model; This invention is based on a gimbal, so establishing a gimbal model is the foundation for realizing this invention, and it has the functions of visible light photography and adjustable light intensity control of the fill light.

[0034] Specifically, the modeling steps include the following:

[0035] Establish a remote sensing area instrument model in advance. C = {List <t>;MC;IP;IN;NC};

[0036] Model Description:

[0037] C (Cradlehead): Intelligent gimbal.

[0038] T(Task): Inspection task T. Then List <t>This is the inspection task list.

[0039] MC (Motion Controller): Based on the gimbal motion controller subsystem, MC = {Speed; Front(); Back(); TurnLeft(); TurnRight();};

[0040] IN(Instrument): Instrument parameter object;

[0041] IP (Image Processor): A visual image processing module IP based on AI pattern recognition algorithms;

[0042] NC (Network Communicator): A network communication component based on SocketTCP;

[0043] Obtain the minimum scale coordinate point (MinPoint), pointer axis center point (CenterPoint), maximum scale coordinate point (MaxPoint), and the minimum (MinValue) and maximum (MaxValue) of the dial.

[0044] S2: Take the image captured by the gimbal as input, combine it with the modeling information, compare it with the algorithm model library, and finally output the reflection characteristics table (resultsdb);

[0045] S2 includes the following steps:

[0046] 1) Extract the target region from the image acquired by the gimbal using a depth detection model;

[0047] 2) The dial and hands in the target region are segmented using a deep semantic segmentation model, resulting in images of the hand region and the dial region.

[0048] 3) For the pointer region image, through skeleton extraction and distance transformation, a straight line is finally fitted to represent the pointer position;

[0049] For the dial area image, morphological filtering is applied, and the transformation matrix M is calculated using the dial contour ellipse. The image points are then corrected using the transformation matrix, and finally the meter reading or reflective characteristics are calculated.

[0050] In step 3), for the dial area image, morphological filtering is first performed, and the outermost contour is extracted from the filtered image. For a rectangular dial, straight line fitting is performed on the four sides of the contour, and then the coordinates of the four intersection points are calculated using the four straight lines. The four points corresponding to the minimum bounding rectangle of the four points are found, and the transformation matrix is ​​calculated.

[0051] For a circular dial, ellipse fitting is performed using contour coordinate points. The fitted ellipse reflects the degree of dial deformation. The coordinate information of the four endpoints is calculated using the ellipse parameters obtained from the fitting. At the same time, the coordinates of the points corresponding to the four endpoints on the circle coinciding with the center of the ellipse are calculated. The transformation matrix M is calculated from these four pairs of coordinate points and used for the correction of points on the subsequent image.

[0052] The deep detection model is either YOLOv4, YOLOv3, or SSD. The deep semantic segmentation model is either the DeepLabv3 model of MobileNetv3, or a deep segmentation model such as SegNet, UET, or PSPNet.

[0053] S3: By comparing the approximation of 600,000 historical recognition results with actual instrument data, the adjustment range of camera zoom and magnification parameters is determined to achieve optimal recognition results. The specific solution is as follows:

[0054] 1) Through extensive nighttime testing, based on the predicted results, the corresponding parameters are adjusted to obtain the instrument recognition results when the camera is zoomed in and out at different magnifications (Zoom and Focus). The results are then normalized and stored in the database.

[0055] 2) Compare the acquired large amount of data with the actual values ​​in the pointer table to find the optimal parameter OP under different time periods and lighting conditions.

[0056] 3) Details: Data acquisition time interval TS, ambient light intensity Intensity, camera zoom value Zoom, camera focus value Focus.

[0057] 4) 12*3600 / TS = Number of items per day.

[0058] 5) For each record, the corresponding recognition results are Fruit, Intensity, Zoom, Focus, and Value (how close to the true value).

[0059] Formula: ABS (Fruit - Value) = Proximity Level

[0060] S4: Obtain the recognition result and reflectivity of the instrument image, and determine whether reflection has occurred based on the reflectivity. If reflection occurs, adjust the brightness of the supplementary light until accurate instrument data recognition results are obtained. Actual test results are as follows: Figure 3 As shown.

[0061] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.< / t> < / t>

Claims

1. A method for adjusting the specular reflection of supplementary lighting at night based on big data analysis, characterized in that, Includes the following steps: Build the database; The collected instrument data is used for nighttime identification to obtain the optimal parameters for the camera to identify the instruments under different time periods and lighting conditions, and these parameters are stored in the database. The parameters include time interval, ambient light intensity, camera zoom value, and zoom value. Move the pan-tilt unit with the fill light to the instrument where the image to be captured; The optimal parameters for that time period are retrieved from the database to adjust the camera parameters of the gimbal and to obtain instrument photos. The instrument photograph is identified to obtain the identification result and reflective characteristics; the reflective characteristics include regional pixel gradient texture, regional pixel brightness, and target region contour shape; based on these reflective characteristics, it is determined whether reflection occurs; If glare occurs, adjust the brightness of the supplementary light until accurate instrument data recognition results are obtained; The nighttime identification of the collected instrument data includes the following steps: The target region in the instrument image is extracted by using a depth detection model, and the pointer region image and dial region image are obtained by using a depth semantic segmentation model. Based on the pointer region image, a straight line representing the pointer position is obtained by skeleton extraction and distance transformation, and by line fitting. Based on the dial area image, morphological filtering is used to calculate the transformation matrix using the dial outline ellipse, and the image points are corrected using the transformation matrix. The process of using a dial area image, employing morphological filtering, calculating a transformation matrix based on the dial contour ellipse, and then correcting image points using the transformation matrix includes the following steps: Morphological filtering is applied to the image of the dial area, and the outermost contour is extracted from the filtered image. For a rectangular dial, straight lines are fitted to the four sides of the outline. The coordinates of the four intersection points are calculated using the four straight lines. The four points corresponding to the minimum bounding rectangle of the four points are found, and the transformation matrix is ​​calculated. For a circular dial, ellipse fitting is performed using contour coordinate points. The fitted ellipse reflects the degree of dial deformation. The coordinate information of the four endpoints is calculated using the ellipse parameters obtained from the fitting. At the same time, the coordinates of the points corresponding to the four endpoints on the circle that coincides with the center of the ellipse are calculated. The transformation matrix M is calculated from these four pairs of coordinate points and used for the correction of points on the subsequent image. In the database, each record corresponds to the recognition result Fruit, ambient light intensity Intensity, camera zoom value Zoom, camera focus value Focus, and the degree of closeness to the true value Level. Among them, the recognition result Fruit - the true value = the level of proximity.

2. The method for adjusting the specular reflection of a supplementary light at night based on big data analysis according to claim 1, characterized in that, The depth detection model is either YOLOv4, YOLOv3, or SSD.

3. The method for adjusting the specular reflection of a supplementary light at night based on big data analysis according to claim 1, characterized in that, The deep semantic segmentation model is the DeepLabv3 model of MobileNetv3, or a deep segmentation model of SegNet, UNET, or PSPNet.

Images