Image quality enhancement method based on light source intensity adaptive adjustment

By building a comprehensive data set and training light source intensity prediction model, adjusting the light source intensity of the mine rescue robot in real time, the problem of low image quality is solved, and clearer image acquisition and higher operating efficiency are achieved.

CN119071641APending Publication Date: 2024-12-03ZHEJIANG UNIV OF TECH

Patent Information

Application Number
CN202411063323.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The images captured by mine rescue robots in dark environments after the disaster are poor, fixed light sources lead to image overexposure and contrast reduction, and traditional image enhancement algorithms are difficult to effectively execute in environments with strict explosion-proof and battery life requirements.

Method used

Build a comprehensive data set, train a light source intensity prediction model, adjust the light source intensity in real time based on the distance and angle information from the light source to the target object, control the light source intensity through the PWM signal, realize adaptive light source intensity adjustment and improve image quality.

Benefits of technology

Achieve a more uniform image brightness distribution, reduce local over-light or over-dark phenomena, significantly improve image clarity and nature, reduce energy consumption, adapt to high temperature and poor ventilation environments, and improve the operation efficiency of rescue robots .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119071641A_ABST
    Figure CN119071641A_ABST
Patent Text Reader

Abstract

The invention discloses an image quality enhancement method based on light source intensity self-adaptive adjustment, which comprises the following steps of: firstly, constructing a data set, and sorting and screening the data set; then, training through the screened data set to obtain a light source intensity prediction model; then, according to the distance and angle information, fed back in real time in the moving process of the robot, from the light source to the target object, the optimal light source intensity is predicted through the light source intensity prediction model; and finally, converting the obtained optimal light source intensity into a PWM (Pulse Width Modulation) signal, and accurately controlling the brightness of the light source through a single chip microcomputer. According to the invention, the phenomenon that the image shot by the mine rescue robot is locally too bright or too dark is effectively reduced, and the details of the image are better reserved, so that the definition of the image is improved; in addition, by flexibly controlling the mechanical arm and accurately illuminating a specific position, the operation efficiency of the rescue robot in a complex and severe post-disaster mine environment is enhanced, the energy consumption is reduced, and the overall working efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of mine environment image enhancement, and in particular to an image quality enhancement method based on adaptive adjustment of light source intensity. Background Art

[0002] After a mine disaster occurs, a mine rescue robot can carry necessary emergency supplies and a variety of sensors to conduct preliminary surveys in the disaster area. Relying on the key information sent back by the robot, the rescue team can formulate an effective rescue plan. However, the lighting conditions in the mine after the disaster are extremely poor, close to complete darkness, which seriously affects the quality of the images taken by the camera, making it difficult for rescuers to obtain accurate underground environmental information and greatly limiting the robot's search and rescue capabilities.

[0003] To improve lighting conditions, a common practice is to install searchlights on the robot. Currently, the design of many mine rescue robots includes installing light sources on the fuselage to improve the lighting inside the mine. However, these lighting devices usually have fixed brightness. When the distance between the light source and the object changes, it is easy to cause problems such as image overexposure and contrast reduction, thereby affecting image clarity. In addition, since the position of the light source is fixed, the light may be blocked by obstacles, and the robot needs to adjust its position frequently when exploring a specific area, which not only increases energy consumption but also reduces efficiency. Although some solutions improve image clarity by adopting image enhancement algorithms, these algorithms require a lot of computing resources. In the actual use environment of mine rescue robots, these algorithms are often difficult to execute effectively because they must meet the strict requirements of explosion protection and endurance, coupled with the possible high temperature and poor ventilation conditions after the disaster. Therefore, it is urgent to develop an image quality enhancement technology suitable for mine rescue robots to solve the problem of low image quality captured by the camera on the current rescue robot. Summary of the invention

[0004] In view of the deficiencies in the prior art, an object of the present invention is to provide an image quality enhancement method based on adaptive adjustment of light source intensity, which is used to improve the image quality of a rescue robot when collecting images in a dark environment of a post-disaster mine.

[0005] The objective of the present invention is achieved through the following technical solution: A method for enhancing image quality based on adaptive adjustment of light source intensity, comprising the following steps:

[0006] 1. Construct a comprehensive data set: Construct a comprehensive data set, the data of which include: the intensity of the light source at the end of the robotic arm, the distance from the light source to the target object, the incident angle of the light source, and image quality evaluation parameters;

[0007] 2. Data set organization and screening: organize the comprehensive data set obtained in step 1, and screen the valid data to obtain a valid data set;

[0008] 3. Training the light source intensity prediction model: learning the effective data set screened in step 2, and training to obtain the light source intensity prediction model;

[0009] 4. Real-time prediction of light intensity: Based on the distance and angle information from the light source to the target object fed back during the robot's movement, the trained light intensity prediction model is used to predict the optimal light intensity;

[0010] 5. Light source intensity control: The optimal light source intensity obtained in step 4 is converted into a PWM signal, transmitted to the single chip microcomputer, and the light source intensity is adjusted to the optimal value.

[0011] Furthermore, the step 1 is specifically as follows:

[0012] Light source incident angle setting: In the simulated post-disaster mine environment, adjust the incident angle between the light source at the end of the robot arm and the plane where the target object is located, and set an angle line at every same angle;

[0013] Setting the distance from the light source to the target object: Start moving from the starting point of each angle along each set angle line in the working space of the robot arm, and move forward a fixed distance each time until reaching the set end point;

[0014] Light intensity setting at the end of the robot arm: gradually increase the light intensity from the lowest brightness to the highest brightness;

[0015] Data collection: gradually increase the light intensity from the lowest brightness to the highest brightness, take multiple pictures at each brightness level, and at the same time, take a set of control pictures at each sampling point, where the sampling point is under sufficient light intensity; calculate the standard deviation, information entropy and average gradient of each picture, and create a comprehensive data set including the intensity of the light source, the distance from the light source to the target object, the incident angle of the light source and the image quality evaluation parameters.

[0016] Furthermore, the step 1 further includes: establishing a relationship between the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, the incident angle of the light source and image quality evaluation parameters; the image quality evaluation parameters include: standard deviation, information entropy and average gradient of the image;

[0017] The light source intensity I on the object surface is expressed as:

[0018]

[0019] Where L represents the intensity of the light source emitted by the end of the robot arm, i represents the incident angle of the light source, d represents the distance from the light source to the target object, and ρ d (x, y) represents the reflectivity distribution of the object surface;

[0020] The relationship between the standard deviation SD of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as:

[0021]

[0022] Among them, α is a constant related to reflectivity. Indicates the standard deviation of the reflectivity of the object surface;

[0023] The relationship between the information entropy H of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as:

[0024] H=log2(L)+log2(cos(i))-2log2(d)+C

[0025] Where C is a constant, e represents the base of natural logarithm;

[0026] The relationship between the average gradient E[AG] of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as:

[0027]

[0028] in, ω x and ω y are the gradient weighting coefficients in the x and y directions.

[0029] Furthermore, the step 2 is specifically as follows:

[0030] Calculate statistical indicators: calculate the mean and standard deviation of image quality evaluation parameters of photos taken at each sampling point, the image quality evaluation parameters include: standard deviation, information entropy, average gradient;

[0031] Setting parameter range: Adopting the “mean ± standard deviation” method, add the mean and standard deviation of each image quality evaluation parameter to determine the effective range of each image quality evaluation parameter;

[0032] Determination of threshold and data screening: Determine the specific threshold of each image quality evaluation parameter according to the effective range; use the threshold as a standard to screen out data that meets the conditions from the data set to obtain a valid data set.

[0033] Furthermore, in step 3, training the light source intensity prediction model includes the following steps:

[0034] Formulate comprehensive image quality scores: Normalize image quality evaluation parameters; set weights for each parameter and add up the weighted values ​​to establish a comprehensive image quality evaluation score;

[0035] Data stratification and sampling: Data that meets the threshold conditions are stratified and sampled. In each layer, the data of one image is randomly selected as part of the test set, and the rest of the data is used for the training set;

[0036] Model training: The intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source are set as feature inputs, and the comprehensive quality evaluation score of the image is used as the target variable. The random forest algorithm is used to train these data to establish the optimal light source intensity prediction model;

[0037] Model evaluation: The mean square error coefficient (RMSE) and the coefficient of determination (R2 score) are used as evaluation indicators to evaluate the prediction accuracy and prediction ability of the model on the test set.

[0038] Furthermore, the step 4 is specifically as follows:

[0039] Distance and angle information acquisition: In the ROS framework, the depth camera at the end of the robotic arm obtains the depth image through OpenCV, calculates the distance from the light source to the target object, and publishes the distance information in the form of a topic; at the same time, the robotic arm obtains the end position through Moveit, calculates the angle between the light source and the target object, and also publishes the angle information in the form of a topic;

[0040] Light source intensity prediction: The light source intensity prediction model subscribes to the published distance and angle information, predicts the image quality scores under different intensities within the effective range of light source intensity, and determines the light source intensity corresponding to the highest score as the optimal light source intensity.

[0041] Furthermore, the step 5 is specifically as follows:

[0042] Establishing the relationship between PWM value and light source intensity: By collecting the intensity of the light source at the end of the robotic arm and the corresponding PWM value data, linear fitting is performed to establish a linear relationship between PWM and the intensity of the light source on the robotic arm;

[0043] Calculation and conversion of PWM values: According to the linear relationship between PWM and the intensity of the light source on the surface of the object, the predicted optimal light source intensity value is converted into the corresponding PWM value;

[0044] The microcontroller adjusts the light source intensity: the calculated PWM value is published through the ROS system, the microcontroller is connected to the ROS system, the PWM value is subscribed in real time, and the light source is adjusted to the intensity predicted by the light source intensity prediction model to achieve precise control of the light source brightness.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. Adaptive light source intensity adjustment: The light source installed at the end of the mechanical arm of the mine rescue robot of the present invention can automatically adjust the light source intensity through the light source intensity prediction model according to the distance and angle information between the light source and the target object fed back in real time. This adaptive adjustment achieves a more uniform image brightness distribution, effectively reduces the phenomenon of local overbrightness or overdarkness in the image, better preserves the details of the image, and thus obtains a clearer picture.

[0047] 2. Direct improvement of image quality: Compared with the traditional method of using a light source with fixed brightness and then improving image quality through subsequent image enhancement technology, the present invention obtains high-quality images by directly increasing the brightness of the environment, which significantly improves the naturalness and practicality of the image effect and enhances the operating efficiency of the rescue robot in complex environments.

[0048] 3. Low energy consumption: The light intensity prediction model in the present invention does not require a large amount of computing resources to run, which meets the strict requirements of rescue robots in terms of explosion-proof performance and endurance, and is adapted to the special needs of harsh environments such as high temperature and poor ventilation in post-disaster mines. In addition, the present invention also allows the robot to accurately illuminate specific locations by flexibly controlling the mechanical arm when exploring the environment. Compared with the fixed light source on the body, there is no need to move the body multiple times, thereby reducing energy consumption and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of the present invention;

[0050] Figure 2 A location map of sampling points for constructing a comprehensive data set experiment for the present invention;

[0051] Figure 3 This is a schematic diagram of the light source at the end of the robot arm of the present invention being 600 mm away from the target object and the incident angle being 60°;

[0052] Figure 4 This is an interpolation curve diagram of light source intensity and image quality score of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with the accompanying drawings:

[0054] like Figure 1 As shown, an image quality enhancement method based on adaptive adjustment of light source intensity is described, and the detailed steps are as follows:

[0055] Step 1: Build a comprehensive dataset

[0056] The steps of constructing a comprehensive data set including the intensity of the light source at the end of the robot arm, the distance to the target object, the incident angle, and the corresponding image quality evaluation parameters include:

[0057] Light source incident angle setting: Figure 2 As shown, it is a sampling point location map for constructing a comprehensive data set experiment. In this embodiment, in a simulated post-disaster mine environment, the angle of the light source fixed at the end of the robot's mechanical arm is adjusted so that the incident angle between the light source and the plane where the target object is located is adjusted from 30 degrees to 90 degrees, and an angle line is set every 5 degrees.

[0058] Setting the distance from light source to target object: also refer to Figure 2 In this embodiment, the robot moves along each set angle line in the working space of the robot arm from the starting point (the farthest point) of each angle, moving forward 50 mm each time until reaching the set end point.

[0059] Light source intensity setting: In this embodiment, the data collection experiment uses a white LED light source, and the luminous intensity is set between 3000 lux and 33000 lux, with 1500 lux as an adjustment step.

[0060] Data collection: At each sampling point, the light intensity gradually increases from the lowest brightness to the highest brightness, and 5 pictures are taken at each brightness level. At the same time, a set of control pictures are taken at each sampling point under sufficient light intensity in the laboratory. In this embodiment, a total of 10,940 pictures are taken under varying light intensity, and a total of 2,188 pictures are taken under sufficient light intensity in the laboratory.

[0061] Establish the relationship between the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, the incident angle of the light source and the image quality evaluation parameters; the image quality evaluation parameters include: the standard deviation, information entropy and average gradient of the image.

[0062] The light source intensity I on the object surface is expressed as:

[0063]

[0064] Where L represents the intensity of the light source emitted by the end of the robot arm, i represents the incident angle of the light source, d represents the distance from the light source to the target object, and ρ d (x,y) represents the reflectivity distribution of the object surface.

[0065] The relationship between the standard deviation SD of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as:

[0066]

[0067] Among them, α is a constant related to reflectivity. Represents the standard deviation of the reflectivity of the object surface.

[0068] The relationship between the information entropy H of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as:

[0069] H=log2(L)+log2(cos(i))-2log2(d)+C

[0070] Where C is a constant, e represents the base of natural logarithms.

[0071] The relationship between the average gradient E[AG] of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as:

[0072]

[0073] in, ω x and ω y are the gradient weighting coefficients in the x and y directions.

[0074] The standard deviation, information entropy and average gradient of each image are calculated to create a comprehensive data set including the intensity of the light source, the distance to the target object, the incident angle and the image quality evaluation parameters.

[0075] Step 2: Dataset organization and screening

[0076] The steps to sort out the data set and filter out the data in the data set include:

[0077] Calculate statistical indicators: Calculate statistical indicators of image quality evaluation parameters in images taken under sufficient lighting conditions in the laboratory. In this embodiment, the calculated means of standard deviation, information entropy and average gradient are 45.20, 5.07 and 39.37 respectively; the corresponding standard deviations are 12.28, 0.29 and 7.64 respectively.

[0078] Set parameter range: Use the "mean ± standard deviation" method to determine the effective range of each parameter. In this embodiment, the effective range of the standard deviation is 39.09 to 52.04, the effective range of the information entropy is 5.02 to 5.25, and the effective range of the average gradient is 43.35 to 61.66.

[0079] Threshold determination and data screening: According to the set parameter range, the specific threshold of each image quality evaluation parameter is determined. In this embodiment, the thresholds are set as follows: the standard deviation is not less than 40, the information entropy is not less than 5.1, and the average gradient is not less than 47. According to these thresholds, the data that meets the conditions are screened out from the data set. Finally, under the condition of changing light source conditions, 10691 pieces of image data that meet the threshold are screened out from a total of 10940 data to obtain a valid data set.

[0080] Step 3: Train the light intensity prediction model

[0081] The steps of using the random forest algorithm to learn the filtered valid data set and train the light source intensity prediction model include:

[0082] Formulate comprehensive image quality score: This embodiment first uses the MinMaxScaler method to normalize the three image quality evaluation indicators of standard deviation, information entropy and average gradient in the data set, so that the data is scaled to the range of [0,1]. Based on these normalized indicators, according to the intensity of the light source, the distance to the target object, the angle of incidence and the correlation coefficient of each image quality evaluation indicator, the weights are set as follows: standard deviation is 0.4, information entropy is 0.35, and average gradient is 0.25. These normalized parameters are multiplied by the corresponding weights and summed to obtain the comprehensive image quality score of each picture.

[0083] Data stratification and sampling: In this embodiment, in the filtered data set, multiple pieces of image quality rating data under the same light source condition are defined as one layer. Through a custom data partitioning function, one picture is randomly selected in each layer as part of the test set, and the remaining pictures are used as the training set.

[0084] Model training: In the training phase, this embodiment uses the intensity of the light source, the distance to the target object, and the angle of incidence as feature inputs, and the comprehensive quality score of the image as the target variable. The random forest algorithm is used to train these data to establish the optimal light source intensity prediction model.

[0085] Model evaluation: This example uses the root mean square error (RMSE) and the coefficient of determination (R2 score) as evaluation indicators. On the test set, the model achieved a root mean square error (RMSE) of 0.01865 and a coefficient of determination (R2) of 0.9746, which shows that the trained model has a high degree of prediction accuracy and strong prediction ability.

[0086] Step 4: Real-time prediction of light intensity

[0087] The steps of using the trained model to predict the optimal light source intensity based on the feedback angle and distance information during the robot movement include:

[0088] Distance and angle information acquisition: Figure 3 As shown, in the embodiment, the depth camera installed at the end of the robotic arm feeds back the distance from the light source to the target object as 600 mm. The robotic arm obtains the end position through the MoveIt control framework and calculates the angle between the light source and the target object, which is 60° in this example. These angle and distance information are published in the form of ROS topics.

[0089] Light source intensity prediction: The light source intensity prediction model in this embodiment subscribes to the published distance and angle information. The model evaluates the image quality scores at different intensities at intervals of 1000 lux within the operating range of light source intensity (3000 to 33000 lux). Figure 4 As shown in the figure, the model generates a curve of light source intensity and image quality score, and determines the optimal light source intensity corresponding to the highest score, which is 19473 lux.

[0090] Step 5: Light source intensity control

[0091] The predicted light source intensity value is converted into a PWM signal and transmitted to the microcontroller to control the light source. The steps of adjusting its intensity to the optimal value include:

[0092] Establishment of the relationship between PWM value and light source intensity: This embodiment uses the least square method to perform linear fitting on the light source intensity and the corresponding PWM value data collected in the experiment, and obtains the linear relationship between PWM and light source intensity as follows:

[0093] P=0.008894×I+7.177

[0094] Where P is the PWM value and I is the light source intensity (unit: lux).

[0095] Calculation and conversion of PWM value: According to the above linear relationship, this embodiment converts the predicted optimal light source intensity value 19473 lux into the corresponding PWM value, and the calculation result is 180.49.

[0096] The single-chip microcomputer adjusts the light source intensity: In this embodiment, the calculated PWM value is sent to the single-chip microcomputer in the light source control module in the form of a ROS topic. After receiving the PWM value, the single-chip microcomputer adjusts the light source to the predicted intensity to achieve accurate control of the light source brightness.

[0097] The image processed by the adaptive light source intensity adjustment method of this embodiment is compared with the original image. The results show that the adjusted image shows a more uniform brightness distribution, significantly reduces the phenomenon of local over-brightness or over-darkness, and effectively retains more detail information. In addition, the adjusted image avoids the problems of overexposure and loss of details, thereby significantly improving the overall quality of the image. This improvement not only enhances the visual effect of the image, but also optimizes the image processing capability of the rescue robot under complex lighting conditions.

Claims

1. A method for enhancing image quality based on adaptive adjustment of light source intensity, characterized in that: The steps include: (1) constructing a comprehensive data set, wherein the data in the data set include: the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, the incident angle of the light source, and image quality evaluation parameters; (2) arranging the comprehensive data set obtained in step (1) and screening valid data to obtain a valid data set; (3) learning the effective data set screened in step (2) and training a light source intensity prediction model; (4) Based on the distance and angle information from the light source to the target object fed back during the robot's motion, the trained light source intensity prediction model is used to predict the optimal light source intensity; (5) Convert the optimal light source intensity obtained in step (4) into a PWM signal, transmit it to the single-chip microcomputer, and adjust the light source intensity to the optimal value.

2. The image quality enhancement method based on adaptive adjustment of light source intensity according to claim 1, characterized in that: The step (1) is specifically: Light source incident angle setting: In the simulated post-disaster mine environment, adjust the incident angle between the light source at the end of the robot arm and the plane where the target object is located, and set an angle line at every same angle; Setting the distance from the light source to the target object: Start moving from the starting point of each angle along each set angle line in the working space of the robot arm, and move forward a fixed distance each time until reaching the set end point; Light intensity setting at the end of the robot arm: gradually increase the light intensity from the lowest brightness to the highest brightness; Data collection: gradually increase the light intensity from the lowest brightness to the highest brightness, take multiple pictures at each brightness level, and at the same time, take a set of control pictures at each sampling point, where the sampling point is under sufficient light intensity; calculate the standard deviation, information entropy and average gradient of each picture, and create a comprehensive data set including the intensity of the light source, the distance from the light source to the target object, the incident angle of the light source and the image quality evaluation parameters.

3. The image quality enhancement method based on adaptive adjustment of light source intensity according to claim 1, characterized in that: The step (1) further includes: establishing a relationship between the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, the incident angle of the light source and image quality evaluation parameters; the image quality evaluation parameters include: standard deviation, information entropy and average gradient of the image; The light source intensity I on the object surface is expressed as: Where L represents the intensity of the light source emitted by the end of the robot arm, i represents the incident angle of the light source, d represents the distance from the light source to the target object, and ρ d (x, y) represents the reflectivity distribution of the object surface; The relationship between the standard deviation SD of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as: Among them, α is a constant related to reflectivity, Indicates the standard deviation of the reflectivity of the surface of an object; The relationship between the information entropy H of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as: H=log2(L)+log2(cos(i))-2log2(d)+C Where C is a constant, e represents the base of natural logarithm; The relationship between the average gradient E[AG] of the image and the intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source is expressed as: in, ω x and ω y are the gradient weighting coefficients in the x and y directions.

4. The image quality enhancement method based on adaptive adjustment of light source intensity according to claim 1, characterized in that: The step (2) is specifically: Calculate the mean and standard deviation of image quality evaluation parameters of the photos taken at each sampling point, wherein the image quality evaluation parameters include: standard deviation, information entropy, and average gradient; Adding the mean and standard deviation of each image quality evaluation parameter to determine the effective range of each image quality evaluation parameter; According to the effective range, a specific threshold value of each image quality evaluation parameter is determined; based on the threshold value, data that meets the conditions is screened out from the comprehensive data set to obtain a valid data set.

5. The image quality enhancement method based on adaptive adjustment of light source intensity according to claim 1, characterized in that: In the step (3), training the light source intensity prediction model comprises the following steps: Normalize the image quality evaluation parameters; set weights for each parameter and add up the weighted values ​​to establish a comprehensive quality evaluation score for the image; The data that meets the threshold condition is sampled in layers. In each layer, one image’s data is randomly selected as part of the test set, and the rest of the data is used for the training set. The intensity of the light source at the end of the robot arm, the distance from the light source to the target object, and the incident angle of the light source are set as the feature input, and the comprehensive quality evaluation score of the image is used as the target variable. The random forest algorithm is used to train these data to establish the optimal light source intensity prediction model. The mean square error coefficient and determination coefficient were used as evaluation indicators to evaluate the prediction accuracy and prediction ability of the model on the test set.

6. The image quality enhancement method based on adaptive adjustment of light source intensity according to claim 1, characterized in that: The step (4) is specifically as follows: In the ROS framework, the depth camera at the end of the robotic arm obtains the depth image through OpenCV, calculates the distance from the light source to the target object, and publishes the distance information in the form of a topic. At the same time, the robotic arm obtains the end position through Moveit, calculates the angle between the light source and the target object, and also publishes the angle information in the form of a topic. The light source intensity prediction model subscribes to the published distance and angle information, predicts the image quality scores under different intensities within the effective range of light source intensity, and determines the light source intensity corresponding to the highest score as the optimal light source intensity.

7. The image quality enhancement method based on adaptive adjustment of light source intensity according to claim 1, characterized in that: The step (5) is specifically as follows: By collecting the intensity of the light source at the end of the robot arm and the corresponding PWM numerical data, linear fitting is performed to establish a linear relationship between PWM and the intensity of the light source on the robot arm; According to the linear relationship between PWM and the intensity of the light source on the surface of the object, the predicted optimal light source intensity value is converted into the corresponding PWM value; The calculated PWM value is published through the ROS system, and the ROS system is connected through the single-chip microcomputer to subscribe to the PWM value in real time, and the light source is adjusted to the intensity predicted by the light source intensity prediction model to achieve precise control of the light source brightness.

Citation Information

Patent Citations

  • Fluorescent area light intensity light supplementing method, device and system for parathyroid gland detection

    CN113376931A

  • Self-adaptive dimming method for LED cold light source of endoscope

    CN117750214A

  • Combined lighting apparatus and method based on image quality control

    WO2016078598A1

  • Light supplementing method and apparatus for facial recognition, and facial recognition device and system therefor

    WO2022166316A1

Cited By

  • ROV underwater adaptive illumination algorithm, storage medium and equipment

    CN119676574A