An imaging parameter optimization method for mobile vision systems

Abstract

The present invention relates to an imaging parameter optimization method for a mobile vision system, comprising: obtaining imaging parameters that affect image quality, and constraining the imaging parameters to obtain image data; constructing a mapping model, and obtaining several groups of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained through training with a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network; with the maximum expected value of the image quality evaluation value as the target, the several groups of imaging parameters are optimized to obtain the optimal image imaging system parameter combination. The present invention combines the intelligent optimization capability of deep learning to automatically adjust imaging parameters to adapt to different scenes and motion states, thereby improving the imaging effect and application performance of mobile machine vision systems in various environments.

Description

Technical Field

[0001] The present invention relates to the field of machine vision technology, and in particular to an imaging parameter optimization method for a mobile vision system. Background Art

[0002] Mobile vision systems play a key role in many fields, such as autonomous driving, intelligent monitoring, and industrial production. However, the imaging effect of mobile vision systems in different environments is affected by many factors, including lighting conditions, camera performance, motion blur, etc. Therefore, it is necessary to optimize the imaging parameters in a targeted manner to improve image quality and system performance. However, traditional imaging parameter setting methods have certain limitations. Traditional imaging parameter optimization methods are mainly based on experience and trial and error, lacking systematicity and efficiency. For example, when adjusting parameters such as exposure time, aperture size, and ISO sensitivity, multiple trials are usually required to obtain the best results, which consumes time and resources.

[0003] In addition, most of the existing automatic optimization methods are designed for static scenes. For mobile vision systems in dynamic environments, especially scenes with high-speed motion or complex actions, these methods are often unable to effectively cope with them, resulting in frequent image blur or distortion, affecting the real-time and reliability of the system.

[0004] At present, with the continuous development of artificial intelligence and computer vision technology, imaging parameter optimization methods based on deep learning have gradually attracted attention. These methods use neural network models and large-scale data sets to automatically adjust imaging parameters to adapt to different environments and task requirements through learning and optimization processes. However, existing deep learning-based methods still have some problems, including high computational complexity and large training data requirements.

[0005] On the other hand, the imaging parameter optimization method based on traditional image processing and optimization algorithms can meet the needs of mobile machine vision systems to a certain extent, but it has limitations in processing complex scenes and improving image quality. Therefore, an imaging parameter optimization method that combines traditional image processing and optimization algorithms with deep learning technology is needed, which can achieve adaptive adjustment in dynamic environments to obtain clear and stable images and ensure the real-time and robustness of the system. Summary of the invention

[0006] The purpose of the present invention is to provide an imaging parameter optimization method for a mobile vision system, which solves the problem that existing automatic optimization methods are difficult to effectively cope with high-speed motion scenes, difficult to handle complex scenes and improve image quality.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] On the one hand, an imaging parameter optimization method for a mobile vision system is provided, comprising:

[0009] Acquiring imaging parameters that affect image quality, and constraining the imaging parameters to acquire image data;

[0010] Constructing a mapping model, and obtaining a plurality of sets of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained by training a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network;

[0011] Taking the maximum expected value of the image quality evaluation value as a target, the plurality of imaging parameters are optimized to obtain the optimal combination of imaging system parameters.

[0012] Optionally, obtaining imaging parameters that affect image quality includes:

[0013] The influence of different imaging parameters on image quality is analyzed by statistical methods, and imaging parameters including exposure time, light source intensity, and the distance between the target and the camera are obtained.

[0014] Optionally, acquiring the image data includes:

[0015] Screening the image data to obtain high-quality images;

[0016] The high-quality image is normalized, denoised, resized, and color-balanced to obtain the image data.

[0017] Optionally, obtaining the image quality evaluation result includes:

[0018] Clarity, power spectrum and signal-to-noise ratio are selected as image quality evaluation indicators, and the image quality evaluation indicators are weighted by entropy weight method to obtain image quality evaluation results corresponding to the image data, wherein the clarity is obtained by Brenner gradient function.

[0019] Optionally, optimizing the imaging parameters with the maximum expected value of the image quality quantization result as a target includes: optimizing the several groups of imaging parameters using a genetic algorithm to obtain an optimal combination of imaging system parameters.

[0020] Optionally, optimizing the plurality of imaging parameters by using a genetic algorithm includes:

[0021] S1, taking the several groups of imaging parameters as variables to be optimized, performing decimal encoding on the variables to be optimized, and initializing the population;

[0022] S2, calculating the fitness of the variables to be optimized encoded in the initialized population through a fitness function, and outputting imaging parameters;

[0023] S3, input the output imaging parameters into the mapping model, obtain the quantization result of the image quality evaluation value corresponding to the output imaging parameters, and determine whether the output quantization result meets the quality evaluation result of the optimal quality image. If so, proceed to S4; if not, perform individual selection, crossover and mutation, obtain the next generation population and return to S2;

[0024] S4. When the output imaging parameters obtained by optimization satisfy the quality evaluation result of the optimal quality image, the optimal image imaging system parameter combination is obtained.

[0025] Optionally, the objective function of the genetic algorithm is:

[0026] k(e,l,d)=|g max -g(e,l,d)|

[0027] Among them, g max is the maximum expected value of image quality, g(e,l,d) is the output result of the mapping model, and k(e,l,d) is the objective function of the genetic algorithm.

[0028] Optionally, the fitness function is:

[0029] Among them, g(e,l,d) is the output result of the mapping model.

[0030] On the other hand, an imaging parameter optimization system for a mobile vision system is provided, comprising:

[0031] A data acquisition module, used to obtain imaging parameters that affect image quality, and constrain the imaging parameters to obtain image data;

[0032] A mapping module, used to construct a mapping model, and obtain a plurality of sets of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained by training a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network;

[0033] The optimization module is used to optimize the plurality of imaging parameters with the maximum expected value of the image quality evaluation value as the target, so as to obtain the optimal combination of imaging system parameters.

[0034] Optionally, the data acquisition module includes: a mobile platform, an image forming device, a navigation and control unit, an image acquisition unit, and an imaging parameter acquisition unit;

[0035] The mobile platform is used to move within the target area;

[0036] The image forming device is used to form an image of a moving target object;

[0037] The navigation and control unit is used to control the motion trajectory of the mobile platform;

[0038] The image acquisition unit is used to acquire image data of the moving target object based on the image imaging device;

[0039] The imaging parameter acquisition unit analyzes the influence of different imaging parameters on image quality through a statistical method, and acquires imaging parameters including exposure time, light source intensity, and the distance between the target object and the camera.

[0040] The beneficial effects of the present invention are:

[0041] 1. The present invention can dynamically adjust according to the real-time environmental conditions and requirements of the mobile vision system, thereby achieving more flexible and intelligent imaging parameter optimization. By optimizing the imaging parameters, it can better adapt to image acquisition under different environmental conditions, thereby improving the clarity of the image and further improving the overall imaging quality.

[0042] 2. The invention can accurately optimize imaging parameters according to actual needs, avoid unnecessary waste of storage resources, and thus improve the efficiency and performance of the system.

[0043] 3. The optimized image of the present invention has better visual effects and can provide a more satisfactory user experience, especially in the shooting and display scenarios of mobile devices.

[0044] 4. Through the automated imaging parameter optimization method, the need for manual intervention can be reduced, the operational complexity can be reduced, and the ease of use and universality of the system can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 A flow chart of an imaging parameter optimization method for a mobile vision system according to an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of a mobile visual image imaging system according to an embodiment of the present invention;

[0048] Figure 3 A flow chart for determining factors affecting image quality according to an embodiment of the present invention;

[0049] Figure 4 is a flow chart of the entropy weight method according to an embodiment of the present invention;

[0050] Figure 5 A flow chart for constructing a convolutional neural network mapping model according to an embodiment of the present invention;

[0051] Figure 6 A flow chart of training and deploying a model according to an embodiment of the present invention;

[0052] Figure 7 The figure is a flow chart of optimizing parameters of an image imaging system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Embodiment 1

[0056] An imaging parameter optimization method for a mobile vision system includes:

[0057] Acquiring imaging parameters that affect image quality, and constraining the imaging parameters to acquire image data;

[0058] Constructing a mapping model, and obtaining a plurality of sets of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained by training a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network;

[0059] Taking the maximum expected value of the image quality evaluation value as a target, the plurality of imaging parameters are optimized to obtain the optimal combination of imaging system parameters.

[0060] In this embodiment, Figure 1 As shown, a method for optimizing imaging parameters of a mobile vision system comprises the following steps:

[0061] S1. Design and build a mobile visual imaging system: During the entire process of the target object moving toward or away from the car, the regional light provided by the light source is reflected into the camera lens to complete the imaging of the corresponding area.

[0062] S2. By analyzing the influence of the imaging system parameters designed in step S1 on the mobile imaging quality, factors having a greater impact on the image quality are determined.

[0063] S3, keep other parameters unchanged, set constraints, and only change the factors with greater influence in step S2. During the whole process of the target object's movement, the imaging system uses the CAN bus to realize control and data communication, set the parameters of the factors with greater influence, and then obtain the initial image data and imaging parameter data under different imaging parameter conditions.

[0064] S4. Screen the images obtained in step S3. If there are many interferences or the features are not obvious or covered, it is necessary to screen the images and then analyze the characteristics of the whole moving process images obtained by the mobile visual imaging system. Combined with the image quality evaluation index, the image data is quantitatively evaluated.

[0065] S5. Perform normalized data processing on the images filtered in step S4 to solve the dimension inconsistency problem of the quantization results, so that the quantization results remain in the range of [0,1]. Images with quantization results less than 0.02 do not meet the evaluation requirements, and the image data that meets the evaluation requirements is saved locally.

[0066] S6. The image quality evaluation index in step S4 is weighted using the entropy weight method, and the image evaluation result data is constructed into a data set and saved locally.

[0067] S7. Preprocess the image data obtained in step S5, including noise removal, resizing, and color balancing, to ensure data quality and consistency. Use a convolutional neural network to build a mapping model between image imaging system parameters and image quality throughout the mobile process. Design the structure of the convolutional neural network, including convolutional layers, pooling layers, fully connected layers, etc., as well as the selection of activation functions. The input of the network includes image imaging system parameters, which can be used as additional channels or input into the fully connected layer. The number of neurons in the output layer matches the dimension of the quantitative evaluation results.

[0068] S8. Combined with the mapping model in step S7, the quality of the image during the whole process of the moving car is improved by using a genetic algorithm. First, based on the mapping model, several sets of imaging parameters corresponding to the preset image quality evaluation values ​​are obtained, and the imaging system parameters are converted into variables to be optimized; then, the image quality value of the whole process of the moving car is converted into an objective function, and an optimization algorithm is selected at the same time, and the mapping model is combined to optimize the variables to be optimized, so that the image quality of the whole process of the moving car is the best.

[0069] like Figure 3 As shown, the specific process of determining the factors that have a greater impact on image quality in step S2 is:

[0070] First, list the possible parameters that affect image quality, such as: photosensitive element (sensor) type and size, lens quality and aperture size, exposure time, light source intensity, target-camera distance, etc. Then, design a series of experiments to change these parameters one by one and record the impact of each parameter change on image quality. Use a certain number of sample images covering different scenes and conditions to ensure the representativeness and reliability of the experiment. Evaluate and analyze the experimental data through visual evaluation or using image quality evaluation indicators (such as SSIM, PSNR, etc.). Compare the image quality differences under different parameter settings. Finally, use statistical methods (such as analysis of variance) to obtain the most significant effects of exposure time, light source intensity, and target-camera distance parameters on image quality.

[0071] In step S3, other parameters are kept unchanged, and only the light source intensity, exposure time, and the distance between the target object and the camera in step S2 are changed. The specific steps are as follows:

[0072] For the light source intensity, it can be achieved by adjusting the brightness of the light source; for the exposure time, it can be achieved by adjusting the exposure value of the camera. In the actual imaging process, in order to ensure the feasibility of the subsequent optimization results and improve the efficiency of the optimization process, these parameters need to be reasonably constrained and limited within a certain range to ensure that the final optimization results can meet the actual imaging requirements and reduce the number of adjustments to the imaging system parameters. The constraints on the variables to be optimized are set as e min ≤e≤e max , l min ≤l≤l max ,d min ≤d≤d max .

[0073] The image quality evaluation index in step S4, for this embodiment, uses the Brenner gradient function to quantitatively evaluate the image clarity. The function calculates the clarity of the image to be evaluated by using the grayscale difference between two pixels separated by two units. Assuming that the grayscale at point f of the acquired image is f(x, y), the calculation formula of the clarity Brenner(f) is: The power spectrum can quantitatively evaluate the blur degree of the moving image. In the whole moving process image acquisition process performed in this embodiment, different camera parameter combination settings will cause the image features to become blurred, causing the high-frequency components of the target area features to be lost, resulting in a decrease in the power spectrum value. The calculation formula for the power spectrum is: In the formula, M and N refer to the length and width pixel values ​​of the image respectively, MxN refers to an image with a pixel size of MxN, and u and v are respectively 0, 1, ..., M-1 and 0, 1, ..., N-1. ; u<M-1; v<N-1. The higher the value of the power spectrum, the more complete the edge of the extracted image feature. In addition, the larger the quantization value of the signal-to-noise ratio of the image, the less noise interference the image is affected by, and the better the image quality. Take the maximum background grayscale of the image as 255, and the signal-to-noise ratio calculation formula is Among them, σ is the standard deviation of the data. The image with a high noise ratio quantization value calculated by this formula is less affected by noise interference and has a higher image quality. However, the image with a low signal-to-noise ratio quantization value has more noise interference, and the image quantization score is low, which can be considered that the image quality is poor. Therefore, clarity, power spectrum and signal-to-noise ratio are selected as image quality evaluation indicators.

[0074] The normalized data processing method in step S5 is specifically as follows:

[0075] Use the maximum value x of a column of data max With the minimum value x min , using the normalization formula The original data set of this column is processed. After the data is processed by the normalization method, all values ​​in the data set will be compressed into the range of [0,1], so as to achieve the purpose of keeping the dimension of the processed data x' consistent.

[0076] The specific steps of constructing the data set in step S5 are:

[0077] First, the factors that have the greatest impact on image quality are determined to be exposure time, light source intensity, and the distance between the target and the camera. These three factors are represented by e, l, and d, respectively. Based on the experience of manually adjusting the parameters of the mobile imaging imaging system, the exposure time e is set to 110μs to 310μs; the light source intensity l is set to an ISO value between 100 and 1600; and the imaging distance d is set to 0cm to 300cm. During the image acquisition process, each time the imaging parameters are changed, the obtained image is named in the format of light source intensity-camera exposure time-distance between the detected object and the camera-suffix through the Python xlwings library and output to the Excel table together with the image quality evaluation index results.

[0078] like Figure 4 As shown, the entropy weight method in step S6 constructs a score matrix through the quantitative evaluation value of the indicator, calculates the weight of the evaluation indicator, and then obtains the weight of the corresponding indicator according to the degree of change of the quantitative evaluation value of the indicator. Finally, in the single indicator evaluation data set, the greater the degree of change of the quantitative evaluation value of the indicator, the higher the weight it occupies. The specific steps for obtaining the indicator weight by the entropy weight method are:

[0079] First, construct the evaluation score matrix X nxm =(x ij ) nxm , and normalize it to obtain the dimensionless evaluation matrix Z of the evaluation index score nxm . Where n is the total number of evaluation indicators, m is the total number of evaluation objects, i = 1, 2, ..., n; j = 1, 2, ..., m. Then, calculate the proportion of the indicator to the sample. By dimensionless processing, calculate the proportion of the i-th indicator and the j-th evaluation object to the entire sample. The calculation formula is Calculate the entropy value of each indicator, the calculation formula is Among them, e i represents the entropy value of the evaluation index i. Then calculate the deviation degree of each index, and the calculation formula is g ei =1-e i , i = 1, 2, ..., n; j = 1, 2, ..., m. Finally, the indicator weight is calculated. The weight calculation formula for the i-th indicator is:

[0080] The image data obtained in step S7 is preprocessed, and the specific steps are: using a Gaussian filter to remove noise in the image, the formula is Among them, m and n refer to the length and width pixel values ​​of the image respectively; σ is the variance, and the non-local mean method is used for denoising. The image is converted to the frequency domain using wavelet transform to remove noise in the frequency domain. Then, according to the needs of the specific task, the image is resized to an appropriate size, ensuring that the aspect ratio of the image is maintained during the resizing process to avoid image deformation. The contrast and brightness of the image are enhanced by adjusting the pixel value distribution of the image; the image is divided into small blocks and histogram equalization is performed separately to avoid over-enhancement caused by global processing.

[0081] In step S7, a convolutional neural network is used to construct a mapping model between the image imaging system parameters and the image quality of the entire moving process, such as Figure 5 As shown, the specific steps are:

[0082] The neural network mapping model is implemented by calling CNN (convolutional neural network regressor) in torch using Python 3.8. In the single evaluation index quantization value mapping model, the number of input features of the model is 3, which are the imaging parameters of the target object, including light source intensity, camera exposure time, and the distance between the detected object and the camera; the output value is the quantization value of the image evaluation index, including the target image signal-to-noise ratio, power spectrum, and image information entropy.

[0083] First, we need to use convolutional neural network (CNN) to extract features and reduce the dimension of the complex and changeable visual modal information input by CCD / CMOS sensors. Suppose the information input amount of CCD sensor array at a certain moment is c is the input feature vector, v is the value of the input feature vector. Through f convolution filters The extracted features are In neural networks, the application of activation functions can introduce nonlinearity, which helps to process complex data patterns while avoiding the problem of gradient vanishing. Then, the maximum pooling operation is used to highlight the significant features in the image and perform dimensionality reduction at the same time. In order to extract richer feature information, convolution-activation-pooling operations are usually performed multiple times. Finally, the obtained feature matrix is ​​converted into a one-dimensional column vector, passed through the neurons of the output layer, and then processed by the sigmoid activation function to output a score between 0 and 1 to describe the quality of the image.

[0084] Use the indicator quantitative evaluation data to train and test the neural network mapping model. Divide the collected data set into training set, validation set and test set, using a ratio of 70%, 15%, and 15%. The evaluation data are stored in sample-1.txt, sample-2.txt and sample-3.txt files respectively. The first three columns of the file are the image imaging system parameters, and the last three columns are the quantitative evaluation values ​​of the single indicator. After the model training is completed, the actual results and prediction results on the test set can be used to verify whether the constructed mapping model prediction is accurate. Use the training set to train the constructed convolutional neural network model, and adjust the network parameters by optimizing the loss function. During the training process, the Adam optimization algorithm is used. The trained convolutional neural network model is deployed to the actual mobile vision system to predict the quantitative evaluation results of the input image imaging system parameters, such as Figure 6 shown.

[0085] like Figure 7 As shown, in step S8, a genetic algorithm is used to optimize the image imaging system parameters, and the specific steps are as follows:

[0086] After completing the construction of the image comprehensive quality mapping model, the genetic algorithm is now combined to optimize the parameters of the image imaging system. The optimization process of the genetic algorithm is implemented through Python3.8. Before use, the main parameters in the genetic algorithm need to be set: population size PoP = 200, crossover probability Pc = 0.2, mutation probability Pv = 0.1, and number of iterations M = 1500. The constraints on the image imaging system parameters are set as follows: exposure time e min =110μs,e max =310μs; light source intensity l min =100,l 1max =1600; imaging distance d min =0,d max =300cm, the optimization of image imaging system parameters can be achieved through genetic algorithm combined with image comprehensive quality mapping model.

[0087] First, the parameters of the image imaging system are converted into variables that need to be optimized. These variables have an important impact on the imaging quality when the robot moves, and are regarded as input features in the mapping model. These input features are variables within the controllable range.

[0088] Next, when performing parameter optimization, it is necessary to encode multiple groups of imaging parameters used by the genetic algorithm that meet the preset image quality evaluation values ​​as variables to be optimized. There are two encoding methods in this algorithm, namely decimal and binary. Compared with the latter, which converts the data before decoding, the former uses the data itself, making the optimization result clearer. After obtaining the initial population, it is necessary to calculate the fitness of the image imaging system parameter combination, output the image imaging parameters, determine whether the algorithm optimization result is feasible, and provide a numerical reference for individual selection in the optimization process. Commonly used fitness functions include penalty functions, objective functions themselves, and objective function conversions. The fitness function Fit(x) used in this embodiment is formulated as g(e,l,d) is the output result of the image quality mapping model.

[0089] Then, the output imaging parameters of the genetic algorithm are input into the mapping model to obtain the output imaging quality corresponding to the output imaging parameters, and to determine whether the output imaging quality image satisfies the quality evaluation result of the optimal quality image. If it does not meet the requirements, the individuals that do not meet the termination conditions need to be selected, gene crossed, and gene mutated. For individual selection, the roulette method is used, and the formula is Where POP is the population size, Fit(x i ) is the individual fitness. Parameters with better fitness within the current iteration number are selected and retained for the next generation. For gene crossover, according to the crossover probability P c Reorganize the parameters at the same position of the two individuals to obtain new imaging system parameters. Choose to use the formula and To simulate gene crossover. For gene mutation, select random mutation mode, according to the mutation probability P v A certain parameter on the individual chromosome is mutated. When the image imaging system parameters obtained by optimization meet the requirements of optimal image quality, the optimization process ends and the optimal image imaging system parameter combination is obtained. The optimal image quality in the genetic algorithm process is set based on prior knowledge.

[0090] Finally, in the parameter optimization problem of the imaging system during the whole process of the vehicle moving, the optimization goal is defined as obtaining high-quality images. The output result of the image quality mapping model is g(e,l,d), and the objective function is constructed based on the above output results. The objective function formula in the genetic algorithm is k(e,l,d)=|g max -g(e,l,d)|. Among them, g max is the maximum expected value of the image quality. According to the image quantization evaluation formula of the whole process of the car moving, during the evaluation process, its score is in the range of [0,1]. max =1 is the maximum expected value.

[0091] Embodiment 2

[0092] On the other hand, an imaging parameter optimization system for a mobile vision system is provided, comprising:

[0093] A data acquisition module, used to obtain imaging parameters that affect image quality, and constrain the imaging parameters to obtain image data;

[0094] A mapping module, used to construct a mapping model, and obtain a plurality of sets of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained by training a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network;

[0095] The optimization module is used to optimize the plurality of imaging parameters with the maximum expected value of the image quality evaluation value as the target, so as to obtain the optimal combination of imaging system parameters.

[0096] The data acquisition module includes: a mobile platform, an image forming device, a navigation and control unit, an image acquisition unit, and an imaging parameter acquisition unit;

[0097] The mobile platform is used to move within the target area;

[0098] The image forming device is used to form an image of a moving target object;

[0099] The navigation and control unit is used to control the motion trajectory of the mobile platform;

[0100] The image acquisition unit is used to acquire image data of the moving target object based on the image imaging device;

[0101] The imaging parameter acquisition unit analyzes the influence of different imaging parameters on image quality through a statistical method, and acquires imaging parameters including exposure time, light source intensity, and the distance between the target object and the camera.

[0102] like Figure 2As shown, in this embodiment, the car is used as a mobile platform for moving within a specified area. In addition, the car can be a remote-controlled car or an autonomous navigation vehicle driven by a motor, depending on the application requirements and budget. The image imaging system can be one or more cameras for imaging the target object. The camera is a CCD / CMOS digital camera. The navigation and control unit of the car is used to control the movement trajectory of the car to ensure that the car can follow the predetermined path or avoid obstacles; an embedded computing unit (such as Raspberry Pi) or a cloud server is used to process data. The workflow of the image imaging system during the whole process of the car moving is as follows: After the car is started, the image imaging system first performs target recognition to determine the target object or area to be imaged. Then, according to the target position and the current position of the car, a path planning scheme is formulated to ensure that the car can effectively move to the specified position. The car starts to move according to the path planning, and the image imaging system continues to collect images of the target object. During the movement, it can be adjusted according to the camera's viewing angle to ensure that the target object is always within the imaging range. Image data is obtained through the above process, and the image data can be used for subsequent target analysis.

[0103] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. An imaging parameter optimization method for a mobile vision system, characterized in that: include: Acquiring imaging parameters that affect image quality, and constraining the imaging parameters to acquire image data; Constructing a mapping model, and obtaining a plurality of sets of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained by training a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network; Taking the maximum expected value of the image quality evaluation value as a target, optimizing the several groups of imaging parameters to obtain the optimal combination of imaging system parameters; Taking the maximum expected value of the image quality quantification result as a target, optimizing the imaging parameters includes: optimizing the several groups of imaging parameters by using a genetic algorithm to obtain an optimal combination of imaging system parameters; The optimization of the plurality of imaging parameters by using a genetic algorithm includes: S1, taking the several groups of imaging parameters as variables to be optimized, performing decimal encoding on the variables to be optimized, and initializing the population; S2, calculating the fitness of the variables to be optimized encoded in the initialized population through a fitness function, and outputting imaging parameters; S3, input the output imaging parameters into the mapping model, obtain the quantization result of the image quality evaluation value corresponding to the output imaging parameters, and determine whether the output quantization result meets the quality evaluation result of the optimal quality image. If so, proceed to S4; if not, perform individual selection, crossover and mutation, obtain the next generation population and return to S2; S4. When the output imaging parameters obtained by optimization satisfy the quality evaluation result of the optimal quality image, the optimal image imaging system parameter combination is obtained.

2. The imaging parameter optimization method for a mobile vision system according to claim 1, characterized in that: The imaging parameters that affect image quality include: The influence of different imaging parameters on image quality is analyzed by statistical methods, and imaging parameters including exposure time, light source intensity, and the distance between the target and the camera are obtained.

3. The imaging parameter optimization method for a mobile vision system according to claim 1, characterized in that: Acquiring the image data includes: Screening the image data to obtain high-quality images; The high-quality image is normalized, denoised, resized, and color-balanced to obtain the image data.

4. The imaging parameter optimization method for a mobile vision system according to claim 3, characterized in that: Obtaining the image quality evaluation result includes: Clarity, power spectrum and signal-to-noise ratio are selected as image quality evaluation indicators, and the image quality evaluation indicators are weighted by entropy weight method to obtain image quality evaluation results corresponding to the image data, wherein the clarity is obtained by Brenner gradient function.

5. The imaging parameter optimization method for a mobile vision system according to claim 1, characterized in that: The objective function of the genetic algorithm is: k(e,l,d)=|g max -g(e,l,d)| Among them, g max is the maximum expected value of image quality, g(e,l,d) is the output result of the mapping model, and k(e,l,d) is the objective function of the genetic algorithm.

6. The imaging parameter optimization method for a mobile vision system according to claim 1, characterized in that: The fitness function is: Among them, g(e,l,d) is the output result of the mapping model.

7. An imaging parameter optimization system for a mobile vision system, characterized in that: include: A data acquisition module, used to obtain imaging parameters that affect image quality, and constrain the imaging parameters to obtain image data; A mapping module, used to construct a mapping model, and obtain a plurality of sets of imaging parameters corresponding to preset image quality evaluation values ​​based on the mapping model, wherein the mapping model is obtained by training a training set, the training set includes imaging parameters and image quality evaluation results corresponding to the image data, and the mapping model is constructed based on a convolutional neural network; An optimization module, used to optimize the plurality of imaging parameters with the maximum expected value of the image quality evaluation value as the target, and obtain the optimal combination of imaging system parameters; Taking the maximum expected value of the image quality quantification result as a target, optimizing the imaging parameters includes: optimizing the several groups of imaging parameters by using a genetic algorithm to obtain an optimal combination of imaging system parameters; The optimization of the plurality of imaging parameters by using a genetic algorithm includes: S1, taking the several groups of imaging parameters as variables to be optimized, performing decimal encoding on the variables to be optimized, and initializing the population; S2, calculating the fitness of the variables to be optimized encoded in the initialized population through a fitness function, and outputting imaging parameters; S3, input the output imaging parameters into the mapping model, obtain the quantization result of the image quality evaluation value corresponding to the output imaging parameters, and determine whether the output quantization result meets the quality evaluation result of the optimal quality image. If so, proceed to S4; if not, perform individual selection, crossover and mutation, obtain the next generation population and return to S2; S4. When the output imaging parameters obtained by optimization satisfy the quality evaluation result of the optimal quality image, the optimal image imaging system parameter combination is obtained.

8. The imaging parameter optimization system for mobile vision system according to claim 7, characterized in that: The data acquisition module includes: a mobile platform, an image forming device, a navigation and control unit, an image acquisition unit, and an imaging parameter acquisition unit; The mobile platform is used to move within the target area; The image forming device is used to form an image of a moving target object; The navigation and control unit is used to control the motion trajectory of the mobile platform; The image acquisition unit is used to acquire image data of the moving target object based on the image imaging device; The imaging parameter acquisition unit analyzes the influence of different imaging parameters on image quality through a statistical method, and acquires imaging parameters including exposure time, light source intensity, and the distance between the target object and the camera.

Images