Photovoltaic cleaning robot sensor data fusion and obstacle avoidance method and system

Abstract

The invention provides a photovoltaic cleaning robot sensor data fusion and obstacle avoidance method and system, and relates to the technical field of photovoltaic cleaning robots, and the method comprises the steps: obtaining multi-source sensor data, carrying out the probability distribution modeling through employing a Gaussian mixture model, and achieving the data fusion through employing a Kalman filtering algorithm of an adaptive covariance matrix, and constructing a three-dimensional semantic map, marking obstacle information, and dynamically adjusting a motion track in combination with global path planning and a fuzzy logic controller of a local obstacle avoidance layer to realize smooth obstacle avoidance of the photovoltaic cleaning robot. The environment perception capability and the obstacle avoidance efficiency of the photovoltaic cleaning robot are improved, the collision risk is reduced, and the safety and the efficiency of photovoltaic panel cleaning operation are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technology of photovoltaic cleaning robots, and particularly to a method and system for sensor data fusion and obstacle avoidance of photovoltaic cleaning robots. Background Art

[0002] With the rapid development of the photovoltaic industry, the scale of photovoltaic power stations is continuously expanding, and the cleaning and maintenance of photovoltaic panels have become a key factor affecting power generation efficiency. The traditional manual cleaning method not only has high costs and low efficiency, but also has potential safety hazards. Therefore, photovoltaic cleaning robots are widely used as an automated cleaning solution. However, the environment of photovoltaic power stations is complex and changeable, and there are various types of obstacles, including photovoltaic brackets, cables, dust accumulation areas, and various temporary obstacles. This poses high requirements for the sensor data fusion and obstacle avoidance of photovoltaic cleaning robots.

[0003] Currently, photovoltaic cleaning robots on the market mainly rely on single sensors for environmental perception, and the data collection is not comprehensive, making it difficult to accurately identify various obstacles in complex environments. In terms of sensor data processing, most use simple data splicing or weighted average methods, lacking the adaptive ability to data noise and environmental changes, resulting in insufficient fusion accuracy and affecting the accuracy of subsequent decisions.

[0004] The existing photovoltaic cleaning robots usually adopt simple stop or detour schemes in obstacle avoidance strategies, lacking in-depth understanding of the environment and intelligent decision-making capabilities. Traditional obstacle avoidance algorithms often only consider a single factor, such as the distance to the obstacle, while ignoring comprehensive factors such as relative speed and cleaning efficiency, resulting in a rigid obstacle avoidance process and even problems such as repeated adjustments and uneven paths, seriously affecting the cleaning efficiency and quality.

[0005] The particularity of the photovoltaic panel surface and the complexity of the installation environment make it difficult for existing obstacle avoidance technologies to balance the relationship between cleaning efficiency and safety. The current technology lacks a method for effectively combining global path planning and local obstacle avoidance, resulting in slow or over-reaction when facing sudden obstacles, which not only affects the cleaning efficiency but also may cause damage to the photovoltaic panels. Summary of the Invention

[0006] Embodiments of the present invention provide a method and system for sensor data fusion and obstacle avoidance of photovoltaic cleaning robots, which can solve the problems in the prior art.

[0007] In the first aspect of the embodiments of the present invention,

[0008] A method for sensor data fusion and obstacle avoidance of a photovoltaic cleaning robot is provided, including:

[0009] Obtain multi-source sensor data of the photovoltaic cleaning robot, where the multi-source sensor data includes surface image data, three-dimensional point cloud data, and attitude data;

[0010] Perform probability distribution modeling on the multi-source sensor data based on the Gaussian mixture model, and use the Kalman filter algorithm with an adaptive covariance matrix for multi-source data fusion, where the adaptive covariance matrix is updated in real time according to the noise characteristics of the multi-source sensor data and environmental changes;

[0011] Construct a three-dimensional semantic map based on the fused multi-source sensor data, and mark the position, type, and size information of obstacles in the three-dimensional semantic map;

[0012] Perform obstacle avoidance path planning on the three-dimensional semantic map. The three-dimensional semantic map includes a global path planning layer and a local obstacle avoidance layer. Among them, the global path planning layer generates an initial path based on the Kalman filter algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance;

[0013] According to the initial path after achieving the smooth obstacle avoidance, control the photovoltaic cleaning robot to move along the set path, and detect obstacles ahead in real time. When an obstacle is detected, combine the current state of the robot and the obstacle feature information to execute the local obstacle avoidance strategy.

[0014] Performing probability distribution modeling on the multi-source sensor data based on the Gaussian mixture model and using the Kalman filter algorithm with an adaptive covariance matrix for multi-source data fusion includes:

[0015] Construct the multi-source sensor data into a data set, and construct a mixture model with multiple Gaussian components for the data set. Each Gaussian component includes a mixing weight, a mean vector, and a covariance tensor;

[0016] Use the expectation maximization algorithm to iteratively optimize the parameters of the mixture model, calculate the expected value of the latent variable by calculating the posterior probability, and update the mixing weight, the mean vector, and the covariance tensor based on the expected value of the latent variable to obtain an innovation sequence;

[0017] Calculate an adaptive factor based on the innovation sequence, multiply the adaptive factor by the adaptive covariance matrix, and obtain a corrected adaptive covariance matrix through the Kalman filter algorithm;

[0018] Use the corrected prediction error covariance to correct the Kalman filter algorithm to obtain a corrected Kalman gain; update the state estimate value and the error covariance based on the corrected Kalman gain to achieve adaptive fusion of the multi-source sensor data.

[0019] Calculating the probability distribution of the features of the photovoltaic panel surface image data and the features of the three-dimensional point cloud data in the multi-source sensor data, and determining the weight coefficient of the feature data according to the result of the probability distribution calculation includes:

[0020] Performing multi-scale decomposition on the photovoltaic panel surface image data in the multi-source sensor data to obtain image features, where the image features include edge features, texture features, and local gradient features; extracting geometric features from the three-dimensional point cloud data in the multi-source sensor data to obtain point cloud features, where the point cloud features include normal vector features, curvature features, and local depth change features;

[0021] Performing differential calculation based on the deviation between the image features and their feature mean to obtain the image probability distribution value, and performing weighted calculation based on the deviation between the point cloud features and their geometric feature mean to obtain the point cloud probability distribution value;

[0022] Determining the weight coefficient according to the covariance between the image features and the point cloud features, and the ratio of the image probability distribution value to the point cloud probability distribution value.

[0023] The global path planning layer generates an initial path based on the Kalman filter algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance, including:

[0024] Obtaining the historical motion trajectory data of the photovoltaic cleaning robot, and generating an initial state prediction result according to the historical motion trajectory data;

[0025] Optimizing the initial state prediction result using the Kalman filter algorithm to obtain an optimized state prediction result;

[0026] Inputting the optimized state prediction result into the path planning model, calculating the cost values of different paths according to the path length, turning angle, obstacle distance, and energy consumption, and the calculation process includes: multiplying each parameter by the corresponding weight coefficient and then summing to obtain the comprehensive cost value, and selecting the path with the smallest comprehensive cost value as the candidate path;

[0027] Smoothing the candidate path, evaluating the feasibility of the candidate path according to the path smoothness, safety distance, and task time, and selecting the optimal path as the global planning path;

[0028] Obtaining the real-time obstacle distance information, relative speed information, and boundary constraint information around the photovoltaic cleaning robot, and combining the global planning path to normalize the obstacle distance, relative speed, and boundary distance and then perform weighted summation to obtain the obstacle avoidance parameter;

[0029] Multiply the obstacle avoidance parameter by a preset speed adjustment coefficient to obtain a speed correction value, and achieve smooth obstacle avoidance based on the obstacle avoidance parameter and the speed correction amount.

[0030] Introduce an adaptive factor to the initial state prediction result, dynamically adjust the state transition matrix and the observation matrix according to the prediction error, and optimize the initial state prediction result using the Kalman filter algorithm. The optimized state prediction result includes:

[0031] Obtain the system state vector, which includes position coordinates, orientation angle, linear velocity, and angular velocity. Calculate the expected value of the error product of the system state vector and the state estimation vector using the expectation operation to obtain the prediction error covariance matrix;

[0032] Calculate the ratio of the prediction error covariance matrix to the theoretical measurement residual covariance matrix to obtain the adaptive factor;

[0033] Construct a diagonal adaptive weight matrix from the adaptive factor according to the dimensions of the system state vector. The diagonal elements of the diagonal adaptive weight matrix are the adaptive factors corresponding to different state components;

[0034] Perform matrix multiplication on the diagonal adaptive weight matrix and the state transition matrix to obtain an adjusted state transition matrix. Perform matrix multiplication on the diagonal adaptive weight matrix and the difference between the current moment state vector and the state prediction value to obtain a prediction error compensation term;

[0035] Left-multiply the state vector of the previous moment by the adjusted state transition matrix and add it to the prediction error compensation term to obtain an optimized state prediction result.

[0036] According to the initial path after achieving the smooth obstacle avoidance, control the photovoltaic cleaning robot to move along the set path, and detect obstacles ahead in real time. When an obstacle is detected, combine the current state of the robot and the obstacle feature information to execute a local obstacle avoidance strategy, including:

[0037] Take the initial path as the set path and control the photovoltaic cleaning robot to move along the set path; during the movement of the photovoltaic cleaning robot, obtain the current position coordinates, current orientation angle, and current speed in real time, and at the same time detect the position coordinates and envelope radius of the obstacle ahead in real time;

[0038] Calculate the position error according to the current position coordinates and the position coordinates of the corresponding path point on the set path, calculate the angle error according to the current orientation angle and the orientation angle of the corresponding path point, and construct a pose error vector;

[0039] Based on the pose error vector, a linear velocity control quantity is obtained by multiplying the position error by a first control gain parameter, and an angular velocity control quantity is obtained by multiplying the position error by a second control gain parameter;

[0040] Subtract the position coordinates of the obstacle from the current position coordinates to obtain a relative position vector, and calculate the modulus of the relative position vector to obtain a relative distance;

[0041] Determine whether the relative distance satisfies a safety distance constraint, where the safety distance constraint is that the relative distance is greater than the sum of the minimum safety distance, the envelope radius, and the robot radius;

[0042] When the safety distance constraint is satisfied, keep the linear velocity control quantity and the angular velocity control quantity unchanged; when the safety distance constraint is not satisfied, execute a local obstacle avoidance strategy.

[0043] Calculate a position error according to the position coordinates of the current position and the corresponding path points of the set path, calculate an angle error according to the current orientation angle and the orientation angle of the corresponding path points, and construct a pose error vector including:

[0044] Obtain the current position coordinates and the current orientation angle of the robot, and find the path point closest to the current position coordinates on the preset path as the corresponding path point;

[0045] Calculate the position error between the current position coordinates and the position coordinates of the corresponding path point in the local coordinate system of the robot respectively, and calculate the included angle between the current orientation angle and the orientation angle of the corresponding path point to obtain an angle error; form a pose error vector with the position error and the angle error.

[0046] In the second aspect of the embodiments of the present invention,

[0047] A photovoltaic cleaning robot sensor data fusion and obstacle avoidance system is provided, including:

[0048] A first unit for obtaining multi-source sensor data of a photovoltaic cleaning robot, where the multi-source sensor data includes surface image data, three-dimensional point cloud data, and attitude data;

[0049] A second unit for performing probability distribution modeling on the multi-source sensor data based on a Gaussian mixture model, and performing multi-source data fusion using a Kalman filtering algorithm with an adaptive covariance matrix, where the adaptive covariance matrix is updated in real time according to the noise characteristics of the multi-source sensor data and environmental changes;

[0050] A third unit for constructing a three-dimensional semantic map according to the fused multi-source sensor data, and marking the position, type, and size information of obstacles in the three-dimensional semantic map;

[0051] The fourth unit is used to plan an obstacle avoidance path for the three-dimensional semantic map. The three-dimensional semantic map includes a global path planning layer and a local obstacle avoidance layer. Among them, the global path planning layer generates an initial path based on the Kalman filtering algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance;

[0052] The fifth unit is used to control the photovoltaic cleaning robot to move along the set path according to the initial path after achieving the smooth obstacle avoidance, and to detect obstacles ahead in real time. When an obstacle is detected, a local obstacle avoidance strategy is executed in combination with the current state of the robot and the obstacle feature information.

[0053] In the third aspect of the embodiments of the present invention,

[0054] There is provided an electronic device, including:

[0055] A processor;

[0056] A memory for storing instructions executable by the processor;

[0057] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0058] In the fourth aspect of the embodiments of the present invention,

[0059] There is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0060] The beneficial effects of this application are as follows:

[0061] By using the Kalman filtering algorithm with an adaptive covariance matrix to fuse multi-source sensor data, high-precision perception of the environment around the photovoltaic cleaning robot is achieved, effectively reducing the limitations and uncertainties of a single sensor, improving the accuracy and robustness of environmental perception, and enabling the robot to obtain more reliable environmental information in the complex and changing photovoltaic panel environment.

[0062] Based on the fused multi-source sensor data, a three-dimensional semantic map is constructed, which can not only accurately mark the position, type, and size information of obstacles, but also provide a comprehensive understanding of the environment for subsequent path planning, enhancing the robot's cognitive ability of the working environment, providing a solid foundation for intelligent decision-making, and effectively improving the accuracy of obstacle recognition.

[0063] Adopt a hierarchical obstacle avoidance path planning strategy, combine the collaborative work of global path planning and local obstacle avoidance layers, and use a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints, achieving smooth obstacle avoidance. This not only ensures the continuity and efficiency of the cleaning operation but also effectively avoids collisions with obstacles, improving the operation safety and adaptability of the photovoltaic cleaning robot in complex environments. Brief Description of the Drawings

[0064] Figure 1 It is a flowchart of the sensor data fusion and obstacle avoidance method for the photovoltaic cleaning robot according to the embodiment of the present invention;

[0065] Figure 2 It is a schematic diagram of the accuracy of sensor data processing under different abnormal rates according to the embodiment of the present invention;

[0066] Figure 3 It is a schematic diagram for comparing the detection accuracy under different pollution coverage rates according to the embodiment of the present invention;

[0067] Figure 4 It is a flowchart of the adaptive Kalman filter algorithm according to the embodiment of the present invention;

[0068] Figure 5 It is a schematic diagram of the influence of the obstacle avoidance gain parameter K_obstacle on the system performance according to the embodiment of the present invention. Detailed Embodiments

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0070] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0071] Figure 1 It is a flowchart of the sensor data fusion and obstacle avoidance method for the photovoltaic cleaning robot according to the embodiment of the present invention. As Figure 1 shown, the method includes:

[0072] Obtain multi-source sensor data of the photovoltaic cleaning robot, where the multi-source sensor data includes surface image data, three-dimensional point cloud data, and attitude data;

[0073] Perform probability distribution modeling on the multi-source sensor data based on the Gaussian mixture model, and use the Kalman filter algorithm with an adaptive covariance matrix for multi-source data fusion, where the adaptive covariance matrix is updated in real time according to the noise characteristics of the multi-source sensor data and environmental changes;

[0074] Construct a three-dimensional semantic map based on the fused multi-source sensor data, and mark the position, type, and size information of obstacles in the three-dimensional semantic map;

[0075] Perform obstacle avoidance path planning on the three-dimensional semantic map, where the three-dimensional semantic map includes a global path planning layer and a local obstacle avoidance layer. Among them, the global path planning layer generates an initial path based on the Kalman filter algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance;

[0076] According to the initial path after achieving the smooth obstacle avoidance, control the photovoltaic cleaning robot to move along the set path, and detect obstacles ahead in real time. When an obstacle is detected, combine the current state of the robot and the obstacle feature information to execute the local obstacle avoidance strategy.

[0077] In an alternative embodiment, performing probability distribution modeling on the multi-source sensor data based on the Gaussian mixture model and using the Kalman filter algorithm with an adaptive covariance matrix for multi-source data fusion includes:

[0078] Construct the multi-source sensor data into a data set, and construct a mixture model with multiple Gaussian components for the data set. Each Gaussian component includes a mixing weight, a mean vector, and a covariance tensor;

[0079] Use the expectation maximization algorithm to iteratively optimize the parameters of the mixture model, calculate the expected value of the hidden variable by calculating the posterior probability, and update the mixing weight, the mean vector, and the covariance tensor based on the expected value of the hidden variable to obtain an innovation sequence;

[0080] Calculate an adaptive factor based on the innovation sequence, multiply the adaptive factor by the adaptive covariance matrix, and obtain a corrected adaptive covariance matrix through the Kalman filter algorithm;

[0081] Use the corrected prediction error covariance to correct the Kalman filter algorithm to obtain a corrected Kalman gain; update the state estimate value and the error covariance based on the corrected Kalman gain to achieve adaptive fusion of the multi-source sensor data.

[0082] In this embodiment, for the problem of multi-source sensor data fusion, a Kalman filter fusion method based on Gaussian mixture model and adaptive covariance matrix is proposed. This method models the probability distribution of multi-source sensor data through a Gaussian mixture model and uses the Kalman filter algorithm with an adaptive covariance matrix to achieve precise fusion of multi-source data.

[0083] Preprocess and construct multi-source sensor data. Assume that the data from different sensors includes information such as position, velocity, and acceleration. Synchronize these data according to timestamps and construct them into a unified data set. This data set can be expressed as D = {x1, x2,..., x n}, where each data point x□ is a multi-dimensional vector containing various measurement values collected by the sensor.

[0084] For the above data set, construct a mixture model with K Gaussian components. Each Gaussian component includes three key parameters: mixture weight π_k, mean vector μ_k, and covariance tensor Σ_k. The mixture weight represents the importance of this Gaussian component in the overall model, and the sum of all weights is 1. The mean vector represents the center of the data distribution, and the covariance tensor describes the degree of dispersion of the data and the correlation between dimensions.

[0085] In practical applications, the appropriate number of Gaussian components can be selected according to the data complexity. For example, for the multi-sensor fusion system of an unmanned aerial vehicle, 3 - 5 Gaussian components can be selected to model the data of sensors such as GPS, IMU, and vision. Initially, the K-means clustering method can be used to group the data, and the initial mean vector and covariance tensor can be calculated based on each group. The mixture weight can be set according to the proportion of the number of data points in each group.

[0086] To optimize the parameters of the mixture model, the Expectation-Maximization (EM) algorithm is used for iteration. This algorithm includes an Expectation step (E-step) and a Maximization step (M-step). In the E-step, calculate the posterior probability that each data point belongs to each Gaussian component, that is, the expected value of the latent variable z. For the data point x□, its posterior probability of belonging to the k-th Gaussian component can be calculated through the current parameters.

[0087] In the M-step, update the parameters of the mixture model based on the calculated posterior probability above. The updated mixture weight is the average posterior probability that each data point belongs to this component; the updated mean vector is the weighted average of all data points according to the posterior probability; the updated covariance tensor takes into account the deviation between the data point and the mean vector. By iterating the E-step and M-step multiple times, the parameters of the mixture model gradually converge to the optimal values.

[0088] After completing the Gaussian mixture model modeling, an innovation sequence of the data is obtained. The innovation sequence represents the difference between the actual observed value and the model predicted value, and is a key indicator in Kalman filtering. According to the statistical characteristics of the innovation sequence, the adaptive factor can be calculated. This adaptive factor reflects the matching degree between the current model and the actual observation. When the deviation is large, the value of the adaptive factor increases, and vice versa.

[0089] For example, assume that in a certain UAV system, GPS and IMU data are fused. When the innovation sequence shows large fluctuations, it indicates that there may be GPS signal interference in the current environment. At this time, the adaptive factor will increase accordingly, reducing the dependence on GPS information and enhancing the weight of IMU data in the fusion result.

[0090] Based on the traditional Kalman filtering algorithm, an adaptive mechanism is introduced. Multiply the calculated adaptive factor by the covariance matrix to obtain the corrected adaptive covariance matrix. This correction process takes into account the real-time characteristics of the data, enabling the filtering algorithm to dynamically adjust the trust in different sensor data.

[0091] For example, when there are abnormal fluctuations in the data of a certain sensor, the corresponding covariance component will be appropriately amplified, reducing the influence of this sensor in the fusion result. In specific implementation, reasonable upper and lower limit thresholds can be set to ensure that the adaptive factor fluctuates within an effective range, avoiding over-adjustment that may lead to system instability.

[0092] Use the corrected prediction error covariance to correct the Kalman filtering algorithm. The gain matrix in traditional Kalman filtering determines the degree of adoption of the observed innovation. By introducing the adaptive covariance matrix, the corrected Kalman gain is obtained. This gain can more reasonably balance the prior information and the new observed data, improving the accuracy of state estimation.

[0093] Update the state estimate value and the error covariance based on the corrected Kalman gain. The update of the state estimate value comprehensively considers the prior predicted value and the observed innovation, and the update of the error covariance reflects the uncertainty of the estimation result. Through this process, the adaptive fusion of multi-source sensor data is achieved.

[0094] In an actual application case, this method has been successfully applied to the multi-sensor fusion of an intelligent driving system. In a certain test, the system fused the data of lidar, millimeter-wave radar, and camera. When the vehicle was driving in a tunnel, due to the change of light conditions, the quality of the camera data decreased. The system automatically increased the adaptive factor corresponding to the camera data and increased the dependence on the lidar and millimeter-wave radar data, ensuring the accuracy of position estimation and controlling the positioning error within 10 centimeters.

[0095] The core advantage of this method lies in its ability to adaptively adjust the fusion strategy according to data characteristics and has strong robustness to abnormal data. At the same time, the introduction of the Gaussian mixture model enables this method to process sensor data with non-Gaussian distributions, expanding the application scenarios. Through the Kalman filtering algorithm with an adaptive covariance matrix, the precise fusion of multi-source sensor data is achieved, improving the system performance.

[0096] Figure 2 Schematic diagram of the processing accuracy of sensor data under different abnormal rates in the embodiments of the present invention:

[0097] This figure shows the comparison of the processing accuracy of three different filtering schemes under different proportions of abnormal sensor data. The horizontal axis represents the proportion of abnormal sensor data, ranging from 5% to 50%; the vertical axis represents the processing accuracy, expressed as a percentage. The figure contains three curves, representing the performance of the technical solution of the present invention (triangle markers), the standard Kalman filter (circle markers), and the particle filter (square markers), respectively. When the proportion of abnormal sensor data is 5%, the processing accuracy of all three schemes is relatively high: about 98% for the technical solution of the present invention, about 95% for the standard Kalman filter, and about 93% for the particle filter. As the proportion of abnormal data increases, the performance of all three schemes shows a downward trend, but at different rates. When the proportion of abnormal data reaches 50%, the technical solution of the present invention can still maintain an accuracy of about 87%, while the standard Kalman filter drops to about 59%, and the particle filter drops to about 55%. From the overall trend, the technical solution of the present invention shows the strongest robustness, with the slowest rate of decrease in accuracy as the proportion of abnormal data increases and the flattest curve. The performance of the standard Kalman filter and the particle filter decreases more significantly, especially after the proportion of abnormal data exceeds 25%, and the accuracy decreases more prominently. This shows that the technical solution of the present invention has obvious advantages in processing high proportions of abnormal data and is more suitable for applications in scenarios where a large amount of abnormal data may occur in sensors.

[0098] In an alternative embodiment, calculating the probability distribution of the features of the photovoltaic panel surface image data and the features of the three-dimensional point cloud data in the multi-source sensor data, and determining the weight coefficient of the feature data according to the result of the probability distribution calculation includes:

[0099] Performing multi-scale decomposition on the photovoltaic panel surface image data in the multi-source sensor data to obtain image features, where the image features include edge features, texture features, and local gradient features; extracting geometric features from the three-dimensional point cloud data in the multi-source sensor data to obtain point cloud features, where the point cloud features include normal vector features, curvature features, and local depth change features;

[0100] The deviation between the image feature and its feature mean is used for differential calculation to obtain the image probability distribution value, and the deviation between the point cloud feature and its geometric feature mean is used for weighted calculation to obtain the point cloud probability distribution value;

[0101] The weight coefficient is determined according to the covariance between the image feature and the point cloud feature and the ratio of the image probability distribution value to the point cloud probability distribution value.

[0102] Edge feature extraction: The Sobel operator is used to process the image to extract the edge information in the horizontal and vertical directions. The gradient magnitude of each pixel point is calculated, and when the gradient magnitude is greater than the preset threshold of 25, it is marked as an edge point. For a 512×512 pixel photovoltaic panel image, usually about 2000 - 3000 effective edge feature points can be extracted. Texture feature extraction: The Local Binary Pattern (LBP) algorithm is applied, with an 8×8 pixel block as the basic unit, and the gray-scale change pattern of the pixels within each block is statistically analyzed. For each block, the gray-scale differences between the central pixel and the surrounding 8 pixels are calculated to form an 8-bit binary code, which is converted into a decimal value as the texture feature value of the block. In standard test samples, the texture feature value usually ranges from 0 to 255. Local gradient feature extraction: Within a 16×16 pixel sliding window, the histogram of gradient directions is calculated. The 360-degree gradient direction is divided into 8 equal intervals, and the frequency of gradient occurrences within each interval is statistically analyzed to form an 8-dimensional feature vector. For the minor defects on the surface of the photovoltaic panel, the local gradient feature usually shows obvious frequency anomalies in the 45 - 135 degree interval.

[0103] The features extracted at three scales are normalized and combined to form a complete feature set of the photovoltaic panel surface image.

[0104] Normal vector feature extraction: For each point, 30 nearest points within its neighborhood are selected, and the principal component analysis method is used to fit the local plane to calculate the normal vector. The normal vector is represented by a three-dimensional unit vector pointing outside the point cloud. For flat areas, the normal vector directions are highly consistent; for defect areas, the normal vector directions change significantly. Curvature feature extraction: Based on the above neighborhood points, the local curvature of each point is calculated. Specifically, it is obtained by calculating the ratio of the average distance from the neighborhood points to the fitted plane to the neighborhood radius. In practical applications, the curvature value of the normal photovoltaic panel surface is usually below 0.001, while the curvature value of the defect area can reach 0.01 - 0.05. Local depth change feature: The standard deviation of the target point and its neighborhood points in the z-axis (depth) direction is calculated. For the minor damage of the photovoltaic panel, the local depth change is usually in the range of 0.5 - 2 mm; for severe damage, the change can reach more than 5 mm.

[0105] After the feature extraction is completed, all feature values are standardized so that they are distributed within the interval [0, 1] for subsequent fusion processing.

[0106] Calculate the mean values of various image features. For example, for edge features, calculate the average of the gradient magnitudes of all extracted edge points; for texture features, calculate the average of the texture feature values of all blocks; for local gradient features, calculate the average of the gradient frequencies in 8 direction intervals. Calculate the deviation of each feature from its mean value. Taking the edge feature as an example, if the gradient magnitude at a certain point is 75 and the mean value is 50, the deviation is 25. Normalize the deviation and convert it to a probability value using the Gaussian function. The larger the deviation, the lower the probability value, indicating a higher likelihood of abnormality for the feature point. Combine the probability values of the three types of features to obtain the final image probability distribution value. In practical applications, the combination weights of edge features, texture features, and local gradient features are set to 0.4, 0.3, and 0.3 respectively.

[0107] Calculate the mean values of various point cloud geometric features. For example, the mean value of the normal vector feature is obtained by calculating the average direction of all normal vectors; the mean value of the curvature feature is the average of all point curvature values; the mean value of the depth change feature is the average of the local depth standard deviations of all points. Calculate the weighted deviation of each feature from its mean value. For the normal vector feature, calculate the angle between the current normal vector and the mean normal vector; for the curvature and depth change features, directly calculate the difference from the mean value. According to the importance of the three types of geometric features, assign weight coefficients of 0.35, 0.35, and 0.3, and calculate the sum of the weighted deviations. Apply the exponential decay function to convert the weighted deviation to a probability value to obtain the point cloud probability distribution value.

[0108] Calculate the covariance between image features and point cloud features. The covariance calculation is based on the pairing of corresponding feature points, and the corresponding relationship is determined by the spatial positions of the feature points. For an image with a resolution of 512×512 and a point cloud data containing approximately 100,000 points, usually about 5000 - 8000 valid feature points can be paired. Calculate the ratio of the image probability distribution value to the point cloud probability distribution value. In an ideal state, if the description capabilities of the two types of features for the target area are the same, the ratio should be close to 1; if the image features are more accurate, the ratio is greater than 1; if the point cloud features are more accurate, the ratio is less than 1. Determine the final weight coefficient based on the covariance and the probability distribution ratio. The larger the covariance, the higher the consistency of the two types of features; the degree to which the probability distribution ratio deviates from 1 reflects the difference in the accuracy of the two types of features.

[0109] In practical applications, for the detection of surface defects on photovoltaic panels, when detecting edge breakage, the weight coefficient of image features is usually set to 0.65 - 0.75, and the weight coefficient of point cloud features is 0.25 - 0.35; when detecting small surface depressions, the weight coefficient of image features is usually set to 0.4 - 0.5, and the weight coefficient of point cloud features is 0.5 - 0.6.

[0110] The weight coefficients determined by the above method can achieve the effective fusion of the surface image data and three-dimensional point cloud data of the photovoltaic panel, improving the accuracy and reliability of defect detection. For a standard photovoltaic panel with a size of 1.5 meters × 1 meter, surface defects as small as 2 millimeters in diameter can be detected by this method, and the detection accuracy rate can reach over 95%.

[0111] Figure 3 The following is a schematic diagram showing the comparison of detection accuracy rates under different pollution coverage rates in the embodiments of the present invention:

[0112] This figure shows the comparison of detection accuracy rates of three different technical solutions under different pollution coverage rate conditions. The horizontal axis represents the pollution coverage rate, ranging from 5% to 50%; the vertical axis represents the detection accuracy rate, expressed as a percentage. The figure contains three curves, respectively representing the performance of this technical solution (triangle marker), regional image features (round marker), and point cloud features only (square marker). At the initial pollution coverage rate of 5%, the detection accuracy rates of the three solutions are: this technical solution is approximately 97.5%, regional image features are approximately 96%, and point cloud features only are approximately 91%. As the pollution coverage rate increases, the performance curves of the three solutions all show a trend of first rising and then falling, but the peak positions and change amplitudes are different. This technical solution reaches the highest accuracy rate of approximately 99% at a pollution coverage rate of 25%, regional image features reach the peak of approximately 97% at 15%, and point cloud features only reach the peak of approximately 95% at 25%. When the pollution coverage rate continues to increase to 50%, the performance of the three solutions all decreases to varying degrees: this technical solution drops to approximately 96.5%, still remaining at a relatively high level; regional image features drop to approximately 90.2%; and point cloud features only drop to approximately 91%. Overall, this technical solution shows the best performance at each pollution coverage rate, and the performance curve is more stable, indicating that this solution has better robustness and adaptability, especially it can still maintain a relatively high detection accuracy rate under high pollution coverage rate conditions.

[0113] In an optional implementation manner, the global path planning layer generates an initial path based on the Kalman filtering algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance, including:

[0114] Obtain the historical motion trajectory data of the photovoltaic cleaning robot, and generate an initial state prediction result according to the historical motion trajectory data;

[0115] Optimize the initial state prediction result by using the Kalman filtering algorithm to obtain an optimized state prediction result;

[0116] Input the optimized state prediction result into the path planning model, calculate the cost values of different paths according to the path length, steering angle, obstacle distance, and energy consumption. The calculation process includes: multiplying each parameter by the corresponding weight coefficient and then summing to obtain the comprehensive cost value, and selecting the path with the minimum comprehensive cost value as the candidate path;

[0117] Smooth the candidate path, evaluate the feasibility of the candidate path according to the path smoothness, safety distance, and task time, and select the optimal path as the global planning path;

[0118] Obtain the real-time obstacle distance information, relative speed information, and boundary constraint information around the photovoltaic cleaning robot. Combine the global planning path to normalize the obstacle distance, relative speed, and boundary distance and then perform weighted summation to obtain the obstacle avoidance parameter;

[0119] Multiply the obstacle avoidance parameter by the preset speed adjustment coefficient to obtain the speed correction value, and achieve smooth obstacle avoidance based on the obstacle avoidance parameter and the speed correction amount.

[0120] Obtain the historical motion trajectory data of the photovoltaic cleaning robot. These data include the state information such as the position, speed, acceleration, and angular velocity of the robot in the past period of time. For example, the system can collect the motion state of the robot in the recent 10 seconds, record the data every 100 milliseconds, and form a historical trajectory containing 100 sets of data points. These historical data are used to predict the initial state of the robot, including the current position coordinates, motion direction, and speed.

[0121] Optimize the initial state prediction result by using the Kalman filter algorithm. This algorithm is based on the motion model and observation model of the robot, and continuously optimizes the state estimation through the iterative process of prediction-update. Specifically, the system first calculates the state prediction value at the current moment according to the historical data, and then corrects the prediction result by combining the actual measurement values of the sensors. For example, when the historical trajectory shows that the robot is moving northeast at a speed of 0.5 m / s, the Kalman filter will predict the next position according to this trend and correct the predicted position through the measured data of the GPS or vision positioning system, and finally obtain a more accurate state estimation.

[0122] After inputting the optimized state prediction results into the path planning model, the system starts to calculate the cost values of different paths. The path cost calculation considers four key factors: path length, turning angle, obstacle distance, and energy consumption. The system assigns weight coefficients to each factor. For example, the weight of the path length is 0.3, the weight of the turning angle is 0.2, the weight of the obstacle distance is 0.4, and the weight of the energy consumption is 0.1. For each possible path, the system standardizes the parameters and multiplies them by the corresponding weights and then sums them up to obtain the comprehensive cost value. For example, for a path that is 10 meters long (standardized value 0.5), with an average turning angle of 15 degrees (standardized value 0.3), a minimum obstacle distance of 1.5 meters (standardized value 0.7), and an estimated energy consumption of 2 kilojoules (standardized value 0.4), its comprehensive cost is 0.5×0.3 + 0.3×0.2 + 0.7×0.4 + 0.4×0.1 = 0.51. The system compares the cost values of all paths and selects the path with the minimum cost as the candidate path.

[0123] Smoothing the candidate path is a crucial step to ensure the stable movement of the robot. The system uses the Bezier curve interpolation method to smooth the path inflection points and replaces the sharp turns in the original path with smooth curves. The smoothed path is evaluated according to three criteria: path smoothness (measured by the curvature change rate, with a threshold set at 0.2 / meter), safety distance (maintaining at least a 0.8-meter distance from obstacles), and task time (not exceeding 1.2 times the preset cleaning time). If the candidate path does not meet these conditions, the system will select the path with the second smallest cost for evaluation until an optimal path that meets the conditions is found as the global planning path.

[0124] During the actual operation process, the system obtains the obstacle information around the photovoltaic cleaning robot in real time. This information comes from lidar, ultrasonic sensors, or vision systems, including obstacle distance (measurement range 0 - 5 meters, accuracy ±2 centimeters), relative speed (measurement range ±2 m / s, accuracy ±0.1 m / s), and photovoltaic panel boundary constraints (not less than 0.2 meters from the edge). The system combines this information with the global planning path for processing. First, it normalizes the obstacle distance, relative speed, and boundary distance so that their value ranges are unified to 0 - 1. Then, the system assigns weights to these three parameters (obstacle distance 0.5, relative speed 0.3, boundary constraint 0.2), and obtains the obstacle avoidance parameter through weighted summation.

[0125] For example, when an obstacle is detected 2 meters ahead, with a relative speed of 0.5 m / s (the obstacle is approaching), and at a distance of 0.5 meters from the edge of the photovoltaic panel, the normalized parameters are: obstacle distance 0.4 (2 / 5 = 0.4), relative speed 0.75 (higher value due to the approaching direction), and boundary constraint 0.6 (0.5 / 0.83 = 0.6, assuming the maximum boundary distance is 0.83 meters). The weighted obstacle avoidance parameter is 0.4×0.5 + 0.75×0.3 + 0.6×0.2 = 0.53.

[0126] Multiply the obstacle avoidance parameter by a preset speed adjustment coefficient (usually set to 1.5) to obtain a speed correction value of 0.795. Based on the magnitude of the obstacle avoidance parameter and the speed correction value, the system dynamically adjusts the movement trajectory of the robot: when the obstacle avoidance parameter is less than 0.3, the robot maintains its original speed and direction; when the obstacle avoidance parameter is between 0.3 and 0.7, the robot decelerates and slightly adjusts its direction to increase the distance from the obstacle; when the obstacle avoidance parameter is greater than 0.7, the robot significantly decelerates (down to at most 30% of the normal speed) and makes a large-angle turn (up to 45 degrees) to avoid the obstacle.

[0127] In this way, the photovoltaic cleaning robot can, on the premise of maintaining the general direction of the global path planning, dynamically adjust its movement trajectory according to real-time environmental information, achieve smooth obstacle avoidance, effectively avoid collisions with obstacles, and at the same time maintain the continuity and stability of the movement trajectory, improving the efficiency and safety of the cleaning operation.

[0128] In an alternative embodiment, an adaptive factor is introduced to the initial state prediction result, the state transition matrix and the observation matrix are dynamically adjusted according to the prediction error, and the Kalman filtering algorithm is used to optimize the initial state prediction result. The optimized state prediction result includes:

[0129] Obtain the system state vector, which includes position coordinates, orientation angle, linear velocity, and angular velocity. Calculate the expected value of the product of the error between the system state vector and the state estimation vector using the expectation operation to obtain the prediction error covariance matrix;

[0130] Calculate the ratio of the prediction error covariance matrix to the theoretical measurement residual covariance matrix to obtain the adaptive factor;

[0131] Construct a diagonal adaptive weight matrix with the adaptive factor according to the dimensions of the system state vector. The diagonal elements of the diagonal adaptive weight matrix are the adaptive factors corresponding to different state components;

[0132] Perform matrix multiplication on the diagonal adaptive weight matrix and the state transition matrix to obtain an adjusted state transition matrix. Perform matrix multiplication on the diagonal adaptive weight matrix and the difference between the state vector at the current moment and the state prediction value to obtain a prediction error compensation term.

[0133] Left-multiply the state vector at the previous moment by the adjusted state transition matrix and add the prediction error compensation term to obtain an optimized state prediction result.

[0134] Obtain an initial state prediction result. This result is obtained based on a traditional prediction model and contains information such as position coordinates, orientation angle, linear velocity, and angular velocity. For example, for a mobile robot, its initial state may be: position coordinates (10.2 meters, 15.7 meters), orientation angle 45 degrees, linear velocity 1.5 meters per second, and angular velocity 0.2 radians per second. These initial prediction values may have certain errors and need to be further optimized.

[0135] Obtain the system state vector, which includes position coordinates, orientation angle, linear velocity, and angular velocity. Taking a mobile robot as an example, the system state vector can be expressed as [x, y, θ, v, ω], where x and y are position coordinates, θ is the orientation angle, v is the linear velocity, and ω is the angular velocity. Assume that the system state vector at the current moment is [10.5 meters, 15.9 meters, 46 degrees, 1.55 meters per second, 0.22 radians per second].

[0136] Calculate the expected value of the error product of the system state vector and the state estimation vector through an expectation operation to obtain the prediction error covariance matrix. Specifically, the system first calculates the difference between the system state vector and the state estimation vector, then calculates the product of these differences, and finally takes the expected value to obtain the prediction error covariance matrix. Taking the above mobile robot as an example, if the state estimation vector is [10.2 meters, 15.7 meters, 45 degrees, 1.5 meters per second, 0.2 radians per second] and the difference is [0.3 meters, 0.2 meters, 1 degree, 0.05 meters per second, 0.02 radians per second], the prediction error covariance matrix can be calculated.

[0137] Obtain the theoretical measurement residual covariance matrix, which reflects the noise level in the measurement process. For a mobile robot, assume that the position measurement accuracy is ±0.1 meter, the angle measurement accuracy is ±0.5 degrees, the speed measurement accuracy is ±0.05 meters per second, and the angular velocity measurement accuracy is ±0.01 radians per second. Then, the theoretical measurement residual covariance matrix can be constructed based on this.

[0138] Calculate the ratio of the predicted error covariance matrix to the theoretical measurement residual covariance matrix to obtain the adaptive factor. The ratio is calculated element by element, that is, each element in the predicted error covariance matrix is divided by the corresponding element in the theoretical measurement residual covariance matrix. For example, the adaptive factor in the x - direction of the position may be 3.0, 2.0 in the y - direction of the position, 2.0 for the orientation angle, 1.0 for the linear velocity, and 2.0 for the angular velocity.

[0139] Construct a diagonal adaptive weight matrix from the adaptive factors according to the dimensions of the system state vector. The diagonal elements of this matrix are the adaptive factors corresponding to different state components. For the above - mentioned mobile robot, the diagonal elements of the diagonal adaptive weight matrix are [3.0, 2.0, 2.0, 1.0, 2.0], representing the adaptive weights of the position x, position y, orientation angle, linear velocity, and angular velocity.

[0140] Multiply the diagonal adaptive weight matrix by the state - transition matrix through matrix multiplication to obtain the adjusted state - transition matrix. The state - transition matrix is a matrix that describes the transition of the system from the previous state to the current state. For example, for a uniform - motion model, the state - transition matrix may contain a time - interval term. Assuming the time interval is 0.1 seconds, the adjusted state - transition matrix will reflect the influence of the adaptive weights of each state quantity on the transition process.

[0141] Calculate the difference between the state vector at the current moment and the state prediction value, such as [0.3 m, 0.2 m, 1°, 0.05 m / s, 0.02 rad / s]. Then perform matrix - multiplication operation between the diagonal adaptive weight matrix and this difference to obtain the prediction - error compensation term, such as [0.9 m, 0.4 m, 2°, 0.05 m / s, 0.04 rad / s].

[0142] Left - multiply the state vector at the previous moment by the adjusted state - transition matrix and add the prediction - error compensation term to obtain the optimized state - prediction result. For example, if the state vector at the previous moment is [10.0 m, 15.5 m, 44°, 1.5 m / s, 0.2 rad / s], after calculation by the adjusted state - transition matrix, the basic prediction value may be [10.15 m, 15.65 m, 44.5°, 1.5 m / s, 0.2 rad / s]. After adding the prediction - error compensation term, the final optimized state - prediction result is [11.05 m, 16.05 m, 46.5°, 1.55 m / s, 0.24 rad / s].

[0143] This adaptive Kalman - filtering method effectively improves the accuracy of state prediction by dynamically adjusting the state - transition matrix and introducing the prediction - error compensation term, and is especially suitable for scenarios with high uncertainty in state changes. In practical applications, this method can be adjusted according to the characteristics of the specific system to obtain the best prediction effect.

[0144] This method has been successfully applied to the navigation systems of various mobile robots. For example, in a certain automated guided vehicle project, after adopting this method, the position prediction error has been reduced from an average of 0.5 meters to 0.15 meters, and the orientation angle prediction error has been reduced from an average of 2 degrees to 0.6 degrees, greatly improving the accuracy and reliability of the handling task.

[0145] Figure 4 The flowchart of the adaptive Kalman filter algorithm according to the embodiment of the present invention is as follows:

[0146] This figure shows a flowchart of system state prediction and optimization, mainly describing the implementation steps of the adaptive Kalman filter algorithm. The system obtains the state vector, including key parameters such as position coordinates, orientation angle, linear velocity, and angular velocity. By performing an expectation operation, the expected value of the product of the errors between these system state vectors and the state estimation vectors is calculated, thereby obtaining the predicted error covariance matrix. Then, the system calculates the ratio of the predicted error covariance matrix to the theoretical measurement residual covariance matrix to obtain the adaptive factor. A diagonal adaptive weight matrix is constructed based on the dimension of the system state vector, where the diagonal elements are the adaptive factors corresponding to different state components. On this basis, the diagonal adaptive weight matrix is multiplied by the state transition matrix to calculate the adjusted state transition matrix. At the same time, the diagonal adaptive weight matrix is multiplied by the difference between the current state vector and the state prediction value to obtain the predicted error compensation term. The state vector at the previous moment is left-multiplied by the adjusted state transition matrix and added to the predicted error compensation term to finally obtain the optimized state prediction result. This complete process reflects the characteristic of the adaptive Kalman filter algorithm to improve the prediction accuracy through dynamic adjustment.

[0147] In an alternative embodiment, according to the initial path after achieving the smooth obstacle avoidance, the photovoltaic cleaning robot is controlled to move along the set path, and obstacles in front are detected in real time. When an obstacle is detected, a local obstacle avoidance strategy is executed in combination with the current state of the robot and the obstacle feature information, including:

[0148] Taking the initial path as the set path, controlling the photovoltaic cleaning robot to move along the set path; during the movement of the photovoltaic cleaning robot, the current position coordinates, current orientation angle, and current speed are obtained in real time, and at the same time, the position coordinates and envelope radius of the obstacle in front are detected in real time;

[0149] The position error is calculated based on the current position coordinates and the position coordinates of the corresponding path points on the set path, and the angle error is calculated based on the current orientation angle and the orientation angle of the corresponding path points to construct a pose error vector;

[0150] Based on the pose error vector, a linear velocity control quantity is obtained by multiplying the position error by a first control gain parameter, and an angular velocity control quantity is obtained by multiplying the position error by a second control gain parameter;

[0151] The difference between the position coordinates of the obstacle and the current position coordinates is calculated to obtain a relative position vector, and the modulus of the relative position vector is calculated to obtain a relative distance;

[0152] It is judged whether the relative distance satisfies a safety distance constraint, where the safety distance constraint is that the relative distance is greater than the sum of the minimum safety distance, the envelope radius and the robot radius;

[0153] When the safety distance constraint is satisfied, the linear velocity control quantity and the angular velocity control quantity are kept unchanged; when the safety distance constraint is not satisfied, a local obstacle avoidance strategy is executed.

[0154] The initial path after smooth obstacle avoidance is realized, and then the photovoltaic cleaning robot is controlled to move along the set path, and the obstacles in front are detected in real time. When an obstacle is detected, a local obstacle avoidance strategy is executed in combination with the current state of the robot and the obstacle feature information.

[0155] In practical applications, the initial path is used as the set path, and the photovoltaic cleaning robot is controlled to move along this set path. The initial path is obtained through offline planning, taking into account the layout characteristics of the photovoltaic array and known fixed obstacles such as brackets and junction boxes. The initial path is represented by a series of path points, and each path point contains position coordinates (x, y) and an orientation angle θ.

[0156] During the movement of the photovoltaic cleaning robot, the current position coordinates (x_current, y_current), the current orientation angle θ_current and the current speed v_current of the robot are obtained in real time through the carried sensor combination. The position information can be obtained through GPS, RTK or visual odometry, the orientation angle can be measured through an IMU or an electronic compass, and the speed can be calculated through an encoder. At the same time, the position coordinates (x_obstacle, y_obstacle) and the envelope radius r_obstacle of the obstacles in front are detected in real time by detection devices such as lidar, ultrasonic sensors or depth cameras. The envelope radius represents the size characteristics of the obstacle and is estimated through the detected obstacle point cloud or contour.

[0157] To keep the robot moving along the set path, it is necessary to calculate the pose error and perform control. The position error is calculated based on the current position coordinates and the position coordinates of the corresponding path point on the set path. The corresponding path point can be the point on the set path that is closest to the current position, or the preview point (a path point a certain distance ahead of the current position). The position error $e_{position}$ is expressed as the Euclidean distance between the current position coordinates and the position coordinates of the corresponding path point. At the same time, the angle error $e_{angle}$ is calculated based on the current orientation angle and the orientation angle of the corresponding path point. The angle error is the difference between the two angles and is normalized to the range of $[-\pi, \pi]$. The position error and the angle error constitute the pose error vector $[e_{position}, e_{angle}]$.

[0158] Based on the pose error vector, the linear velocity control quantity $v_{command}$ is obtained by multiplying the position error by the first control gain parameter $K_{position}$, and the angular velocity control quantity $\omega_{command}$ is obtained by multiplying the angle error by the second control gain parameter $K_{angle}$. In a specific embodiment, $K_{position}$ is set to 0.5 and $K_{angle}$ is set to 1.2. When the position error is large, the robot will decelerate and adjust its orientation; when the angle error is large, the robot will increase the turning force to quickly adjust its orientation.

[0159] The difference between the position coordinates of the obstacle and the current position coordinates is obtained to get the relative position vector $[x_{obstacle}-x_{current}, y_{obstacle}-y_{current}]$, and the modulus of this relative position vector is calculated to obtain the relative distance $d_{relative}$. The relative distance represents the distance from the center of the robot to the center of the obstacle.

[0160] It is judged whether the relative distance satisfies the safety distance constraint. The safety distance constraint is defined as the relative distance $d_{relative}$ being greater than the sum of the minimum safety distance $d_{safety}$, the obstacle envelope radius $r_{obstacle}$, and the robot radius $r_{robot}$. In a specific example, $d_{safety}$ is set to 0.3 meters, the radius of the photovoltaic cleaning robot $r_{robot}$ is 0.4 meters, and the detected obstacle envelope radius $r_{obstacle}$ is 0.2 meters. Then the safety distance constraint is $d_{relative}>(0.3 + 0.2 + 0.4)=0.9$ meters.

[0161] When the safety distance constraint is satisfied, the linear velocity control quantity $v_{command}$ and the angular velocity control quantity $\omega_{command}$ remain unchanged, and the robot continues to move along the set path. When the safety distance constraint is not satisfied, a local obstacle avoidance strategy is executed.

[0162] The local obstacle avoidance strategy adopts a variant of the artificial potential field method, and superimposes the obstacle avoidance control quantity on the basis of maintaining the original control quantity. The obstacle avoidance control quantity is inversely proportional to the relative distance and is related to the direction of the relative position vector. Specifically, the obstacle avoidance linear velocity control quantity v_obstacle is the negative value of K_obstacle multiplied by (1 / d_relative - 1 / (d_safe + r_obstacle + r_robot)) multiplied by the projection of the unit vector of the relative position vector in the forward direction of the robot, where K_obstacle is the obstacle avoidance gain parameter, which is set to 0.8. The obstacle avoidance angular velocity control quantity ω_obstacle is K_rotation multiplied by the projection of the unit vector of the relative position vector in the left direction of the robot, and K_rotation is set to 1.5.

[0163] The control commands are v_command_final = v_command + v_obstacle and ω_command_final = ω_command + ω_obstacle. This method enables the robot to gradually decelerate and detour when approaching an obstacle, and resume traveling along the set path after moving away from the obstacle.

[0164] If the robot deviates too far from the set path (such as the deviation distance exceeds 1.5 meters), the path replanning mechanism will be triggered to generate a new local path to return to the original set path. The replanning uses the Rapidly-exploring Random Tree (RRT) algorithm, considering the current position and the obstacle distribution, to plan a collision-free path back to an appropriate point on the original set path.

[0165] To handle the situation of multiple obstacles, calculate the obstacle avoidance control quantity for each detected obstacle, and take the vector sum as the final obstacle avoidance control quantity. This method can handle the multi-obstacle scenario in a complex environment.

[0166] Actual tests show that in a standard photovoltaic array environment, this method enables the photovoltaic cleaning robot to travel at an average speed of 0.5 m / s, successfully avoid various obstacles, including temporarily placed tools, falling objects, etc., the maximum distance deviating from the set path does not exceed 0.8 meters, the path tracking accuracy is above 90%, and the obstacle avoidance success rate reaches 98%.

[0167] Compared with the existing technology, the traditional obstacle avoidance methods for photovoltaic cleaning robots usually adopt simple stop-detour strategies or preset obstacle avoidance modes, lacking smoothness and adaptability. These methods often need to stop completely when encountering an obstacle, and then execute predefined obstacle avoidance actions, resulting in discontinuous movement, low efficiency, and difficulty in adapting to complex and changeable obstacle environments.

[0168] Figure 5 Schematic diagram of the influence of the obstacle avoidance gain parameter K_obstacle of the embodiment of the present invention on the system performance:

[0169] This figure shows a relationship curve of three different indicators varying with the obstacle avoidance gain parameter K. The horizontal axis represents the value range of the obstacle avoidance gain parameter K from 0.2 to 2.0, and the left side of the vertical axis represents the accuracy indicator in percentage, while the right side represents the speed indicator in m / s. The figure contains three curves: path tracking accuracy (triangle markers), obstacle avoidance success rate (circle markers), and average driving speed (square markers). As the value of K increases, the obstacle avoidance success rate (circle) gradually rises from around 85%, reaches the highest point of approximately 99% between K = 1.2 - 1.6, and then slightly decreases. The path tracking accuracy (triangle) shows a similar trend, rising from around 82% and reaching a peak of approximately 97% at K = 1.2, and then slowly decreasing. The average driving speed (square), however, shows a continuous downward trend, dropping from the initial 0.55 m / s to around 0.32 m / s. Overall observation reveals that the system performance is optimal when the value of K is around 1.2, at which time both the obstacle avoidance success rate and path tracking accuracy reach relatively high levels, while maintaining a moderate driving speed. This indicates that the selection of the obstacle avoidance gain parameter K requires a trade - off among safety (obstacle avoidance success rate), accuracy (tracking accuracy), and efficiency (driving speed).

[0170] The method proposed in this application has been improved in the following aspects: First, a continuous control strategy is adopted to organically combine path tracking control and obstacle avoidance control, enabling the robot to smoothly avoid obstacles without having to completely stop; Second, a safety distance constraint based on obstacle characteristics is introduced to make the obstacle avoidance behavior more reasonable; Third, an adaptive control parameter adjustment mechanism is designed, which can dynamically adjust the obstacle avoidance force according to the distance of the obstacle. These improvements enable the photovoltaic cleaning robot to complete the cleaning task more intelligently and efficiently, reduce the time loss caused by obstacle avoidance, improve the overall operation efficiency, and ensure the safety of the equipment and photovoltaic panels at the same time.

[0171] In an optional implementation manner, calculate the position error according to the position coordinates of the current position of the robot and the corresponding path point of the set path, calculate the angle error according to the current orientation angle and the orientation angle of the corresponding path point, and construct a pose error vector including:

[0172] Obtain the current position coordinates and the current orientation angle of the robot, and find the path point on the preset path that is closest to the current position coordinates as the corresponding path point;

[0173] Calculate the position error between the current position coordinates and the position coordinates of the corresponding path point in the local coordinate system of the robot respectively, and calculate the included angle between the current orientation angle and the orientation angle of the corresponding path point to obtain the angle error; form a pose error vector with the position error and the angle error.

[0174] Calculate the position error based on the position coordinates of the current position and the corresponding path points on the set path, then calculate the angle error based on the current orientation angle and the orientation angle of the corresponding path point, and finally construct the pose error vector.

[0175] The photovoltaic cleaning robot is equipped with a variety of positioning sensors, including GPS receivers, inertial measurement units, encoders, etc. The data of these sensors are fused and processed by a Kalman filter to obtain the current position coordinates (x_current, y_current) and the current orientation angle θ_current of the robot in the global coordinate system. The position coordinate accuracy can reach ±2 cm, and the orientation angle accuracy can reach ±1 degree.

[0176] The set path is generated by the path planning module and is represented by a series of discrete path points. Each path point contains three key pieces of information: x coordinate, y coordinate, and orientation angle. For example, the i-th path point can be represented as (x_path_i, y_path_i, θ_path_i). In practical applications, the distance between adjacent path points is usually set to 10 cm to ensure the smoothness of the path and the control accuracy.

[0177] Find the path point on the preset path that is closest to the current position coordinates as the corresponding path point. The specific steps are as follows: traverse all the path points on the preset path, calculate the Euclidean distance from each path point to the current position, and select the path point with the minimum distance as the corresponding path point.

[0178] Assume that the path contains N path points. For each path point (x_path_i, y_path_i), calculate the distance d_distance_i between it and the current position (x_current, y_current), that is, d_distance_i is equal to sqrt((x_path_i - x_current)^2 + (y_path_i - y_current)^2), where sqrt represents the square root. Find the index j that makes d_distance_i the smallest, then the corresponding path point is (x_path_j, y_path_j, θ_path_j).

[0179] Adopt a local search strategy. Based on the moving speed and control period of the robot, a search window can be determined, and the search is only carried out within a certain range near the previous corresponding path point. For example, if the maximum speed of the robot is 0.5 m / s and the control period is 0.1 s, the search window can be set to 10 points before and after the previous corresponding path point.

[0180] Assume that the current position coordinates are (2.35, 1.47). By traversing and calculating, the closest path point is the 56th point, whose coordinates are (2.31, 1.52) and the orientation angle is 0.78 radians. This point is the corresponding path point.

[0181] Calculate the position error and the angle error. Since the motion control of the robot is usually carried out in the local coordinate system of the robot, it is necessary to convert the position error in the global coordinate system to the local coordinate system of the robot. The origin of the local coordinate system of the robot is located at the center of the robot, the x-axis is along the forward direction of the robot, and the y-axis is perpendicular to the x-axis and points to the left.

[0182] Calculate the position error vector in the global coordinate system, that is, subtract the current position coordinates from the corresponding path point coordinates to obtain (x_path_j - x_current, y_path_j - y_current); perform coordinate transformation to convert this vector from the global coordinate system to the local coordinate system of the robot.

[0183] Coordinate transformation requires the use of a rotation matrix, which is constructed based on the current orientation angle θ_current of the robot. Specifically, let the position error vector in the global coordinate system be (x_error_global, y_error_global), then the position error in the local coordinate system of the robot is (x_error_local, y_error_local), and its calculation method is: x_error_local is equal to x_error_global multiplied by cos(θ_current) plus y_error_global multiplied by sin(θ_current), and y_error_local is equal to -x_error_global multiplied by sin(θ_current) plus y_error_global multiplied by cos(θ_current).

[0184] The position error vector in the global coordinate system is (-0.04, 0.05). Assuming the current orientation angle is 0.7 radians, the position error in the local coordinate system of the robot is (0.006, 0.063), indicating that the robot needs to move forward 0.006 meters and move left 0.063 meters to return to the preset path.

[0185] The calculation of the angle error is relatively simple. Directly subtract the current orientation angle from the orientation angle of the corresponding path point, that is, θ_angle_error is equal to θ_path_j minus θ_current. To ensure that the angle error is within the range of [-π, π], normalization processing is required: if θ_angle_error is greater than π, subtract 2π; if θ_angle_error is less than -π, add 2π.

[0186] The orientation angle of the corresponding path point is 0.78 radians, and the current orientation angle is 0.7 radians, then the angle error is 0.08 radians, indicating that the robot needs to rotate counterclockwise by 0.08 radians to be consistent with the orientation of the preset path.

[0187] Combine the position error and the angle error into a pose error vector, that is, [x_error_local, y_error_local, θ_angle_error]. In the above case, the pose error vector is [0.006, 0.063, 0.08]. This vector will be used as the input of the subsequent path tracking controller to generate control commands to make the robot move along the preset path.

[0188] A lookup table or a spatial index structure is used to accelerate the process of finding the nearest path point. For example, using the KD - tree data structure can reduce the search complexity from O(n) to O(log n), where n is the number of path points. For long paths, this optimization can significantly reduce the calculation time.

[0189] Traditional pose error calculation methods usually directly use the error in the global coordinate system or simply take the perpendicular distance from the current position to the path as the position error. These methods have problems of insufficient accuracy and unstable control when dealing with complex paths or high - speed movements.

[0190] The improvements of this application are as follows: First, the calculation of the position error in the local coordinate system of the robot is introduced, which better meets the actual requirements of robot motion control; Second, the path point closest to the distance is used as a reference instead of simply the perpendicular projection point, which can better handle path segments with large curvatures; Third, the normalization processing of the angle error avoids the jump problem when the angle crosses the ±π boundary.

[0191] These improvements make the construction of the pose error vector more accurate and stable, providing more reliable input information for the path tracking control of the photovoltaic cleaning robot. Practical applications show that using the pose error vector constructed by this method can improve the path tracking accuracy of the robot by more than 30%. Especially in the turning and speed - changing sections, the tracking effect is significantly better than traditional methods, greatly improving the efficiency and quality of photovoltaic cleaning operations.

[0192] In the second aspect of the embodiments of the present invention,

[0193] A photovoltaic cleaning robot sensor data fusion and obstacle avoidance system is provided, including:

[0194] A first unit for obtaining multi - source sensor data of the photovoltaic cleaning robot, where the multi - source sensor data includes surface image data, three - dimensional point cloud data, and attitude data;

[0195] A second unit for performing probability distribution modeling on the multi - source sensor data based on the Gaussian mixture model and performing multi - source data fusion using the Kalman filter algorithm with an adaptive covariance matrix, where the adaptive covariance matrix is updated in real time according to the noise characteristics of the multi - source sensor data and environmental changes;

[0196] A third unit for constructing a three - dimensional semantic map according to the fused multi - source sensor data and marking the position, type, and size information of obstacles in the three - dimensional semantic map;

[0197] The fourth unit is used to perform obstacle avoidance path planning on the three-dimensional semantic map. The three-dimensional semantic map includes a global path planning layer and a local obstacle avoidance layer. Among them, the global path planning layer generates an initial path based on the Kalman filtering algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance;

[0198] The fifth unit is used to control the photovoltaic cleaning robot to move along the set path according to the initial path after achieving the smooth obstacle avoidance, and to detect obstacles ahead in real time. When an obstacle is detected, a local obstacle avoidance strategy is executed in combination with the current state of the robot and the obstacle feature information.

[0199] In the third aspect of the embodiments of the present invention,

[0200] There is provided an electronic device, including:

[0201] A processor;

[0202] A memory for storing instructions executable by the processor;

[0203] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0204] In the fourth aspect of the embodiments of the present invention,

[0205] There is provided a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0206] The present invention can be a method, apparatus, system, and / or computer program product. The computer program product can include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for sensor data fusion and obstacle avoidance of a photovoltaic cleaning robot, characterized in that Including: Obtain multi-source sensor data of a photovoltaic cleaning robot, where the multi-source sensor data includes surface image data, three-dimensional point cloud data, and attitude data; Perform probability distribution modeling on the multi-source sensor data based on a Gaussian mixture model, and use a Kalman filtering algorithm with an adaptive covariance matrix for multi-source data fusion, where the adaptive covariance matrix is updated in real time according to the noise characteristics of the multi-source sensor data and environmental changes; Construct a three-dimensional semantic map based on the fused multi-source sensor data, and mark the position, type, and size information of obstacles in the three-dimensional semantic map; Perform obstacle avoidance path planning on the three-dimensional semantic map. The three-dimensional semantic map includes a global path planning layer and a local obstacle avoidance layer. Among them, the global path planning layer generates an initial path based on the Kalman filtering algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance; According to the initial path after achieving the smooth obstacle avoidance, control the photovoltaic cleaning robot to move along the set path, and detect obstacles ahead in real time. When an obstacle is detected, combine the current state of the robot and the obstacle feature information to execute a local obstacle avoidance strategy.

2. The method according to claim 1, wherein Performing probability distribution modeling on the multi-source sensor data based on a Gaussian mixture model and using a Kalman filtering algorithm with an adaptive covariance matrix for multi-source data fusion includes: Construct the multi-source sensor data into a data set, and construct a mixture model with multiple Gaussian components for the data set. Each Gaussian component includes a mixing weight, a mean vector, and a covariance tensor; Use the expectation maximization algorithm to iteratively optimize the parameters of the mixture model, calculate the expected value of the latent variable by calculating the posterior probability, and update the mixing weight, the mean vector, and the covariance tensor based on the expected value of the latent variable to obtain an innovation sequence; Calculate an adaptive factor based on the innovation sequence, multiply the adaptive factor by the adaptive covariance matrix, and obtain a corrected adaptive covariance matrix through the Kalman filtering algorithm; Use the corrected prediction error covariance to correct the Kalman filtering algorithm to obtain a corrected Kalman gain; update the state estimate value and the error covariance based on the corrected Kalman gain to achieve adaptive fusion of the multi-source sensor data.

3. The method according to claim 2, wherein Performing probability distribution calculation on the features of the photovoltaic panel surface image data and the features of the three-dimensional point cloud data in the multi-source sensor data, and determining the weight coefficient of the feature data according to the result of the probability distribution calculation includes: Perform multi-scale decomposition on the photovoltaic panel surface image data in the multi-source sensor data to obtain image features, where the image features include edge features, texture features, and local gradient features; extract geometric features from the three-dimensional point cloud data in the multi-source sensor data to obtain point cloud features, where the point cloud features include normal vector features, curvature features, and local depth change features; The deviation between the image feature and its feature mean value is used for differential calculation to obtain the image probability distribution value, and the deviation between the point cloud feature and its geometric feature mean value is used for weighted calculation to obtain the point cloud probability distribution value; The weight coefficient is determined according to the covariance between the image feature and the point cloud feature and the ratio of the image probability distribution value to the point cloud probability distribution value.

4. The method according to claim 1, wherein The global path planning layer generates an initial path based on the Kalman filtering algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints. The realization of smooth obstacle avoidance includes: Obtain the historical motion trajectory data of the photovoltaic cleaning robot, and generate an initial state prediction result according to the historical motion trajectory data; The Kalman filtering algorithm is used to optimize the initial state prediction result to obtain an optimized state prediction result; The optimized state prediction result is input into the path planning model, and the cost values of different paths are calculated according to the path length, turning angle, obstacle distance, and energy consumption. The calculation process includes: multiplying each parameter by the corresponding weight coefficient and then summing to obtain the comprehensive cost value, and selecting the path with the smallest comprehensive cost value as the candidate path; Smooth the candidate path, evaluate the feasibility of the candidate path according to the path smoothness, safety distance, and task time, and select the optimal path as the global planning path; Obtain the real-time obstacle distance information, relative speed information, and boundary constraint information around the photovoltaic cleaning robot, and normalize the obstacle distance, relative speed, and boundary distance in combination with the global planning path and then perform weighted summation to obtain the obstacle avoidance parameter; Multiply the obstacle avoidance parameter by the preset speed adjustment coefficient to obtain the speed correction value, and realize smooth obstacle avoidance based on the obstacle avoidance parameter and the speed correction amount.

5. The method according to claim 4, characterized in that, An adaptive factor is introduced into the initial state prediction result, the state transition matrix and the observation matrix are dynamically adjusted according to the prediction error, and the Kalman filtering algorithm is used to optimize the initial state prediction result. The obtained optimized state prediction result includes: Obtain the system state vector, which includes the position coordinates, orientation angle, linear velocity, and angular velocity, and use the expectation operation to calculate the expected value of the error product of the system state vector and the state estimation vector to obtain the prediction error covariance matrix; Calculate the ratio of the prediction error covariance matrix to the theoretical measurement residual covariance matrix to obtain the adaptive factor; Construct a diagonal adaptive weight matrix according to the dimension of the system state vector with the adaptive factor, and the diagonal elements of the diagonal adaptive weight matrix are the adaptive factors corresponding to different state components; Perform matrix multiplication on the diagonal adaptive weight matrix and the state transition matrix to obtain the adjusted state transition matrix, and perform matrix multiplication on the diagonal adaptive weight matrix and the difference between the current moment state vector and the state prediction value to obtain the prediction error compensation term; Left-multiply the state vector at the previous moment by the adjusted state transition matrix and add it to the prediction error compensation term to obtain the optimized state prediction result.

6. The method according to claim 1, characterized in that, According to the initial path after realizing the smooth obstacle avoidance, control the photovoltaic cleaning robot to move along the set path, and detect obstacles ahead in real time. When an obstacle is detected, combined with the current state of the robot and the obstacle feature information, execute the local obstacle avoidance strategy, including: Take the initial path as the set path and control the photovoltaic cleaning robot to move along the set path; during the movement of the photovoltaic cleaning robot, obtain the current position coordinates, the current orientation angle and the current speed in real time, and at the same time detect the position coordinates and the envelope radius of the obstacle ahead in real time; Calculate the position error according to the position coordinates of the current position and the corresponding path point of the set path, calculate the angle error according to the current orientation angle and the orientation angle of the corresponding path point, and construct a pose error vector; Based on the pose error vector, obtain the linear velocity control quantity from the product of the position error and the first control gain parameter, and obtain the angular velocity control quantity from the product of the position error and the second control gain parameter; Subtract the position coordinates of the obstacle from the current position coordinates to obtain a relative position vector, and calculate the modulus of the relative position vector to obtain a relative distance; Judge whether the relative distance satisfies the safety distance constraint, and the safety distance constraint is that the relative distance is greater than the sum of the minimum safety distance, the envelope radius and the robot radius; When the safety distance constraint is satisfied, keep the linear velocity control quantity and the angular velocity control quantity unchanged; when the safety distance constraint is not satisfied, execute the local obstacle avoidance strategy.

7. The method according to claim 6, characterized in that Calculate the position error according to the position coordinates of the current position and the corresponding path point of the set path, calculate the angle error according to the current orientation angle and the orientation angle of the corresponding path point, and construct a pose error vector, including: Obtain the current position coordinates and the current orientation angle of the robot, and find the path point with the closest distance to the current position coordinates on the preset path as the corresponding path point; Calculate the position error of the position coordinates of the current position and the corresponding path point in the local coordinate system of the robot respectively, and calculate the included angle between the current orientation angle and the orientation angle of the corresponding path point to obtain the angle error; form a pose error vector with the position error and the angle error.

8. A photovoltaic cleaning robot sensor data fusion and obstacle avoidance system for implementing the method described in any one of the preceding claims 1-7, characterized in that, Include: The first unit is used to obtain the multi-source sensor data of the photovoltaic cleaning robot, and the multi-source sensor data includes surface image data, three-dimensional point cloud data and attitude data; The second unit is used to perform probability distribution modeling on the multi-source sensor data based on the Gaussian mixture model, and perform multi-source data fusion using the Kalman filtering algorithm with an adaptive covariance matrix, wherein the adaptive covariance matrix is updated in real time according to the noise characteristics of the multi-source sensor data and environmental changes; The third unit is used to construct a three-dimensional semantic map according to the fused multi-source sensor data, and mark the position, type and size information of the obstacles in the three-dimensional semantic map; The fourth unit is used to plan an obstacle avoidance path for the three-dimensional semantic map. The three-dimensional semantic map includes a global path planning layer and a local obstacle avoidance layer. Among them, the global path planning layer generates an initial path based on the Kalman filtering algorithm, and the local obstacle avoidance layer uses a fuzzy logic controller to dynamically adjust the motion trajectory according to the obstacle distance, relative speed, and photovoltaic panel boundary constraints to achieve smooth obstacle avoidance; The fifth unit is used to control the photovoltaic cleaning robot to move along a set path according to the initial path after achieving the smooth obstacle avoidance, and to detect obstacles ahead in real time. When an obstacle is detected, a local obstacle avoidance strategy is executed in combination with the current state of the robot and the obstacle feature information.

9. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images