An explosion-proof intelligent visual navigation method and system based on camera

Abstract

The present invention discloses an explosion-proof intelligent visual navigation method and system based on a camera, and relates to the field of visual navigation technology. The method comprises the following steps: obtaining characteristic data of obstacles; obtaining a fused similarity matrix; obtaining a navigation demand change curve and an environment change curve; combining the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and applying a prediction model based on the comprehensive data set, the fused similarity matrix and the label data to obtain the matching score of the obstacles; taking obstacles with high matching degree as part of the navigation path; the system comprises a characteristic data acquisition module, a similarity fusion module, a data extraction module, a matching degree acquisition module and a navigation path confirmation module. The present invention not only makes decisions based on the characteristics of a single obstacle, but can also integrate the interactions and correlations of multiple obstacles to provide a more comprehensive and effective navigation solution.

Description

Technical Field

[0001] The present invention relates to the field of visual navigation technology, and in particular to a camera-based explosion-proof intelligent visual navigation method and system. Background Art

[0002] Intelligent visual navigation is a technology that uses visual perception technology to identify and understand the environment, thereby achieving autonomous navigation and path planning. It mainly relies on cameras as sensors, combined with image processing and machine learning algorithms to capture and analyze environmental data to achieve accurate identification and geographic positioning of the surrounding environment. Intelligent visual navigation greatly improves the environmental adaptability and decision-making intelligence of the automation system by integrating advanced image recognition technology and intelligent algorithms. It is one of the key technologies in the field of modern automation and artificial intelligence.

[0003] The application of explosion-proof intelligent visual navigation methods based on cameras in hazardous environments is extremely important, especially in chemical plants, oil refineries and other places, which are full of flammable and explosive substances. Traditional navigation equipment or methods may not be enough to cope with the complexity and potential dangers. The use of specific explosion-proof cameras can not only withstand the harsh conditions of these environments, but also provide the necessary safety guarantees.

[0004] In the prior art, for example, Chinese patent 201910276428.1 discloses a passive visual navigation unmanned forklift system and navigation method for explosion-proof warehouses, which includes explosion-proof cameras, explosion-proof controllers, sensors, alarms, and mechanical anti-collision devices to achieve unmanned operation of passive visual navigation in explosion-proof warehouses to avoid harm to human health. However, the above method still has the following shortcomings: it processes and analyzes the real-time beacon information of the forklift and the corresponding stored beacon information to determine the real-time position of the forklift. However, in a dynamic environment, the interactions between obstacles may change frequently, such as in a high-mobility warehouse. If these changes cannot be updated and responded to quickly, the effectiveness of its decision-making will be limited.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes a camera-based explosion-proof intelligent visual navigation method and system to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a camera-based explosion-proof intelligent visual navigation method is provided, and the camera-based explosion-proof intelligent visual navigation method comprises the following steps:

[0009] S1. Acquire environmental images based on explosion-proof cameras, identify obstacles, and obtain feature data of obstacles.

[0010] S2. Based on the feature data of the obstacles, the similarity of the overlapping positions of the obstacles and the cosine similarity between the vectors are calculated, and similarity fusion is performed to obtain a fused similarity matrix.

[0011] S3. Obtain historical navigation log data, and use time series analysis to obtain the periodicity and trend of navigation demand, obtain a navigation demand change curve, and use an event detection algorithm to identify significant points of environmental change, and obtain an environmental change curve.

[0012] S4. Combine the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and apply the prediction model based on the comprehensive data set, the fusion similarity matrix and the label data to obtain the obstacle matching score.

[0013] S5. Consider obstacles with a high matching degree as part of the navigation path, and plan a navigation path from the starting point to the end point, passing through all selected obstacles with a high matching degree.

[0014] Furthermore, based on the characteristic data of the obstacles, similarity calculation of the overlapping positions of the obstacles and cosine similarity calculation between vectors are performed, and similarity fusion is performed to obtain a fused similarity matrix, which includes the following steps:

[0015] S21. Based on the position and size information of the obstacle, represent each obstacle as a geometric shape;

[0016] S22, obtaining the overlapping area between two obstacles and calculating the intersection area;

[0017] S23, calculating the area of ​​the minimum geometric shape of all spaces of the two obstacles to obtain the union area;

[0018] S24, the Jaccard index is calculated by the ratio of the intersection area to the union area;

[0019] S25, constructing a vector with the characteristic data of the obstacle, and using the various dimension values ​​of the vector to perform multiplication and accumulation to obtain the dot product result of the obstacle, and obtaining the calculation modulus of the obstacle by calculating the square root of the sum of the squares of the components of the vector;

[0020] S26. Divide the dot product of the two vectors by the modulus length of each vector to obtain the cosine similarity;

[0021] S27. For each pair of obstacles, the Jaccard index and the cosine similarity score are fused and a fused similarity matrix is ​​constructed.

[0022] Furthermore, obtaining historical navigation log data, and using time series analysis to obtain the periodicity and trend of navigation demand, obtaining a navigation demand change curve, and using an event detection algorithm to identify significant points of environmental change, obtaining an environmental change curve includes the following steps:

[0023] S31, collecting historical navigation log data, and cleaning the historical navigation log data;

[0024] S32. Apply time series analysis methods to identify periodic patterns in data;

[0025] S33. Use trend analysis techniques to identify long-term trends;

[0026] S34. Use a time series diagram to display the navigation demand change curve;

[0027] S35. Obtain significant points of environmental change based on the Bayesian model and environmental monitoring data to obtain an environmental change curve.

[0028] Furthermore, obtaining significant points of environmental change based on the Bayesian model and environmental monitoring data to obtain an environmental change curve includes the following steps:

[0029] S351. Obtain environmental monitoring data, and perform data cleaning and data coding;

[0030] S352, defining a Gaussian kernel function and setting a bandwidth parameter for calculating the similarity between two data points;

[0031] S353, using the Gaussian kernel function to iteratively calculate the data of all environmental monitoring points to obtain a similarity matrix;

[0032] S354, establishing a Bayesian model based on prior knowledge and newly acquired data, and dynamically updating the similarity matrix according to the new data and the results of Bayesian inference;

[0033] S355, performing spectral clustering analysis on the updated similarity matrix to obtain data grouping results;

[0034] S356: Use the grouping results to identify significant points of environmental change, and based on the identified significant points and related environmental events, draw an environmental change curve representing the frequency and pattern of obstacle changes.

[0035] Furthermore, the formula of the Gaussian kernel function is:

[0036]

[0037] Where K(x, y) represents the similarity value between two data points x and y;

[0038] σ represents the bandwidth parameter.

[0039] Further, performing spectral clustering analysis on the updated similarity matrix to obtain data grouping results includes the following steps:

[0040] S3551. constructing an adjacency matrix of the undirected weighted graph based on the updated similarity matrix, and constructing a corresponding graph Laplacian matrix from the adjacency matrix, and performing standardization;

[0041] S3552. Calculate the eigenvectors of the normalized graph Laplacian matrix to identify the structure and grouping of data.

[0042] Furthermore, the statistical characteristics of the navigation demand change curve and the environment change curve are combined to obtain a comprehensive data set, and based on the comprehensive data set, the fusion similarity matrix and the label data, the prediction model is applied to obtain the matching score of the obstacle, including the following steps:

[0043] S41, merging the mean and trend characteristic values ​​of the navigation demand change and the environment change to obtain a comprehensive data set;

[0044] S42, integrating the fused similarity matrix into the comprehensive dataset and adding a corresponding similarity score for each obstacle pair;

[0045] S43, determining a matching score standard for obstacles, and assigning a matching score to each obstacle in the comprehensive data set as a label;

[0046] S44, using the comprehensive data set and the matching score to train the random forest model to obtain a prediction model;

[0047] S45. Use the built-in variable importance evaluation function of random forest to obtain the influence of the feature on the prediction matching score of the prediction model;

[0048] S46. Use cross-validation techniques to test the prediction model, and adjust the prediction model parameters and feature selection based on the prediction model performance and feature importance feedback;

[0049] S47. Use the trained prediction model to predict the matching score of the new obstacle to obtain the matching score of the obstacle.

[0050] Furthermore, using the built-in variable importance evaluation function of random forest, obtaining the influence of the feature on the prediction matching score of the prediction model includes the following steps:

[0051] S451. For each decision tree, use samples that have not been trained by the decision tree to evaluate the performance of the prediction model;

[0052] S452, randomly permuting the value of each feature, and obtaining the effect of the permutation on the out-of-bag error of the prediction model, and obtaining the importance score of each feature by comparing the out-of-bag errors before and after the permutation;

[0053] S453, averaging the feature importance scores calculated by all decision trees to obtain an average importance score for each feature;

[0054] S454. Sort the features according to the obtained average importance scores.

[0055] Furthermore, when determining the obstacle matching score standard, the obstacle matching score calculation formula is:

[0056] S = w1P - w2T;

[0057] Where P represents the navigation success rate of obstacles;

[0058] T represents the average delay time caused by obstacles;

[0059] w1 and w2 represent the weights of the navigation success rate of obstacles and the average delay time caused by obstacles.

[0060] According to another aspect of the present invention, a camera-based explosion-proof intelligent visual navigation system is provided. The camera-based explosion-proof intelligent visual navigation system includes a feature data acquisition module, a similarity fusion module, a data extraction module, a matching degree acquisition module and a navigation path confirmation module.

[0061] Among them, the feature data acquisition module is used to acquire environmental images based on the explosion-proof camera, identify obstacles, and acquire feature data of the obstacles.

[0062] The similarity fusion module calculates the similarity of the overlapping positions of the obstacles and the cosine similarity between the vectors based on the feature data of the obstacles, and performs similarity fusion to obtain a fused similarity matrix.

[0063] The data extraction module is used to obtain historical navigation log data, and use the time series analysis method to obtain the periodicity and trend of navigation demand to obtain the navigation demand change curve, and use the event detection algorithm to identify the significant points of environmental change to obtain the environmental change curve.

[0064] The matching degree acquisition module is used to combine the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and based on the comprehensive data set, the fusion similarity matrix and the label data, the prediction model is applied to obtain the matching degree score of the obstacle.

[0065] The navigation path confirmation module is used to take obstacles with high matching degree as part of the navigation path, and plan a navigation path from the starting point to the end point, passing through all the selected obstacles with high matching degree.

[0066] The beneficial effects of the present invention are:

[0067] (1) The present invention provides a camera-based explosion-proof intelligent visual navigation method and system, which can ensure the safe acquisition of images in flammable and explosive environments and reduce the risk of safety accidents by using explosion-proof cameras. By obtaining a fused similarity matrix and using the fused similarity matrix for the training and application of the prediction model, the relative position and feature similarity between obstacles can be taken into account, which helps to improve the accuracy of the prediction matching score. Because similar obstacles will have similar effects on the navigation system, the similarity matrix reveals the potential spatial and functional connections between obstacles, allowing the navigation system to better plan the path. The data of the fused similarity matrix allows the prediction model to make decisions not only based on the characteristics of a single obstacle, but also to integrate the interactions and correlations of multiple obstacles to provide a more comprehensive and effective navigation solution.

[0068] (2) Time series analysis provides periodic and trend analysis of navigation needs, helping to predict future navigation needs and optimize resource allocation. Event detection algorithms can quickly identify environmental changes, allowing the navigation system to adjust its strategy in a timely manner and improve the speed and accuracy of its response to environmental changes. By constantly learning and adapting to environmental changes and changes in navigation needs, the prediction model can better adapt to complex and changing operating environments, which is the key to efficient navigation and avoiding potential obstacles. At the same time, multi-source data is integrated, and the matching degree of obstacles is predicted using a random forest model to provide more scientific and accurate decision support. Through continuous training and model adjustment, the navigation system can adapt to environmental changes and improve the reliability and efficiency of long-term operation. Accurate matching scores make path planning more efficient, ensuring that the most appropriate path is used to reach the destination. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0070] Figure 1 is a flow chart of a camera-based explosion-proof intelligent visual navigation method according to an embodiment of the present invention;

[0071] Figure 2The figure is a principle block diagram of a camera-based explosion-proof intelligent visual navigation system according to an embodiment of the present invention.

[0072] In the figure:

[0073] 1. Feature data acquisition module; 2. Similarity fusion module; 3. Data extraction module; 4. Matching degree acquisition module; 5. Navigation path confirmation module. DETAILED DESCRIPTION

[0074] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0075] According to an embodiment of the present invention, a camera-based explosion-proof intelligent visual navigation method and system are provided.

[0076] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a camera-based explosion-proof intelligent visual navigation method is provided, and the camera-based explosion-proof intelligent visual navigation method comprises the following steps:

[0077] S1. Obtain environmental images based on explosion-proof cameras, identify obstacles such as reactors, storage tanks, and pipelines (equivalent to a node on the navigation path), and obtain feature data of the obstacles, which include location and size information.

[0078] Specifically, an explosion-proof camera is used to capture real-time images in a specific environment. The captured images undergo a series of preprocessing to improve the efficiency and accuracy of subsequent processing, including noise removal, contrast enhancement, and image cropping.

[0079] Use image recognition technology to identify specific obstacles (such as reactors, storage tanks, and pipelines) in the image and extract their key features, such as position and size. The position can be determined by the pixel coordinates in the image, while the size can be compared with a reference object of known size or calculated by the imaging parameters of the camera.

[0080] S2. Based on the feature data of the obstacles, the similarity of the overlapping positions of the obstacles and the cosine similarity between the vectors are calculated, and similarity fusion is performed to obtain a fused similarity matrix.

[0081] In a further embodiment, based on the feature data of the obstacles, similarity calculation of overlapping positions of the obstacles and cosine similarity calculation between vectors are performed, and similarity fusion is performed to obtain a fused similarity matrix, which includes the following steps:

[0082] S21. Based on the position and size information of the obstacle, represent each obstacle as a geometric shape.

[0083] S22. Obtain the overlapping area between two obstacles and calculate the intersection area.

[0084] S23. Calculate the area of ​​the minimum geometric shape of all spaces between the two obstacles to obtain the union area.

[0085] S24. The Jaccard index is calculated by the ratio of the intersection area to the union area. The Jaccard index is a statistic used to compare the similarities and differences between sample sets. The Jaccard index is defined as the ratio between the intersection size and the union size of two sets, and its value ranges from 0 to 1.

[0086] S25. Construct a vector using the characteristic data of the obstacle, and use the values ​​of each dimension of the vector to perform multiplication and accumulation to obtain a dot product result of the obstacle, and obtain the calculation modulus of the obstacle by calculating the square root of the sum of the squares of each component of the vector.

[0087] S26. Divide the dot product of the two vectors by the modulus of each vector to obtain the cosine similarity.

[0088] Specifically, the features of each obstacle (e.g., length, width, height) are converted into a vector. The dot product (also called inner product or scalar product) is an operation on two vectors that multiplies the corresponding coordinates of the two vectors and sums them. The magnitude (or length) of a vector is the magnitude of the vector and is obtained by calculating the square root of the sum of the squares of the vector's components. Cosine similarity is calculated by dividing the dot product of the two vectors by the product of their magnitudes. This value ranges from -1 to 1. For example, if the cosine similarity of two obstacles is close to 1, it means that they are very similar in direction; if it is close to 0, it means that they are orthogonal; if it is -1, they are completely opposite in direction.

[0089] S27. For each pair of obstacles, the Jaccard index and the cosine similarity score are fused and a fused similarity matrix is ​​constructed.

[0090] For each pair of obstacles, the Jaccard and cosine similarity scores are fused. The fusion can be performed by simple averaging, weighted averaging, or other statistical methods. For example, the Jaccard score focuses more on the actual physical overlap, while the cosine score focuses on the attribute similarity, and the weights of the two will be adjusted according to specific navigation requirements.

[0091] In order to facilitate understanding of the above technical solutions of the present invention, the working principle or operation mode of the present invention in the actual process is described in detail below.

[0092] Specifically, determine that the obstacle is a geometric shape, for example, the reactor is circular and the storage tank is rectangular. Calculate the intersection area of ​​the two obstacles, for example, the intersection area of ​​the reactor and the storage tank is 20 square meters. Calculate the union area, for example, the union area of ​​the reactor and the storage tank is 100 square meters. The Jaccard index is 20 / 100=0.2. Construct a feature vector and calculate the dot product, for example, the reactor vector is [1, 0, 2], the tank vector is [2, 1, 0], and the dot product is 4. The cosine similarity is 0.8. The fusion score is (0.2+0.8) / 2=0.5.

[0093] S3. Obtain historical navigation log data, and use time series analysis to obtain the periodicity and trend of navigation demand, obtain a navigation demand change curve, and use an event detection algorithm to identify significant points of environmental change, and obtain an environmental change curve.

[0094] In a further embodiment, obtaining historical navigation log data, and using a time series analysis method to obtain the periodicity and trend of navigation requirements, obtaining a navigation requirement change curve, and using an event detection algorithm to identify significant points of environmental changes, obtaining an environmental change curve includes the following steps:

[0095] S31. Collect historical navigation log data, including timestamp, navigation start and end points, route selection, usage frequency and other information, and clean the historical navigation log data to remove outliers and missing data to ensure data quality and provide accurate input for analysis;

[0096] S32. Apply time series analysis methods (such as autoregressive moving average (ARMA) models, seasonally decomposed time series (STL), or Fourier transforms) to identify periodic patterns in the data, which involves identifying recurring patterns at specific time intervals, such as daily or weekly peaks in navigation demand.

[0097] S33. Use trend analysis techniques (such as moving averages, exponential smoothing, or linear regression models) to identify long-term trends, such as an increase or decrease in navigation demand.

[0098] Specifically, select ARMA (autoregressive moving average model), STL (seasonal decomposition method) or Fourier transform according to the data characteristics. For the ARMA model, determine the order of autoregression (AR) and moving average (MA). Estimate the model parameters using maximum likelihood estimation. Apply the model to the data and check the goodness of fit. Check whether the residual sequence is free of autocorrelation, that is, the white noise test. Adjust the model parameters based on the test results. Use the selected model to identify periodic patterns. For STL, decompose and adjust the seasonal component.

[0099] S34. Use a time series graph to display the navigation demand change curve; a time series graph is a graph used to display the trend of data changes over time, which helps to intuitively understand how navigation demand changes over time.

[0100] S35. Obtain significant points of environmental change based on the Bayesian model and environmental monitoring data to obtain an environmental change curve.

[0101] In a further embodiment, obtaining significant points of environmental changes based on the Bayesian model and environmental monitoring data to obtain an environmental change curve includes the following steps:

[0102] S351. Obtain environmental monitoring data, such as the location of obstacles, timestamps of appearance and disappearance, etc., and perform data cleaning and data encoding; data encoding is to convert raw data into a format that can be processed by a computer, such as converting a timestamp into a date and time format, and converting location information into numerical data.

[0103] S352. Define a Gaussian kernel function and set a bandwidth parameter for calculating the similarity between two data points.

[0104] S353. Use the Gaussian kernel function to iteratively calculate the data of all environmental monitoring points to obtain a similarity matrix that reflects the similarity between the data points.

[0105] It should be noted that first, it is necessary to select a suitable Gaussian kernel function, including the setting of bandwidth parameters. Convert the data of all environmental monitoring points into feature vector form, and initialize the similarity matrix, using a zero matrix or a unit matrix as the initial value. Use the Gaussian kernel function to calculate the similarity of the feature vector and update the value of the similarity matrix. An iterative algorithm can be used to update the similarity matrix, such as iteratively solving the Gaussian kernel function. During the iterative calculation process, it is necessary to set a convergence condition, such as setting a maximum number of iterations or setting a threshold, and stop the iteration when the change in the similarity matrix is ​​less than the threshold or reaches the maximum number of iterations. After the iterative calculation is completed, the obtained similarity matrix reflects the degree of similarity between each data point.

[0106] S354, establishing a Bayesian model based on prior knowledge and newly acquired data, and dynamically updating the similarity matrix according to the new data and the results of Bayesian inference;

[0107] Specifically, the prior distribution of the Bayesian model is determined based on the existing prior knowledge and experience. That is, the probability distribution of environmental changes, such as the frequency and duration of obstacles. Newly acquired environmental monitoring data is obtained, including information such as the location of obstacles, the timestamps of their appearance and disappearance. These data will be used to update the Bayesian model. The likelihood function is calculated based on the new data, that is, the probability of observing the data under given prior conditions. This usually involves comparing the new data with the prior distribution of the Bayesian model to determine the impact of the data on the model parameters. Using Bayes' theorem, the prior distribution, likelihood function and new data are combined to perform Bayesian inference to obtain the posterior distribution. The posterior distribution reflects the update and adjustment of the model parameters under the consideration of prior information and new data. According to the posterior distribution obtained by Bayesian inference, the relevant parameters or weights in the similarity matrix are dynamically updated. That is, it is achieved by incorporating the information of the posterior distribution into the process of similarity calculation. The updated similarity matrix is ​​evaluated and verified, and the results before and after the update are compared to check whether the update is consistent with the actual situation of the prior knowledge and new data.

[0108] S355, perform spectral clustering analysis on the updated similarity matrix to obtain data grouping results; spectral clustering analysis is a clustering method based on graph theory, which converts the similarity matrix into a graph and obtains data grouping results by dividing the graph. Such grouping results help identify significant points of environmental changes.

[0109] S356. Use the grouping results to identify significant points of environmental change, and based on the identified significant points and related environmental events, draw an environmental change curve that represents the frequency and pattern of obstacle changes, reflecting the changing trends and patterns of obstacles in different time periods, which helps to understand and predict environmental changes.

[0110] In a further embodiment, the formula of the Gaussian kernel function is:

[0111]

[0112] In the formula, K(x, y) represents the similarity value between two data points x and y; data points x and y specifically refer to the feature data of obstacles identified from the environmental images captured by the explosion-proof camera. These feature data usually include information such as the location and size of the obstacle, which is converted into a numerical vector;

[0113] σ represents the bandwidth parameter, which controls the width of the Gaussian kernel function, thus affecting the similarity calculation result. The larger the bandwidth parameter, the flatter the curve of the Gaussian kernel function, and the more sensitive the similarity calculation is to data changes; conversely, the smaller the bandwidth parameter, the steeper the curve of the Gaussian kernel function, and the less sensitive the similarity calculation is to data changes.

[0114] In a further embodiment, performing spectral clustering analysis on the updated similarity matrix to obtain data grouping results includes the following steps:

[0115] S3551. constructing an adjacency matrix of the undirected weighted graph based on the updated similarity matrix, and constructing a corresponding graph Laplacian matrix from the adjacency matrix, and performing standardization;

[0116] S3552. Calculate the eigenvectors of the normalized graph Laplacian matrix to identify the main structures and groupings of the data.

[0117] In order to facilitate understanding of the above technical solutions of the present invention, the working principle or operation mode of the present invention in the actual process is described in detail below.

[0118] For historical navigation log data, navigation start and end points (A to B), path selection (Route 1, Route 2), usage frequency (peak at 8 am and 5 pm every day), etc. ARMA model is applied to identify periodic patterns. Using moving average to identify long-term trends, it is found that navigation demand has a slight upward trend every Friday. The time series chart shows the change curve of navigation demand. The chart shows that from 2023-04-01 to 2023-04-30, navigation demand has an obvious periodic pattern during the morning and evening peak hours on weekdays. Update the model using prior knowledge (obstacles usually appear on weekdays) and new data (obstacles appear on 2023-04-20). Mark 2023-04-20 as a significant change point because the frequency of obstacles on this day is abnormally high. Set the bandwidth parameter to 0.5 and calculate the similarity matrix. Convert the similarity matrix into a graph, and obtain two main data groups through spectral clustering. Based on the grouping results, the frequency and pattern of obstacles during the morning and evening peak hours are plotted, revealing the regularity of obstacle appearance.

[0119] S4. Combine the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and apply the prediction model based on the comprehensive data set, the fusion similarity matrix and the label data to obtain the obstacle matching score.

[0120] In a further embodiment, the statistical features of the navigation demand change curve and the environment change curve are combined to obtain a comprehensive data set, and based on the comprehensive data set, the fusion similarity matrix and the label data, the prediction model is applied to obtain the matching score of the obstacle, including the following steps:

[0121] S41. The mean and trend characteristic values ​​of the navigation demand change and the environmental change are combined to obtain a comprehensive data set for describing the characteristics of obstacles and changes in the environment.

[0122] S42, integrate the fused similarity matrix into the comprehensive data set, and add the corresponding similarity score for each obstacle pair. Specifically, for each obstacle, calculate the average of its similarity scores with all other obstacles in the data set, that is, the total similarity score, and integrate it into the comprehensive data set. An indicator reflecting the overall similarity between the obstacle and all other obstacles in the data set. For a new obstacle, if there is no similarity data with other obstacles, use the other features of the obstacle and the average of the total similarity scores of obstacles in the data set for prediction.

[0123] S43. Determine a matching score standard for the obstacle, for example, based on past navigation success rate, delay time, etc., and assign a matching score to each obstacle in the comprehensive data set as a label.

[0124] S44, using the comprehensive data set (features) and matching scores (labels) to train a random forest model, and configuring model parameters, such as the number of trees, maximum depth, etc., to obtain a prediction model;

[0125] S45. Use the built-in variable importance assessment function of random forest to obtain the influence of features on the prediction matching score of the prediction model. This helps to understand which features are more important to the prediction effect of the model.

[0126] S46. Use cross-validation technology to test the model to ensure that the model does not overfit, and adjust the prediction model parameters and feature selection based on the prediction model performance and feature importance feedback to optimize the prediction accuracy.

[0127] Specifically, first use evaluation indicators such as accuracy, precision, recall, F1-score, etc. to evaluate the performance of the model on the test data. If the model performance is poor, you need to consider adjusting the model parameters, such as increasing the number or depth of trees, adjusting the learning rate, etc. to improve the performance of the model. Models such as random forests can provide feature importance evaluation results. Feature importance reflects the degree of influence of each feature on the model's prediction results. If some features contribute less to the model's prediction, you can consider reducing the number of features or adjusting the feature weights to simplify the model and improve the prediction effect. Using cross-validation techniques can more comprehensively evaluate the performance of the model.

[0128] Based on the feature importance evaluation results, select and retain the features that have an important impact on the model prediction, and remove the features that have a small impact on the model. This can reduce the complexity of the model and improve the generalization ability of the model.

[0129] S47. Use the trained prediction model to predict the matching score of the new obstacle to obtain the matching score of the obstacle.

[0130] Specifically, for example, the navigation demand change curve shows that the demand increases from 5 to 7 pm every Friday, and the environment change curve shows that the frequency of obstacles increases during these periods. Combine the mean and trend feature values ​​of the two curves to create a data set containing time, demand increase, and obstacle frequency. The similarity score between obstacles A and B is 0.8, the similarity score between obstacles B and C is 0.5, and then calculate the total similarity score for each obstacle. Assign matching score labels to the obstacles in the comprehensive data set. Train the random forest model using the comprehensive data set and matching scores. The random forest model makes the final prediction by building multiple decision trees and combining their predictions. Each decision tree is built based on a different random subset of the input features, and each tree outputs a prediction. The final prediction of the random forest model is the average of the predictions of these individual trees (regression problem) or the majority vote (classification problem).

[0131] In a further embodiment, using the built-in variable importance evaluation function of the random forest, obtaining the influence of the feature on the prediction matching score of the prediction model includes the following steps:

[0132] S451. For each decision tree, use samples that have not been trained by the decision tree to evaluate the performance of the prediction model.

[0133] S452. Randomly permute the value of each feature, and obtain the effect of the permute on the out-of-bag error of the prediction model, and obtain the importance score of each feature by comparing the out-of-bag error before and after the permute.

[0134] S453. For each feature, its importance score is usually measured by the increase in its out-of-bag error before and after permutation. The greater the error increase, the greater the impact of the feature on the model. The feature importance scores calculated by all decision trees are averaged to obtain the average importance score of each feature.

[0135] S454. Sort the features according to the average importance score obtained. The higher the score, the more critical the role of the feature in the prediction model. Analyze the top-ranked features and explain their role in the model and their specific impact on the prediction target. This can help understand the decision-making process of the model and guide subsequent feature engineering and model optimization.

[0136] In a further embodiment, when determining the matching score standard of the obstacle, the calculation formula of the matching score of the obstacle is:

[0137] S = w1P - w2T;

[0138] Where P represents the navigation success rate of obstacles; T represents the average delay time caused by obstacles; w1 and w2 represent the weights of the navigation success rate of obstacles and the average delay time caused by obstacles.

[0139] The range of navigation success rate is usually 0 to 1, with 1 indicating 100% success rate. Delay time is expressed as the average delay time caused by obstacles (in minutes or seconds). The weight setting depends on how important you think navigation success rate and delay time are. For example, if delay time has a greater impact on user satisfaction, you can increase the value of w2 accordingly.

[0140] S5. Consider obstacles with a high matching degree as part of the navigation path, and plan a navigation path from the starting point to the end point, passing through all selected obstacles with a high matching degree.

[0141] For example, obstacle A has a high matching score.

[0142] Obstacle B, with a high matching score.

[0143] Obstacle C, matching score is medium.

[0144] Obstacle D, low matching score.

[0145] Select all obstacles with high matching degree as required points.

[0146] Use a path planning algorithm (such as Dijkstra) to calculate the shortest path from the starting point to the end point through obstacles A and B. The path starts from the starting point, goes to obstacle A, then from obstacle A to obstacle B, and finally goes straight from obstacle B to the end point.

[0147] like Figure 2 As shown, according to another embodiment of the present invention, there is provided an explosion-proof intelligent visual navigation system based on a camera, the explosion-proof intelligent visual navigation system based on a camera comprises a feature data acquisition module 1, a similarity fusion module 2, a data extraction module 3, a matching degree acquisition module 4 and a navigation path confirmation module 5;

[0148] The feature data acquisition module 1 is used to acquire environmental images based on an explosion-proof camera, identify obstacles, and acquire feature data of the obstacles.

[0149] The similarity fusion module 2 calculates the similarity of the overlapping positions of the obstacles and the cosine similarity between the vectors based on the feature data of the obstacles, and performs similarity fusion to obtain a fused similarity matrix.

[0150] The data extraction module 3 is used to obtain historical navigation log data, and use the time series analysis method to obtain the periodicity and trend of navigation demand to obtain the navigation demand change curve, and use the event detection algorithm to identify the significant points of environmental change to obtain the environmental change curve.

[0151] The matching degree acquisition module 4 is used to combine the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and based on the comprehensive data set, the fusion similarity matrix and the label data, apply the prediction model to obtain the matching degree score of the obstacle.

[0152] The navigation path confirmation module 5 is used to take obstacles with a high matching degree as part of the navigation path, and plan a navigation path from the starting point to the end point, passing through all the selected obstacles with a high matching degree.

[0153] In summary, the present invention provides a camera-based explosion-proof intelligent visual navigation method and system, which can ensure the safe acquisition of images in flammable and explosive environments by using explosion-proof cameras, and reduce the risk of safety accidents. By obtaining a fused similarity matrix and using the fused similarity matrix for the training and application of the prediction model, the relative position and feature similarity between obstacles can be taken into account, which helps to improve the accuracy of the prediction matching score. Because similar obstacles have similar effects on the navigation system, the similarity matrix reveals the potential spatial and functional connections between obstacles, allowing the navigation system to better plan the path. The data of the fused similarity matrix enables the prediction model to make decisions not only based on the characteristics of a single obstacle, but also to integrate the interactions and correlations of multiple obstacles to provide a more comprehensive and more effective navigation solution. Through time series analysis, periodic and trend analysis of navigation needs is provided to help predict future navigation needs and optimize resource allocation. Through the event detection algorithm, environmental changes are quickly identified, so that the navigation system can adjust strategies in time and improve the response speed and accuracy to environmental changes. Continuously learning and adapting to changes in environmental changes and navigation needs, the prediction model can better adapt to complex and changeable operating environments, which is the key to efficient navigation and avoiding potential obstacles. At the same time, it integrates multi-source data and uses the random forest model to predict the matching degree of obstacles, providing more scientific and accurate decision support. Through continuous training and model adjustment, the navigation system can adapt to changes in the environment and improve the reliability and efficiency of long-term operation. Accurate matching scores make path planning more efficient and ensure that the most appropriate path is used to reach the destination.

[0154] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A camera-based explosion-proof intelligent visual navigation method, characterized in that: The camera-based explosion-proof intelligent visual navigation method comprises the following steps: S1. Obtain environmental images based on explosion-proof cameras, identify obstacles, and obtain characteristic data of obstacles; S2. Based on the feature data of the obstacles, similarity calculation of the overlapping positions of the obstacles and cosine similarity calculation between vectors are performed, and similarity fusion is performed to obtain a fused similarity matrix; S3. Obtain historical navigation log data, and use time series analysis to obtain the periodicity and trend of navigation demand, obtain a navigation demand change curve, and use an event detection algorithm to identify significant points of environmental change, obtain an environmental change curve; S4. Combining the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and applying the prediction model based on the comprehensive data set, the fusion similarity matrix and the label data to obtain the matching score of the obstacle; S5, taking obstacles with high matching degrees as part of the navigation path, and planning a navigation path from the starting point to the end point, passing through all selected obstacles with high matching degrees; The method of combining the statistical features of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and applying the prediction model to obtain the matching score of the obstacle based on the comprehensive data set, the fusion similarity matrix and the label data includes the following steps: S41, merging the mean and trend characteristic values ​​of the navigation demand change and the environment change to obtain a comprehensive data set; S42, integrating the fused similarity matrix into the comprehensive dataset and adding a corresponding similarity score for each obstacle pair; S43, determining a matching score standard for obstacles, and assigning a matching score to each obstacle in the comprehensive data set as a label; S44, using the comprehensive data set and the matching score to train the random forest model to obtain a prediction model; S45. Use the built-in variable importance evaluation function of random forest to obtain the influence of the feature on the prediction matching score of the prediction model; S46. Use cross-validation techniques to test the prediction model, and adjust the prediction model parameters and feature selection based on the prediction model performance and feature importance feedback; S47. Use the trained prediction model to predict the matching score of the new obstacle to obtain the matching score of the obstacle.

2. The camera-based explosion-proof intelligent visual navigation method according to claim 1, characterized in that: The method of calculating the similarity of overlapping positions of obstacles and the cosine similarity between vectors based on the feature data of obstacles, and performing similarity fusion to obtain a fused similarity matrix includes the following steps: S21. Based on the position and size information of the obstacle, represent each obstacle as a geometric shape; S22, obtaining the overlapping area between two obstacles and calculating the intersection area; S23, calculating the area of ​​the minimum geometric shape of all spaces of the two obstacles to obtain the union area; S24, the Jaccard index is calculated by the ratio of the intersection area to the union area; S25, constructing a vector with the characteristic data of the obstacle, and using the various dimension values ​​of the vector to perform multiplication and accumulation to obtain the dot product result of the obstacle, and obtaining the calculation modulus of the obstacle by calculating the square root of the sum of the squares of the components of the vector; S26. Divide the dot product of the two vectors by the modulus length of each vector to obtain the cosine similarity; S27. For each pair of obstacles, the Jaccard index and the cosine similarity score are fused and a fused similarity matrix is ​​constructed.

3. The camera-based explosion-proof intelligent visual navigation method according to claim 2 is characterized in that: The method of obtaining historical navigation log data, obtaining the periodicity and trend of navigation demand using a time series analysis method, obtaining a navigation demand change curve, and identifying significant points of environmental change using an event detection algorithm to obtain an environmental change curve comprises the following steps: S31, collecting historical navigation log data, and cleaning the historical navigation log data; S32. Apply time series analysis methods to identify periodic patterns in data; S33. Use trend analysis techniques to identify long-term trends; S34. Use a time series graph to display the navigation demand change curve; S35. Obtain significant points of environmental change based on the Bayesian model and environmental monitoring data to obtain an environmental change curve.

4. The camera-based explosion-proof intelligent visual navigation method according to claim 3 is characterized in that: The method of obtaining significant points of environmental changes based on the Bayesian model and environmental monitoring data to obtain an environmental change curve comprises the following steps: S351. Obtain environmental monitoring data, and perform data cleaning and data coding; S352, defining a Gaussian kernel function and setting a bandwidth parameter for calculating the similarity between two data points; S353, using the Gaussian kernel function to iteratively calculate the data of all environmental monitoring points to obtain a similarity matrix; S354, establishing a Bayesian model based on prior knowledge and newly acquired data, and dynamically updating the similarity matrix according to the new data and the results of Bayesian inference; S355, performing spectral clustering analysis on the updated similarity matrix to obtain data grouping results; S356: Use the grouping results to identify significant points of environmental change, and based on the identified significant points and related environmental events, draw an environmental change curve representing the frequency and pattern of obstacle changes.

5. The camera-based explosion-proof intelligent visual navigation method according to claim 4 is characterized in that: The formula of the Gaussian kernel function is: ； In the formula, K ( x , y ) represents two data points x and y The similarity value between them; σ Indicates the bandwidth parameter.

6. The camera-based explosion-proof intelligent visual navigation method according to claim 5, characterized in that: The performing spectral clustering analysis on the updated similarity matrix to obtain the data grouping result comprises the following steps: S3551. constructing an adjacency matrix of the undirected weighted graph based on the updated similarity matrix, and constructing a corresponding graph Laplacian matrix from the adjacency matrix, and performing standardization; S3552. Calculate the eigenvectors of the normalized graph Laplacian matrix to identify the structure and grouping of data.

7. The camera-based explosion-proof intelligent visual navigation method according to claim 1, characterized in that: The method of using the built-in variable importance evaluation function of the random forest to obtain the influence of the feature on the prediction matching score of the prediction model includes the following steps: S451. For each decision tree, use samples that have not been trained by the decision tree to evaluate the performance of the prediction model; S452, randomly permuting the value of each feature, and obtaining the effect of the permutation on the out-of-bag error of the prediction model, and obtaining the importance score of each feature by comparing the out-of-bag errors before and after the permutation; S453, averaging the feature importance scores calculated by all decision trees to obtain an average importance score for each feature; S454. Sort the features according to the obtained average importance scores.

8. The camera-based explosion-proof intelligent visual navigation method according to claim 7, characterized in that: When determining the obstacle matching score standard, the obstacle matching score calculation formula is: ； In the formula, P Indicates the navigation success rate of obstacles; T represents the average delay time caused by obstacles; w 1 and w 2 represents the weight of the navigation success rate of obstacles and the average delay time caused by obstacles.

9. An explosion-proof intelligent visual navigation system based on a camera, used to implement the explosion-proof intelligent visual navigation method based on a camera according to any one of claims 1 to 8, characterized in that: The camera-based explosion-proof intelligent visual navigation system includes a feature data acquisition module, a similarity fusion module, a data extraction module, a matching acquisition module and a navigation path confirmation module; The feature data acquisition module is used to acquire environmental images based on an explosion-proof camera, identify obstacles, and acquire feature data of the obstacles; The similarity fusion module calculates the similarity of the overlapping positions of the obstacles and the cosine similarity between the vectors based on the feature data of the obstacles, and performs similarity fusion to obtain a fused similarity matrix; The data extraction module is used to obtain historical navigation log data, and use the time series analysis method to obtain the periodicity and trend of navigation requirements to obtain a navigation requirement change curve, and use the event detection algorithm to identify significant points of environmental changes to obtain an environmental change curve; The matching degree acquisition module is used to combine the statistical characteristics of the navigation demand change curve and the environment change curve to obtain a comprehensive data set, and based on the comprehensive data set, the fusion similarity matrix and the label data, apply the prediction model to obtain the matching degree score of the obstacle; The navigation path confirmation module is used to take obstacles with a high matching degree as part of the navigation path, and plan a navigation path from the starting point to the end point, passing through all selected obstacles with a high matching degree.

Images