Method, System, Device and Storage Medium for Optimizing Dash Cam Images

Abstract

The present invention belongs to the technical field of image processing, and specifically provides a method, system, device and storage medium for optimizing images of a driving recorder; the method includes the following steps: obtaining image data and environmental parameter data of a target scene in real time through a driving recorder and an acquisition module deployed on a mining vehicle; outputting the current environmental state type through an environmental state classification model based on the environmental parameter data; calculating its weight score through a preset severity scoring model according to the environmental parameter deviation degree of each sub-state in the environmental state type, and determining the priority order of the image optimization algorithm according to the scoring result; sequentially calling the image optimization algorithms associated with each sub-state according to the priority order to process the image data and generate optimized image data. Through this solution, the problems of poor image quality and monitoring failure of a mining driving recorder in a harsh environment are solved, and the visibility of the image data of key driving information is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and particularly relates to a method, system, device and storage medium for optimizing the images of a driving recorder. Background Art

[0002] The mining driving recorder is a vehicle-mounted recording device specially designed for special working environments such as mines and mining areas, and is mainly used to monitor and record the driving status, operation process and surrounding environment of mining vehicles (such as transport vehicles, forklifts, mining trucks, etc.) to improve the level of safety production management and the ability of accident traceability.

[0003] The mining environment is usually relatively harsh, such as insufficient light, a lot of dust, high humidity, and there may also be frequent vibrations, etc.; these environmental factors will affect the image quality captured by the driving recorder, resulting in the driving recorder being unable to perform effective monitoring and recording, and thus may lead to the occurrence of safety accidents.

[0004] To avoid the occurrence of the above accidents, the present invention provides a method, system, device and storage medium for optimizing the images of a driving recorder to solve the above problems. Summary of the Invention

[0005] In order to overcome the deficiencies of the prior art, the present invention provides a method, system, device and storage medium for optimizing the images of a driving recorder to solve the problems in the prior art.

[0006] One embodiment of the present invention provides a method for optimizing the images of a driving recorder, including the following steps:

[0007] Real-time obtain the image data and environmental parameter data of the target scene through a driving recorder and a collection module deployed on a mining vehicle;

[0008] Based on the environmental parameter data, output the current environmental state type through an environmental state classification model, and the environmental state type includes at least one or a combination of a low-light state, a high-dust state, a high-frequency vibration state, and a high-humidity state;

[0009] According to the environmental parameter deviation degrees of each sub-state in the environmental state type, calculate its weight score through a preset severity scoring model, and determine the priority order of the image optimization algorithm according to the scoring result; wherein, the severity scoring model satisfies the following conditions:

[0010] Allocate dynamic weight coefficients for different sub-state types, and the weight coefficients are positively correlated with the degree of negative impact of the sub-state on the image quality;

[0011] Calculate a score based on the deviation degree of each sub - state parameter value from the preset threshold. The greater the deviation degree, the higher the score. When multiple sub - states co - exist, preferentially execute the image optimization algorithm associated with the sub - state with the highest score;

[0012] Call the image optimization algorithms associated with each sub - state in the order of priority to process the image data and generate the optimized image data.

[0013] In one embodiment, in the step of outputting the current environmental state type through the environmental state classification model based on the environmental parameter data, the environmental state classification model determines the environmental state type in the following way:

[0014] Compare the environmental parameter data with the preset thresholds corresponding to each sub - state in the environmental state type;

[0015] Based on the comparison result, combined with whether the environmental parameter value exceeds the preset threshold and its deviation degree, determine the current environmental state type.

[0016] In one embodiment, the preset threshold is dynamically determined by at least one of the following methods:

[0017] Match the real - time vehicle position with the preset mine area map, and load the regional environmental reference value associated with the current position as the environmental parameter threshold. The preset mine area map includes the boundary coordinates of each area, the environmental parameter reference value, and the risk level;

[0018] Dynamically calculate the current environmental parameter threshold according to the statistical result of the environmental parameter data within the historical time window.

[0019] In one embodiment, in the step of matching the real - time vehicle position with the preset mine area map and loading the regional environmental reference value associated with the current position as the environmental parameter threshold, the determination method of the regional reference value specifically includes the following steps;

[0020] Collect multi - dimensional input features, where the multi - dimensional input features at least include the mine operation plan, equipment operation status, and real - time meteorological data;

[0021] Input the multi - dimensional input features into the pre - trained time - series prediction model, and the time - series prediction model outputs the prediction result of the environmental parameter reference value within a preset time period;

[0022] Update the reference value of the corresponding area in the mine area map according to the prediction result.

[0023] In one embodiment, in the step of calculating the weight score through the preset severity scoring model according to the environmental parameter deviation degree of each sub - state in the environmental state type and determining the priority order of the image optimization algorithm according to the scoring result, the following steps are also included:

[0024] Receive the real-time vehicle scheduling data sent by the mine car scheduling system, where the data includes real-time positioning coordinates and the current operation stage identifier;

[0025] Based on the vehicle scheduling data and the risk levels of each area in the mining area map, identify the risk level of the area where the vehicle is currently located, and the risk levels include high risk, medium risk, and low risk;

[0026] According to the risk level and the current operation stage identifier, obtain the sub-state type associated with the high-risk area through a preset mapping table, and increase the weight coefficient by a preset increment;

[0027] When the sub-state score after the weight coefficient adjustment exceeds the preset threshold, trigger the preloading instruction of the corresponding image optimization algorithm.

[0028] In one embodiment, in the step of calculating the weight score according to the deviation degree of the environmental parameters of each sub-state in the environmental state type through a preset severity scoring model, and determining the priority order of the image optimization algorithm according to the scoring result, the construction of the preset severity scoring model includes the following steps:

[0029] Obtain a sample data set, where the data set contains historical environmental parameter data, corresponding image quality evaluation data, and annotations of the influence level of each sub-state type on image quality;

[0030] Based on preset rules, assign initial weight coefficients to each sub-state type, and the initial weight coefficients are positively correlated with the predefined negative impact degree of the sub-state on image quality;

[0031] Construct a severity scoring model, use the environmental parameter deviation degree, initial weight coefficient, and image quality evaluation data as input features, and train the severity scoring model using a machine learning algorithm to obtain a trained severity scoring model for receiving the deviation degree of the environmental parameters of each sub-state in real time and outputting the weight score of each sub-state;

[0032] Use the test set to verify the performance of the severity scoring model, and optimize the parameters of the severity scoring model according to the verification results.

[0033] In one embodiment, when performing the step of assigning initial weight coefficients to each sub-state type based on preset rules, it includes the following steps:

[0034] Define multiple influence parameters for each sub-state type, and the influence parameters at least include image quality influence parameters, safety risk parameters, and data reliability parameters;

[0035] Based on the influence parameters, an initial weight coefficient is generated through weighted calculation, and the weight distribution rule of the weighted calculation is dynamically updated according to real-time data or environmental changes;

[0036] When it is detected that the data reliability parameter is lower than the preset threshold, the automatic recalibration of the weight coefficient is triggered.

[0037] This application also relates to a driving recorder image optimization system, including:

[0038] A data acquisition module, configured to obtain image data and environmental parameter data of a target scene in real time through a driving recorder and an acquisition module deployed on a mining vehicle;

[0039] An environment determination module, configured to output the current environmental state type based on the environmental parameter data through an environmental state classification model, where the environmental state type includes at least one or a combination of a low-light state, a high-dust state, a high-frequency vibration state, and a high-humidity state;

[0040] A calculation module, configured to calculate its weight score through a preset severity scoring model according to the environmental parameter deviation degree of each sub-state in the environmental state type, and determine the priority order of the image optimization algorithm according to the scoring result;

[0041] An image optimization module, configured to sequentially call image optimization algorithms associated with each sub-state to process the image data according to the priority order, and generate optimized image data.

[0042] This application also relates to a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned driving recorder image optimization method are implemented.

[0043] This application also relates to a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned driving recorder image optimization method are implemented.

[0044] The above-mentioned driving recorder image optimization method, system, device, and storage medium provided by the above embodiments have the following beneficial effects:

[0045] 1. By collecting multi-source environmental data such as dust, light, vibration, and humidity in real time, and based on a severity scoring model of parameter deviation and dynamic weight, under the factor of intelligently determining the current environmental state, the optimal image optimization algorithm selected for image optimization is realized for targeted optimization, breaking through the misjudgment limitation caused by traditional solutions relying on manual experience or a single threshold. Call the targeted optimization algorithm in the order of priority to avoid problems such as image detail loss and artifact diffusion caused by the superposition of multiple environmental factors, improve the visibility of image data of key driving information (such as obstacles and roadway signs), and solve the problems of poor image quality and monitoring failure of mine driving recorders in harsh environments.

[0046] 2. By dynamically comparing the environmental parameter data with the preset thresholds corresponding to each sub-state and combining deviation quantification evaluation, accurate classification of the environmental state is realized. For example, when the dust concentration exceeds the threshold but the deviation is low, misjudgment as a high-dust state is avoided, reducing the triggering of ineffective algorithms; at the same time, when the parameter value is close to the threshold, the determination sensitivity is enhanced through deviation weighting to prevent missed judgment of real abnormal states. Solve the problems of misjudgment and missed judgment caused by traditional environmental state classification relying on fixed thresholds, improve the accuracy of environmental state classification, and provide reliable input for subsequent priority decision-making.

[0047] 3. Based on the vehicle's real-time position, load the regional reference value, or dynamically calculate the threshold according to historical data to achieve scene adaptation of the environmental parameter threshold. For example, a higher dust threshold is adopted in the mining face area to avoid misjudgment of normal dust, and a dynamic statistical threshold is adopted in the transportation roadway to adapt to the change of vehicle flow. Solve the problem of threshold adaptation caused by environmental differences in multiple regions of the mining area, reduce the misjudgment rate and improve the resource utilization rate in different regions, especially suitable for mine scenes with complex geological conditions and variable operation stages.

[0048] 4. By receiving mine car scheduling data, dynamically adjusting the weight coefficient in combination with the risk level and operation stage identifier, and preloading high-priority algorithms, accurate allocation of computing resources is realized, solving the problem of optimization delay caused by resource competition in multiple interference scenarios. At the same time, improve the execution efficiency of key algorithms and ensure real-time monitoring capabilities in extreme scenarios such as sharp increase in dust and intensified vibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on the structures shown in these drawings.

[0050] Figure 1A flowchart of an image optimization method for a driving recorder provided by an embodiment of the present invention;

[0051] Figure 2 An implementation flowchart of step S200 of an image optimization method for a driving recorder provided by an embodiment of the present invention

[0052] Figure 3 An implementation flowchart of step S300 of an image optimization method for a driving recorder provided by an embodiment of the present invention;

[0053] Figure 4 An implementation flowchart of step S300 of an image optimization method for a driving recorder provided by an embodiment of the present invention;

[0054] Figure 5 A principle block diagram of a computer device provided by an embodiment of the present invention. Detailed implementation manners

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with 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 the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0056] It should be noted that if there are directional indications (such as up, down, left, right, front, back...) involved in the embodiments of the present invention, the directional indications are only used to explain the relative positional relationship and movement conditions between components in a specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0057] In addition, if there are descriptions such as "first" and "second" involved in the embodiments of the present invention, the descriptions of "first" and "second" are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, if "and / or" or "and / or" appears throughout the text, its meaning includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or a solution where A and B are satisfied simultaneously. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the protection scope required by the present invention.

[0058] Refer to Figure 1, one embodiment of the present invention provides a method for optimizing the images of a driving recorder, including the following steps:

[0059] S100. Real-time obtain the image data and environmental parameter data of the target scene through the driving recorder and the acquisition module deployed on the mining vehicle;

[0060] S200. Based on the environmental parameter data, output the current environmental state type through the environmental state classification model, and the environmental state type includes at least one or a combination of low light state, high dust state, high frequency vibration state, and high humidity state;

[0061] S300. According to the environmental parameter deviation degrees of the sub-states in the environmental state type, calculate their weight scores through a preset severity scoring model, and determine the priority order of the image optimization algorithms according to the scoring results; wherein, the severity scoring model satisfies the following conditions:

[0062] Allocate dynamic weight coefficients for different sub-state types, and the weight coefficients are positively correlated with the degree of negative impact of the sub-states on the image quality;

[0063] Calculate the score according to the deviation degree of each sub-state parameter value from the preset threshold. The greater the deviation degree, the higher the score. When multiple sub-states coexist, preferentially execute the image optimization algorithm associated with the sub-state with the highest score;

[0064] S400. Sequentially call the image optimization algorithms associated with each sub-state according to the priority order to process the image data and generate the optimized image data.

[0065] As described in step S100 above, through the driving recorder (such as a multi-spectral camera) deployed on the mining vehicle and its integrated multi-type sensors (such as a light sensor, a dust concentration detector, a three-axis accelerometer, a humidity sensor, etc.), synchronously collect the real-time image data and environmental parameters of the vehicle operation scene. The environmental parameters include but are not limited to light intensity, air dust concentration, vehicle vibration frequency, and environmental humidity data. During the acquisition process, through the timestamp synchronization technology, ensure the temporal and spatial consistency of the image frames and the corresponding environmental parameters, providing a data basis for subsequent analysis with spatio-temporal correlation.

[0066] As described in step S200 above, the collected environmental parameter data is preprocessed (such as normalized) and then input into a pre-trained environmental state classification model. This model is constructed based on machine learning algorithms (such as an environmental state classification model based on the random forest algorithm), and can identify the current environmental state type according to the characteristic combinations of environmental parameter data, including single states or composite states (for example, there is a high-frequency vibration state or there are both low light and high dust states at the same time). For example: if the dust concentration > 200 μg / m³ and the light intensity < 10 Lux → it is determined as a high dust + low light combined state; if the vibration frequency > 30 Hz and the humidity > 80%RH → it is determined as a high-frequency vibration + high humidity combined state. The classification results are output in the form of state labels, which are used to guide the formulation of subsequent optimization strategies. For example, in the transportation roadway, the dust concentration is 300 μg / m³ and the light intensity is 5 Lux → it is classified as a high dust + low light state; in the mining face, the vibration frequency is 45 Hz and the humidity is 85%RH → it is classified as a high-frequency vibration + high humidity state.

[0067] As described in step S300 above, for each sub-state type obtained by classification, calculate the degree of deviation between its environmental parameters and the preset threshold (for the preset threshold, it is dynamically set according to the specific mine environment in practice and is not limited here). For the deviation calculation: deviation degree = (real-time value - preset threshold) / preset threshold × 100%; for example, the real-time value of the dust concentration is 280 μg / m³ and the preset threshold is 200 μg / m³ → deviation degree = 40%.

[0068] Then, the severity scoring model assigns dynamic weight coefficients to each sub-state according to the preset dynamic weight rules (for example, the weight coefficient of the high dust state = 0.7, the weight coefficient of the low light state = 0.6, the high-frequency vibration state = 0.5, the high humidity state = 0.3). Then, combined with the deviation degree values of each sub-state type and the influence degree of each sub-state on the image quality, a priority score is generated. The scoring rule is satisfied: the sub-state that has a greater impact on the image quality (such as high dust obscuring the key area) has a higher weight; when multiple sub-states coexist, the system automatically selects the sub-state with the highest score as the primary optimization target. For example, the score of the high dust state = 0.7 (weight coefficient) × 40% (deviation degree) = 28 points; the score of the low light state = 0.6 (weight coefficient) × 80% (deviation degree) = 48 points. Therefore, the system will give priority to executing the low light enhancement algorithm.

[0069] Among them, for the definition of the influence degree of each sub-state on the image quality:

[0070] High dust state: It directly obscures the content of the picture, may completely lose key information (such as obstacles), and has the greatest impact.

[0071] Low light state: It causes the overall picture to be dim, but some details may be retained, and the impact degree is the second.

[0072] High-frequency vibration: It causes image blurring, but if the object is stationary, it can still be recognized, and the impact degree is relatively low.

[0073] When calculating the priority of each image optimization algorithm, it is not just about looking at the magnitude of the deviation value. Instead, it is necessary to calculate the degree of environmental abnormality through the deviation and superimpose the influence ability of the environmental type itself through the impact degree.

[0074] Through this design, misjudgments of "only looking at the magnitude of the value" can be avoided. For example, the high-frequency vibration parameter may have a very large deviation value due to vehicle bumps, but its destructiveness to the image may be lower than that of slight dust. And it meets the actual safety requirements: prioritize solving the problems that have the greatest impact on monitoring and accident tracing (such as dust obscuring key obstacles), rather than simply dealing with numerical anomalies.

[0075] As described in step S400 above, an image optimization algorithm library is pre-constructed, which contains a variety of different image optimization algorithms to correspond to image optimization in different environmental states. For example, in the low-light state, it corresponds to a low-light enhancement algorithm such as the Retinex algorithm, which focuses on global illumination compensation; in the high-dust state, it corresponds to a dust noise reduction algorithm such as non-local means (NL-Means) filtering and a defogging processing algorithm, etc.; in the high-frequency vibration state, it corresponds to a vibration compensation algorithm such as optical flow-based electronic image stabilization (EIS); in the high-humidity state, it corresponds to a humidity correction algorithm such as a non-uniform humidity correction algorithm. Among them, for the specific algorithm type, it can be adjusted according to the actual mine environment, and it is not limited to the algorithm types described above. The above algorithm types are only examples for the understanding of those skilled in the art. After determining the priority order of the image optimization algorithms according to the severity scoring model, a resource scheduling queue containing multiple image optimization algorithms is established, and the optimization algorithms matching the sub-states are called in turn according to the priority score from high to low. For the composite environmental state, a parallel processing architecture is adopted to prioritize the allocation of computing resources for high-priority algorithms, and at the same time, through algorithm cascading or data fusion technology, the collaborative improvement of multi-dimensional image quality is achieved.

[0076] In this embodiment, by real-time collecting multi-source environmental data such as dust, light, vibration, and humidity, and based on the severity scoring model of parameter deviation and dynamic weight, under the factors of the current environmental state, the optimal image optimization algorithm selected for image optimization is intelligently determined to achieve targeted optimization, breaking through the misjudgment limitations caused by traditional solutions relying on manual experience or a single threshold. The targeted optimization algorithms are called in priority order to avoid problems such as image detail loss and artifact diffusion caused by the superposition of multiple environmental factors, improving the visibility of image data of key driving information (such as obstacles and roadway signs), and solving the problems of poor image quality and monitoring failure of mine-used driving recorders in harsh environments.

[0077] In this embodiment, it is assumed that in a roadway of a mining area, a mining truck encounters various harsh environments during transportation: the roadway lighting is insufficient, resulting in a dim image; the dust raised by the vehicle in front forms a fog-like obstruction; and at the same time, the bumpy road surface causes the driving recorder to shake continuously. The system identifies the composite state of "low light + high dust + high-frequency vibration" through the environmental state classification model, and calculates the priority scores of each sub-state through the severity scoring model:

[0078] High dust state: Since the dust seriously obscures the contour of the road ahead, the score is the highest, triggering the real-time defogging algorithm, and restoring the image of the obscured obstacle by analyzing the optical scattering characteristics of dust particles;

[0079] Low light state: Based on the image after defogging processing, start the adaptive brightness enhancement algorithm to focus on improving the visibility of details at the roadway edge and the cargo area;

[0080] High-frequency vibration state: Combine the vehicle motion data and adopt the dynamic image stabilization algorithm to eliminate the image jitter, ensuring that the clarity of the key frame image meets the requirements for accident traceability.

[0081] After multi-level optimization, the originally blurred and distorted image is converted into a monitoring image that can clearly identify the roadway structure, vehicle position, and road surface foreign objects. By dynamically adjusting the processing order, the system preferentially solves the dust occlusion problem that has the greatest impact on safety monitoring under the constraint of limited computing power, significantly improving the usability of the driving record image in complex environments.

[0082] Refer to Figure 2 , in one of the embodiments, in step S200, the environmental state classification model determines the environmental state type in the following manner:

[0083] S210. Compare the environmental parameter data with the preset thresholds corresponding to each sub-state in the environmental state type;

[0084] S220. According to the comparison result, combined with whether the environmental parameter value exceeds the preset threshold and its deviation degree, determine the current environmental state type.

[0085] As described in the above steps, preset determination thresholds are set for various environmental sub-states:

[0086] Low light state: Define the lowest safety threshold for light intensity. For example, if the light intensity threshold ≤ 10 Lux, it is used to determine whether it reaches the dark light environment that affects image visibility;

[0087] High dust state: Set the critical value of dust concentration. For example, if the dust concentration threshold ≥ 200 μg / m³, it is used to distinguish normal dust emission from dust pollution that seriously obscures the line of sight;

[0088] High-frequency vibration state: Determine the threshold range of acceleration fluctuation. For example, if the vibration frequency threshold ≥ 30 Hz, identify abnormal vibrations that may cause image blurring;

[0089] High-humidity state: Define the humidity saturation threshold. For example, if the humidity threshold ≥ 80%RH, judge the risk level of lens fogging.

[0090] For the preset thresholds, they can be dynamically adjusted according to the specific mine environment in practice and are not limited to the specific thresholds described above. The above specific thresholds are only examples for the understanding of those skilled in the art.

[0091] By comparing the real-time collected environmental parameters with the above thresholds item by item, generate a preliminary state trigger signal. For example, when the reading of the light sensor continuously falls below the low-light threshold, mark the low-light state as a sub-state to be confirmed.

[0092] Perform secondary verification and deviation degree fusion analysis based on the comparison results:

[0093] Threshold exceedance verification: Check whether additional conditions such as the duration of parameter overrun and fluctuation stability are met to avoid misjudgment caused by instantaneous interference (e.g., a brief dust raise does not trigger the high-dust state);

[0094] Deviation degree weighted calculation: For parameters exceeding the threshold, calculate the percentage amplitude of their deviation from the threshold reference value to quantify the degree of environmental abnormality;

[0095] Composite state determination: When multiple parameters exceed the threshold simultaneously, generate a composite type label according to the deviation degree ratio combination of each sub-state. For example, if the deviation degree of dust concentration reaches 70% and the deviation degree of light reaches 50%, then determine it as a "high-dust + low-light" composite state.

[0096] In this embodiment, by dynamically comparing the environmental parameter data with the preset thresholds corresponding to each sub-state and combining the deviation degree quantitative evaluation, the accurate classification of the environmental state is realized. For example, when the dust concentration exceeds the threshold but the deviation degree is low, avoid misjudging it as a high-dust state and reduce the triggering of ineffective algorithms; at the same time, when the parameter value is close to the threshold, improve the determination sensitivity through deviation degree weighting to prevent missing real abnormal states. Solve the problems of misjudgment and missing judgment caused by the traditional environmental state classification relying on fixed thresholds, improve the accuracy of environmental state classification, and provide reliable input for subsequent priority decision-making.

[0097] In this embodiment, it is assumed that after a heavy rain in an open-pit mining area, a mining truck successively passes through two typical areas:

[0098] (1) Waterlogged pothole area:

[0099] Environmental parameters: Humidity continuously higher than the threshold by 20%, vibration frequency exceeding the safe range;

[0100] System determination: The calculated humidity deviation is medium (30%), and the vibration deviation is high (65%). However, since the humidity does not reach the fogging critical condition, it is finally determined as a single "high-frequency vibration" state.

[0101] (2) Semi-closed loading area:

[0102] Environmental parameters: The light intensity is 50% below the threshold, and the dust concentration exceeds the standard and continues to rise;

[0103] System determination: The light deviation reaches 80% (severely insufficient), and the dust deviation reaches 60% (rapid accumulation), triggering a "low light + high dust" composite state. The system automatically increases the image processing frame rate to prioritize the coordinated execution of dust filtration and local light compensation algorithms.

[0104] Through a dual verification mechanism, the system successfully avoids misjudging the short-term splashing water humidity fluctuation as a high humidity state, and at the same time accurately identifies the complex harsh working conditions in the loading area, providing a reliable classification basis for subsequent optimization.

[0105] In one of the embodiments, in step S210, the preset threshold is dynamically determined by at least one of the following methods:

[0106] S211. Match the preset mining area map according to the real-time position of the vehicle, and load the regional environmental reference value associated with the current position as the environmental parameter threshold. The preset mining area map includes the boundary coordinates of each area, the environmental parameter reference value, and the risk level;

[0107] S212 Dynamically calculate the current environmental parameter threshold according to the statistical results of the environmental parameter data within the historical time window.

[0108] As described in step S211 above, the dynamic setting of the threshold is achieved by the following method:

[0109] (1) Spatial association mapping:

[0110] Pre-store a digitalized mining area environmental map. Different functional areas (such as mining faces, transportation roadways, loading and unloading areas, etc.) are divided in the map, and historical environmental reference values (such as typical light ranges, dust concentration baselines, vibration intensity means) are marked for each area;

[0111] When the vehicle enters a specific area identified by the positioning module (such as Beidou / GPS + inertial navigation), the corresponding environmental parameter reference value of the area is automatically called as the dynamic threshold;

[0112] For example: A lower dust concentration threshold is adopted for the transportation roadway section (because dust is likely to accumulate in the enclosed space), while a higher dust threshold is set for the open-pit mining area to match the characteristics of the open environment.

[0113] Risk level adaptation:

[0114] According to the regional risk levels marked on the map (such as high-risk collapse areas, conventional operation areas), a correction factor is applied to the reference value. In high-risk areas, the sensitivity thresholds of dust / vibration are automatically reduced to achieve more stringent environmental anomaly monitoring.

[0115] As described in step S212 above, the dynamic calculation of the threshold is achieved in the following manner:

[0116] (1) Sliding time window statistics:

[0117] Continuously collect environmental parameter data within a time window of a preset duration (such as the previous operation cycle);

[0118] Perform statistical calculations on the window data (such as moving average, standard deviation analysis, percentile statistics) to generate a reference value reflecting the current environmental background;

[0119] For example: Set a dynamic threshold based on the 90th percentile value of the dust concentration in the previous time period to avoid interference from occasional peaks.

[0120] (2) Trend adaptive adjustment:

[0121] When it is detected that the environmental parameters continuously change unidirectionally (such as the humidity linearly rising over time), an exponentially weighted algorithm is used to update the threshold, enabling the threshold to automatically drift with the evolution of the environment;

[0122] The dynamic threshold remains synchronized with the real-time environmental background to prevent misjudgment or missed judgment caused by the drift of the environmental baseline.

[0123] In this embodiment, the regional reference value is loaded based on the real-time position of the vehicle, or the threshold is dynamically calculated according to historical data to achieve scenario adaptability of the environmental parameter threshold. For example, a higher dust threshold is adopted in the mining face area to avoid misjudgment of normal dust, and a dynamic statistical threshold is used in the transportation roadway to adapt to the change of vehicle flow. It solves the problem of threshold adaptation caused by environmental differences in multiple regions of the mining area, can reduce the misjudgment rate and improve resource utilization in different regions, and is especially suitable for mine scenarios with complex geological conditions and variable operation stages.

[0124] In one of the embodiments, in step S211, the method for determining the regional reference value specifically includes the following steps;

[0125] S211a. Collect multi-dimensional input features, where the multi-dimensional input features at least include mine operation plans, equipment operating status, and real-time meteorological data;

[0126] S211b. Input the multi-dimensional input features into a pre-trained time series prediction model, and the time series prediction model outputs a prediction result of the environmental parameter reference value within a preset time period;

[0127] S211c. Update the baseline value of the corresponding area in the mining area map according to the prediction result.

[0128] As described in step S211a above, collect multi-dimensional input features in the following ways:

[0129] (1) Mine operation plan features:

[0130] Obtain the planned mining volume, the number of deployed operation machines, and the work schedule of the current mine area from the production management system in real time;

[0131] Extract operation parameters related to dust generation and equipment vibration (such as blasting time period, transportation vehicle frequency).

[0132] (2) Equipment operation status features:

[0133] Collect the operation data of surrounding operation equipment through in-vehicle Internet of Things terminals, including but not limited to engine load rate, hydraulic system pressure, and vehicle moving speed;

[0134] Analyze the impact of the equipment cluster operation mode on the local environment (such as the simultaneous operation of multiple devices intensifying dust diffusion).

[0135] (3) Real-time meteorological data features:

[0136] Access the data stream of the mining area meteorological station to obtain wind speed, wind direction, temperature, humidity, and precipitation probability;

[0137] Combine the microclimate model to predict the impact of air flow in the roadway on the dust settlement speed and humidity distribution.

[0138] As described in step S211b above, the time series prediction model operates through the following mechanisms:

[0139] (1) Model architecture:

[0140] Adopt a multi-variable time series prediction model based on deep learning. Its input layer receives the standardized multi-dimensional feature matrix, and the output layer generates the baseline value of environmental parameters for a preset future time period (such as the next operation cycle);

[0141] Inside the model, the attention mechanism is used to dynamically weight the influence of different features. For example, in a dry climate, the prediction weight of wind speed on the dust baseline value is strengthened.

[0142] (2) Prediction logic:

[0143] Encode the mine operation plan features into a periodic event sequence, convert the equipment status features into a spatial distribution heat map, and construct a time decay factor for meteorological data;

[0144] Calculate the trend of the reference values of various environmental parameters (such as dust concentration, vibration intensity) through the feature fusion layer, and output the quantitative prediction results.

[0145] As described in step S211c above, the reference value update mechanism includes:

[0146] (1) Dynamic mapping update:

[0147] Associate the prediction results with the spatial grid coordinates of the mining area map, and refresh the reference values of environmental parameters grid by grid according to the regional coverage;

[0148] Set a shorter reference value update period for high-risk areas (such as poorly ventilated roadways) to ensure the timeliness of prediction.

[0149] (2) Feedback verification:

[0150] Perform residual analysis on the actually collected environmental parameters and the predicted reference values, and trigger model retraining when the deviation continuously exceeds the tolerance range;

[0151] Retain the latest environmental data through the sliding window mechanism, and continuously optimize the adaptability of the prediction model to complex working conditions.

[0152] In this embodiment, by fusing multi-dimensional features such as mine operation plans, equipment status, and meteorological data, and using a time series prediction model to prospectively adjust the reference values, the real-time performance and prediction accuracy of threshold setting are significantly improved. This solution reduces the prediction error of the reference values and improves the response speed to abnormal events, effectively coping with the sudden changes in the mine environment and solving the problem that the traditional reference value update lags behind the environmental mutations.

[0153] Refer to Figure 3 , in one of the embodiments, in step S300, the following steps are further included:

[0154] S301. Receive the real-time vehicle scheduling data sent by the mine car scheduling system, and the data includes real-time positioning coordinates and the current operation stage identifier;

[0155] S302. Based on the vehicle scheduling data and the risk levels of each area in the mining area map, identify the risk level of the area where the vehicle is currently located, and the risk levels include high risk, medium risk, and low risk;

[0156] S303. According to the risk level and the current operation stage identifier, obtain the sub-state type associated with the high-risk area through a preset mapping table, and increase the weight coefficient by a preset increment;

[0157] S304. When the sub-state score after the weight coefficient adjustment exceeds the preset threshold, trigger the preloading instruction of the corresponding image optimization algorithm.

[0158] As described in step S301 above, establish a data link with the central dispatching system in the mining area through the vehicle-mounted communication module, and receive scheduling data packets containing the following content in real time:

[0159] Real-time positioning coordinates: Based on the fusion technology of the Beidou positioning system and UWB indoor positioning, obtain the real-time position of the vehicle in the mining area map;

[0160] Current operation stage identifier: Parse the operation stage code from the scheduling instruction (such as "empty vehicle driving", "heavy load transportation", "fixed-point loading and unloading").

[0161] As described in step S302 above, the risk level identification mechanism includes:

[0162] Spatial grid matching: Map the vehicle coordinates to the encrypted grid coordinate system of the mining area map, and query the preset risk level label of the grid where it is located;

[0163] Dynamic risk correction: When the vehicle is in the "heavy load transportation" stage, automatically raise the risk level of the roadway turning area and the steep slope section by one level (for example, the original medium risk is raised to high risk).

[0164] As described in step S303 above, the weight adjustment rules are as follows:

[0165] High-risk area: Increase the weight increment for the "high dust" and "high-frequency vibration" sub-states (for example, the basic weight is increased from 0.4 to 0.6);

[0166] Medium-risk area: Only increase the weight for the sub-states strongly related to the current operation stage (for example, give priority to increasing the weight of "low light" during the loading and unloading stage);

[0167] Low-risk area: Maintain the basic weight coefficient.

[0168] As described in step S304, the preloading trigger logic is:

[0169] When the sub-state score exceeds the preloading threshold after weight adjustment, send an algorithm preloading instruction to the image processing unit;

[0170] The preloading process includes: allocating GPU computing resources, loading the neural network model weight file, and initializing the algorithm parameter buffer.

[0171] In this embodiment, by receiving the ore truck scheduling data, dynamically adjusting the weight coefficient in combination with the risk level and the operation stage identifier, and preloading the high-priority algorithm, the precise allocation of computing resources is achieved, solving the optimization delay problem caused by resource competition in multiple interference scenarios. This solution improves the execution efficiency of key algorithms and ensures the real-time monitoring ability in extreme scenarios such as a sharp increase in dust and intensified vibration.

[0172] In this embodiment, assume a scenario of transportation operations in a high-risk area:

[0173] A certain mining truck enters the transportation roadway of the fault zone marked as "high-risk" according to the dispatching instructions. The dispatching data received by the system includes:

[0174] Position coordinates: Matched to the "West Area Fault Zone - High-risk" grid in the map;

[0175] Operation phase identifier: "Heavy-load transportation".

[0176] The system performs the following processing:

[0177] Identify the risk level as high-risk, and determine the weight coefficients of "high dust" and "high-frequency vibration" to be strengthened through the mapping table;

[0178] Calculate the adjusted sub-state score: The score of "high dust" rises from 32 points to 48 points (exceeding the pre-loaded threshold of 40 points);

[0179] Trigger the pre-loading instruction: Load the defogging algorithm model into the video memory in advance, and initialize the motion estimation module of the vibration compensation algorithm.

[0180] When the vehicle actually encounters a sharp increase in dust caused by falling rocks in the fault zone, the defogging algorithm is in a ready state, achieving sub-second response and avoiding picture loss caused by algorithm loading delay.

[0181] In this embodiment, assume a scenario of loading and unloading operations in a medium-risk area:

[0182] A certain wide-body dump truck performs ore loading at the temporary loading and unloading point marked as "medium-risk". The system receives the data:

[0183] Position coordinates: Matched to "East Area Temporary Yard - Medium-risk";

[0184] Operation phase identifier: "Fixed-point loading and unloading".

[0185] Processing flow:

[0186] According to the characteristics of the operation phase, preferentially increase the weight of "low light" (insufficient night-time supplementary lighting in the loading and unloading area);

[0187] After adjustment, the score of "low light" reaches 36 points (threshold 35 points), triggering the pre-loading of the low illuminance enhancement algorithm;

[0188] When the loading and unloading machinery blocks part of the lighting, the system immediately applies the pre-loaded local brightness enhancement algorithm to ensure that the image of the grab operation is clearly traceable.

[0189] Refer to Figure 4 , in one of the embodiments, in step S300, the construction of the preset severity scoring model includes the following steps:

[0190] S310: Acquire a sample data set, the data set including historical environmental parameter data, corresponding image quality evaluation data, and annotations of the impact level of each sub-state type on image quality;

[0191] S320, allocating an initial weight coefficient for each sub-state type based on a preset rule, wherein the initial weight coefficient is positively correlated with a predefined negative impact degree of the sub-state on image quality;

[0192] S330: Constructing a severity scoring model, using the environmental parameter deviation, initial weight coefficient, and image quality evaluation data as input features, and using a machine learning algorithm to train the severity scoring model to obtain a trained severity scoring model for receiving the environmental parameter deviation of each sub-state in real time and outputting a weight score for each sub-state;

[0193] S340. Use the test set to verify the performance of the severity scoring model, and optimize the severity scoring model parameters based on the verification results.

[0194] As described in step S310 above, the sample data set is constructed in the following manner:

[0195] (1) Data collection:

[0196] Collect time series data of multi-dimensional environmental parameters (light, dust, vibration, humidity) during historical operations of mining vehicles;

[0197] Simultaneously collect original images of the corresponding time period and image quality scores annotated by experts (including indicators such as clarity, contrast, and key target recognizability);

[0198] Each data sample is marked with the effective sub-state type and its combined impact level (such as high dust alone, high dust and low light combined).

[0199] (2) Data preprocessing:

[0200] Perform time alignment, outlier removal, and missing value interpolation on the original data;

[0201] Quantitative indicators (such as the SSIM structural similarity index) are generated through image quality evaluation algorithms and fused with manual annotation results to form comprehensive quality labels.

[0202] As described in step S320 above, the initial weight distribution rules include:

[0203] Expert knowledge guidance: Based on mining safety regulations and image processing experience, the image quality damage level of each sub-state is preset (for example: high dust > low light > high frequency vibration > high humidity);

[0204] Literature benchmark reference: Combining the research conclusions of optical imaging, quantifying the influence coefficients of different environmental factors on the signal-to-noise ratio of the image sensor, and converting them into normalized initial weights;

[0205] Influence degree mapping: Establish an association matrix between the sub-state types and typical image defects (such as fogging, blurring, low contrast), and determine the weight ratio through matrix eigenvalue decomposition.

[0206] As described in step S330 above, the model construction process includes:

[0207] (1) Feature engineering:

[0208] The input features include the deviation degree of environmental parameters (continuous value), initial weights (static coefficients), and real-time image quality indicators (dynamic feedback);

[0209] Perform dimensionality reduction on high-dimensional features and extract feature combinations strongly correlated with the score (such as the non-linear interaction feature between the dust deviation degree and the image fogging degree).

[0210] (2) Model training:

[0211] Use the Gradient Boosting Decision Tree (GBDT) algorithm to construct a regression model with the goal of minimizing the mean square error of the prediction score and the real image quality loss;

[0212] Introduce an attention mechanism to dynamically adjust the weight coupling relationship of different sub-states under composite working conditions;

[0213] Prevent overfitting through regularization constraints to ensure the generalization ability of the model under unknown environmental combinations.

[0214] As described in step S340 above, the model verification and optimization methods include:

[0215] (1) Performance evaluation:

[0216] Use an independent test set to calculate the Pearson correlation coefficient and mean absolute error (MAE) between the prediction score and the real image quality indicator;

[0217] Conduct special stress tests on high-risk working condition samples (such as extreme value combinations of multiple sub-states).

[0218] (2) Parameter tuning:

[0219] Based on the Bayesian optimization algorithm, search for the optimal combination of model hyperparameters (learning rate, tree depth, regularization coefficient);

[0220] Adopt an online learning mechanism to incrementally update the newly collected operation data to the training set and continuously iterate the model version.

[0221] In this embodiment, the limitations of traditional fixed rules or pure data-driven models are broken through, so that the comprehensive decision-making accuracy and stability of the system are increased in complex interference scenarios such as sudden changes in dust concentration and sudden drops in light intensity, and problems such as insufficient scene adaptability caused by the static and single environmental state evaluation model in the traditional image optimization system are solved.

[0222] In one of the embodiments, in step S320, when allocating the initial weight coefficients of each sub-state type based on preset rules, the following steps are included:

[0223] S321. Define multiple influence parameters for each sub-state type, and the influence parameters at least include image quality influence parameters, safety risk parameters, and data reliability parameters;

[0224] S322. Based on the influence parameters, generate initial weight coefficients through weighted calculation, and the weight allocation rules of the weighted calculation are dynamically updated according to real-time data or environmental changes;

[0225] S323. When it is detected that the data reliability parameter is lower than the preset threshold, trigger the automatic recalibration of the weight coefficient.

[0226] As described in step S321 above, the sub-state influence parameters are defined through the following dimensions:

[0227] (1) Image quality influence parameters:

[0228] Quantify the influence degree of the sub-state on the key image indicators (such as edge sharpness, color fidelity, noise level);

[0229] Establish the mapping relationship between the sub-state type and the image defect (such as high dust corresponding to the atomization index, and low light corresponding to the signal-to-noise ratio attenuation rate).

[0230] (2) Safety risk parameters:

[0231] Associate the safety level of the area where the sub-state is located (such as the risk coefficient of abnormal dust concentration in a high-risk area is doubled);

[0232] Combine the characteristics of the operation stage (such as the sensitivity of vibration to monitoring during loading and unloading operations is increased).

[0233] (3) Data reliability parameters:

[0234] Evaluate the confidence of sensor data (including signal stability, historical false alarm rate, equipment health status);

[0235] Monitor the integrity of the data acquisition link (such as the impact of transmission delay and packet loss rate on timing consistency).

[0236] As described in step S322 above, the dynamic weighted calculation mechanism includes:

[0237] (1)Parameter normalization processing:

[0238] Perform standardization conversion on each influencing parameter (such as Min-Max normalization, Z-Score standardization) to eliminate the dimension difference;

[0239] Introduce a time decay factor for the safety risk parameter to reduce the long-term influence weight of historical risk events.

[0240] (2)Dynamic weight allocation:

[0241] Based on the output of the real-time environmental state classification model, automatically select a preset weight allocation strategy (such as giving priority to increasing the weight of the safety risk parameter in the "underground transportation mode");

[0242] Adopt a sliding window to statistically analyze the characteristics of the real-time data stream (such as parameter fluctuation frequency) and dynamically adjust the weighting ratio of each influencing parameter.

[0243] As described in the above step S323, automatically recalibrate the trigger condition and execution logic:

[0244] (1)Reliability threshold monitoring:

[0245] When the data reliability parameter is continuously lower than the preset threshold (such as the sensor confidence level < 80% and continuous timeout), generate a calibration trigger signal;

[0246] Force the start of recalibration for multi-source data conflict scenarios (such as contradictions between the light sensor and the image brightness analysis result).

[0247] (2)Calibration execution process:

[0248] Switch to the backup data source or enable virtual sensor data compensation based on the physical model;

[0249] Use historical reliable data to fill back the current parameters and recalculate the initial weight coefficient;

[0250] Temporarily freeze the score output during the calibration period until the data reliability returns to the safe range.

[0251] In this embodiment, by simultaneously evaluating the image quality influencing parameter, the safety risk parameter, and the data reliability parameter, the limitation of the traditional method relying only on a single environmental parameter is broken through, making the weight allocation more in line with the actual working condition requirements. Through the weight allocation rule, it is dynamically adjusted according to real-time data and environmental changes, solving the problem of insufficient adaptability of the fixed weight model in a complex mine environment.

[0252] It should be understood that the sequence numbers of the steps in the above embodiments do not imply the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0253] In one embodiment, an image optimization system for a driving recorder is provided. This image optimization system for a driving recorder corresponds to the image optimization method for a driving recorder in the above embodiment. This image optimization system for a driving recorder includes:

[0254] A data acquisition module, configured to obtain image data and environmental parameter data of a target scene in real time through a driving recorder and an acquisition module deployed on a mining vehicle;

[0255] An environment determination module, configured to output the current environmental state type through an environmental state classification model based on the environmental parameter data, where the environmental state type includes at least one or a combination of a low-light state, a high-dust state, a high-frequency vibration state, and a high-humidity state;

[0256] A calculation module, configured to calculate its weight score through a preset severity scoring model according to the environmental parameter deviation degree of each sub-state in the environmental state type, and determine the priority order of the image optimization algorithm according to the scoring result;

[0257] An image optimization module, configured to sequentially call image optimization algorithms associated with each sub-state to process the image data in the priority order to generate optimized image data.

[0258] For the specific limitations on an image optimization system for a driving recorder, reference may be made to the limitations on an image optimization method for a driving recorder in the foregoing, which will not be elaborated here. Each module in the above image optimization system for a driving recorder can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in a computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above modules.

[0259] In one embodiment, a computer device is provided. This computer device may be a server, and its internal structure diagram may be as Figure 5As shown. The computer device includes a processor, a memory, a network interface, and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store data, data processing, predictive analysis, etc. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for optimizing the images of a driving recorder.

[0260] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements a method for optimizing the images of a driving recorder.

[0261] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, it implements a method for optimizing the images of a driving recorder.

[0262] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0263] Those skilled in the art can clearly understand that, for the convenience and conciseness of description, only the above-mentioned division of each functional unit and module is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0264] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application 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 of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An image optimization method for a driving recorder, characterized in that, It includes the following steps: Real-time obtain the image data and environmental parameter data of the target scene through the driving recorder and the acquisition module deployed on the mining vehicle; Based on the environmental parameter data, output the current environmental state type through the environmental state classification model, and the environmental state type includes at least one of low light state, high dust state, high frequency vibration state, and high humidity state; According to the environmental parameter deviation degree of each sub-state in the environmental state type, calculate its weight score through the preset severity scoring model, and determine the priority order of the image optimization algorithm according to the scoring result; wherein, the severity scoring model satisfies the following conditions: Allocate dynamic weight coefficients for different sub-state types, and the weight coefficients are positively correlated with the negative impact degree of the sub-state on the image quality; Calculate the score according to the deviation degree of each sub-state parameter value from the preset threshold. The greater the deviation degree, the higher the score. When multiple sub-states coexist, preferentially execute the image optimization algorithm associated with the sub-state with the highest score; Call the image optimization algorithms associated with each sub-state in sequence according to the priority order to process the image data and generate the optimized image data; Among them, the construction of the preset severity scoring model includes the following steps: Obtain a sample data set, which includes historical environmental parameter data, corresponding image quality evaluation data, and annotations of the influence level of each sub-state type on the image quality; Allocate initial weight coefficients for each sub-state type based on preset rules, and the initial weight coefficients are positively correlated with the predefined negative impact degree of the sub-state on the image quality; Construct a severity scoring model, use the environmental parameter deviation degree, initial weight coefficients, and image quality evaluation data as input features, and train the severity scoring model using a machine learning algorithm to obtain a trained severity scoring model for receiving the environmental parameter deviation degree of each sub-state in real time and outputting the weight score of each sub-state; Verify the performance of the severity scoring model using the test set and optimize the parameters of the severity scoring model according to the verification results.

2. The method for optimizing the image of a driving recorder according to claim 1, wherein In the step of outputting the current environmental state type through the environmental state classification model based on the environmental parameter data, the environmental state classification model determines the environmental state type in the following way: Compare the environmental parameter data with the preset thresholds corresponding to each sub-state in the environmental state type; According to the comparison result, combine whether the environmental parameter value exceeds the preset threshold and its deviation degree to determine the current environmental state type.

3. The method for optimizing the images of a driving recorder according to claim 2, wherein, The preset threshold is dynamically determined by at least one of the following methods: Match the real-time position of the vehicle with the preset mining area map, and load the regional environmental reference value associated with the current position as the environmental parameter threshold. The preset mining area map includes the boundary coordinates of each area, the environmental parameter reference value, and the risk level; Dynamically calculate the current environmental parameter threshold according to the statistical result of the environmental parameter data within the historical time window.

4. The method for optimizing the image of a driving recorder according to claim 3, wherein, In the step of matching the real-time position of the vehicle with the preset mining area map and loading the regional environmental reference value associated with the current position as the environmental parameter threshold, the determination method of the regional environmental reference value specifically includes the following steps; Collect multi-dimensional input features, where the multi-dimensional input features at least include mine operation plans, equipment operation status, and real-time meteorological data; Input the multi-dimensional input features into a pre-trained time series prediction model, and the time series prediction model outputs a prediction result of the environmental parameter baseline value within a preset time period; Update the baseline value of the corresponding area in the mining area map according to the prediction result.

5. The method for optimizing the image of a driving recorder according to claim 3, wherein, In the step of calculating the weight score of each sub-state in the environmental state type through a preset severity scoring model according to the deviation degree of the environmental parameters of each sub-state, and determining the priority order of the image optimization algorithm according to the scoring result, the following steps are further included: Receive real-time vehicle scheduling data sent by the mine car scheduling system, where the data includes real-time positioning coordinates and the current operation stage identifier; Based on the vehicle scheduling data and the risk levels of each area in the mining area map, identify the risk level of the area where the vehicle is currently located, and the risk levels include high risk, medium risk, and low risk; According to the risk level and the current operation stage identifier, obtain the sub-state type associated with the high-risk area through a preset mapping table, and increase its weight coefficient by a preset increment; When the sub-state score after the weight coefficient adjustment exceeds a preset threshold, trigger a preloading instruction for the corresponding image optimization algorithm.

6. The method for optimizing the image of a driving recorder according to claim 1, characterized in that, When performing the step of allocating the initial weight coefficients of each sub-state type based on a preset rule, the following steps are included: Define multiple influencing parameters for each sub-state type, and the influencing parameters at least include image quality influencing parameters, safety risk parameters, and data reliability parameters; Based on the influencing parameters, generate initial weight coefficients through weighted calculation, and the weight distribution rule of the weighted calculation is dynamically updated according to real-time data or environmental changes; When it is detected that the data reliability parameter is lower than a preset threshold, trigger automatic recalibration of the weight coefficient.

7. A driving recorder image optimization system for implementing the steps of a driving recorder image optimization method according to any one of claims 1-6, characterized in that, Include: A data acquisition module for real-time acquiring image data and environmental parameter data of a target scene through a driving recorder and an acquisition module deployed on a mining vehicle; An environment determination module for outputting the current environmental state type through an environmental state classification model based on the environmental parameter data, and the environmental state type includes at least one of a low-light state, a high-dust state, a high-frequency vibration state, and a high-humidity state; A calculation module for calculating the weight score of each sub-state in the environmental state type through a preset severity scoring model according to the deviation degree of the environmental parameters of each sub-state, and determining the priority order of the image optimization algorithm according to the scoring result; An image optimization module for sequentially calling image optimization algorithms associated with each sub-state to process the image data according to the priority order, and generating optimized image data.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of a method for optimizing the image of a driving recorder as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of a method for optimizing the image of a driving recorder as described in any one of claims 1-6.

Images