Mining locomotive face recognition control system

Abstract

The invention relates to the technical field of biological feature recognition, in particular to a mining locomotive face recognition control system which comprises a reference feature calibration module, a structural deviation analysis module, a reverse compensation modeling module, a periodic identity binding module and an operation authority locking module. According to the method, by measuring and calculating the deviation between the vertical distance of key points of eyebrows and eyes of a driver and reference data, the longitudinal extrusion degree of a safety helmet on the face of the driver is accurately quantified, pixel-level reconstruction is conducted on a pressed deformation area through a reverse weighted compensation and geometric longitudinal stretching algorithm, feature distortion caused by soft tissue displacement and eye crack closing is eliminated, and the safety of the driver is improved. Recovery data truly reflecting physiological features are generated for comparison, the problem of misrecognition caused by wearing of the protector is solved, an identity confirmation result is dynamically bound with an operation cycle, a verification identifier is automatically cleared when it is monitored that the locomotive stops for a long time and the speed returns to zero, and identity verification accuracy and locomotive management and control safety are ensured when the protector is worn.

Description

Technical Field

[0001] This invention relates to the field of biometric identification technology, and in particular to a face recognition control system for mining locomotives. Background Technology

[0002] The field of biometric identification technology refers to a set of technologies that use inherent physiological or behavioral characteristics of the human body to distinguish and confirm identities. These technologies include methods for acquiring biometric information, feature representations, comparison and judgment processes, and control or management mechanisms that are combined with specific application scenarios.

[0003] Among them, the mining locomotive facial recognition control system refers to a system applied in the mining environment to confirm the identity of locomotive operators and associate it with locomotive control. It identifies the personnel entering or operating the mining locomotive and establishes a corresponding relationship with the locomotive control behavior. Typically, an imaging device is set up at the locomotive driver's position to capture images of the operator's face. In the captured facial images, facial feature description information is formed based on the preset facial contour positions of specific parts such as the eyes, nose, and mouth. This feature description information is then compared one by one with the pre-stored personnel facial feature records. Based on the comparison results, a corresponding control judgment is generated and the judgment result is used in the locomotive start-up or operation control process.

[0004] Existing technologies typically rely on static contour localization and matching of pre-stored feature records, ignoring the continuous downward mechanical pressure exerted by safety helmets on the forehead and eye socket areas during mining operations. This pressure causes longitudinal displacement of the brow bone soft tissue and alters the degree of eye opening and closing, disrupting the originally stable facial geometric topology. This results in spatial deviations between the collected features and the standard template, leading to legitimate drivers being misjudged as unqualified due to facial deformation caused by force. Furthermore, the lack of adaptive correction mechanisms for such physical deformations makes it difficult for the system to maintain high-precision recognition in scenarios where protective gear is worn. This leads to a lower identity verification pass rate and hinders efficient locomotive startup, failing to meet the continuous operation requirements under complex working conditions. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a face recognition control system for mining locomotives. The technical solution is as follows:

[0006] On the one hand, a facial recognition control system for mining locomotives was provided, which includes:

[0007] The baseline feature calibration module acquires the initial detection facial image of the current locomotive driver in front of the explosion-proof camera, determines whether the initial detection facial image appeared in the previous operating cycle of the locomotive, and acquires the current facial image of the driver to be verified.

[0008] The structural deviation analysis module extracts the key pixel coordinates in the current facial image of the driver to be verified, determines whether the facial structure of the driver to be verified is being squeezed by the helmet, and obtains the facial compression analysis results.

[0009] The reverse compensation modeling module, if the facial compression analysis result indicates that the facial structure of the driver to be verified is compressed by the helmet, corrects the facial geometry of the driver to be verified by affine transformation to obtain the restored driver facial feature vector.

[0010] The periodic identity binding module traverses the pre-set standard facial feature database of registered drivers, matches it with the currently restored facial feature vector of the driver, binds the identity information of the driver to be verified, and obtains the verification identifier for a single locomotive operation cycle.

[0011] The operation permission interlocking module adds a trigger request for the current driver to start the locomotive based on the locomotive single operation cycle verification identifier, and obtains the locomotive face recognition operation permission reset control result.

[0012] As a further aspect of the present invention, the current facial image of the driver to be verified includes panoramic pixel data of the driver's face while wearing a helmet, and image data of the eye and eyebrow regions for key point extraction. The facial compression analysis result is specifically a facial structure compression state judgment instruction generated based on the numerical deviation between the current vertical pixel distance between the eyebrows and eyes and the driver's baseline eyebrow-eye distance, and the ratio of the deviation between the distances between the left and right eye regions. The restored driver facial feature vector includes compensated eyebrow-eye region pixel coordinate data calculated based on preset facial deformation recovery parameters, and corrected eye geometric structure data generated based on a longitudinal stretching transformation model. The locomotive single-run cycle verification identifier includes a unique driver identity confirmation identifier generated after passing cosine similarity matching, and a locomotive run cycle identifier that establishes a mapping relationship with the identity confirmation identifier. The locomotive face recognition run permission reset control result is specifically a storage clear operation instruction for the locomotive single-run cycle verification identifier generated when the locomotive's real-time running speed returns to zero and the duration exceeds the preset stopping time.

[0013] As a further aspect of the present invention, the reference feature calibration module includes:

[0014] The feature cache retrieval submodule, when the current locomotive driver is in front of the explosion-proof camera, calls the explosion-proof camera to collect the initial detection facial image of the current locomotive driver, searches the preset storage space to see if there is cached facial feature data that has established a mapping relationship with the previous operating cycle, converts the initial detection facial image into a feature vector to be compared, and calculates the similarity value between the feature vector to be compared and the cached facial feature data as the feature consistency verification coefficient.

[0015] The vertical reference distance calculation submodule determines that the locomotive is in a new operating cycle to be verified if the feature consistency verification coefficient is less than the preset continuity threshold. In the new operating cycle, the explosion-proof camera is used to collect the reference facial image of the locomotive driver wearing a safety helmet. The pixel coordinates of the pupil center point and the pixel coordinates of the lower edge center point of the eyebrow in the reference facial image are identified, and the Euclidean distance between the two in the vertical direction is calculated to obtain the driver's reference eyebrow-eye distance. If the feature consistency verification coefficient is greater than or equal to the continuity threshold, the locomotive is in a continuous operating cycle. The reference facial image and the driver's reference eyebrow-eye distance stored in the previous cycle are retrieved from the preset storage space.

[0016] The eye fissure height reference extraction submodule locates the upper and lower edge pixels of the locomotive driver's left eyelid and the upper and lower edge pixels of the right eyelid in the reference facial image, measures the vertical span between the two and performs mean value processing to determine the driver's reference eye fissure height, integrates the driver's reference eyebrow-eye distance, and obtains the current facial image of the driver to be verified.

[0017] As a further aspect of the present invention, the structural deviation analysis module includes:

[0018] The key point vertical distance extraction submodule extracts the vertical position of the current lower edge pixel coordinates of the eyebrow and the current center point pixel coordinates of the pupil in the current facial image of the driver to be verified in the image coordinate system, and calculates the current eyebrow-eye vertical pixel distance between the current lower edge pixel coordinates of the eyebrow and the current center point pixel coordinates of the pupil.

[0019] The deformation deviation numerical calculation submodule calculates the difference between the current vertical pixel distance between the eyebrows and eyes and the driver's reference eyebrow and eye distance, determines the amount of change in the eyebrow and eye distance, compares it with a preset deformation fluctuation threshold, determines whether there is stress deformation in the eyebrow and eye area, and obtains the eyebrow and eye distance deviation analysis result.

[0020] The symmetry ratio determination submodule, based on the eyebrow-eye distance deviation analysis results, calculates the distance deviation values ​​of the left eye region and the right eye region in the current facial image of the driver to be verified, as well as the quotient of the distance deviation values ​​of the left eye region and the right eye region, respectively, when the numerical deviation of the eyebrow-eye distance exceeds the preset deformation fluctuation threshold. If the quotient is within the preset symmetry tolerance range, it is determined that the current facial structure is squeezed by the helmet, and a facial compression analysis result is generated.

[0021] As a further aspect of the present invention, the reverse compensation modeling module includes:

[0022] The regional pixel weighted compensation submodule, if the facial compression analysis result indicates that the current facial structure is being compressed by a safety helmet, identifies the pixel coordinates of the eyebrow and eye region in the current facial image of the driver to be verified, performs inverse weighted compensation calculation on the pixel coordinates of the eyebrow and eye region, adjusts the spatial distribution position of the pixels in the compressed region, and obtains the compensated pixel coordinates of the eyebrow and eye region.

[0023] The eye geometry longitudinal correction submodule performs vertical geometric correction on the pixels in the eye socket region of the current facial image of the driver to be verified by longitudinal stretching, based on the driver's reference eye fissure height, to correct the eye fissure closure or deformation caused by the helmet compression, and generate the corrected eye geometry structure.

[0024] The feature vector restoration and reconstruction submodule integrates the compensated eyebrow and eye region pixel coordinates with the corrected eye geometry, extracts key facial feature points, and reconstructs a multi-dimensional vector reflecting the current locomotive driver's facial physiological characteristics as the restored driver's facial feature vector.

[0025] As a further aspect of the present invention, the periodic identity binding module includes:

[0026] The cosine similarity calculation submodule traverses the preset standard facial feature database of registered drivers, extracts the original facial feature data of drivers in the standard facial feature database of registered drivers one by one, calculates the cosine similarity between the restored driver facial feature vector and the original facial feature data and sorts them, and filters the matching items with the largest cosine similarity.

[0027] The unique identity generation submodule compares the cosine similarity of the maximum cosine similarity matching item with a preset identity verification threshold. If the preset identity verification threshold is met, the identity of the current locomotive driver is confirmed, the index information of the current locomotive driver in the registered driver standard facial feature database is locked, and a driver identity confirmation identifier is generated.

[0028] The cycle verification identifier combination submodule establishes a one-way mapping relationship for locomotive operation cycle identifiers in a preset storage space, and splices the locomotive operation cycle identifiers with the driver identity confirmation identifiers to form a locomotive single operation cycle verification identifier.

[0029] As a further aspect of the present invention, the execution permission locking module includes:

[0030] The startup request response verification submodule receives the current locomotive driver's trigger request for locomotive startup, retrieves and verifies whether the locomotive single operation cycle verification identifier is stored in the preset storage space. If the verification is successful, the locomotive is allowed to enter the runnable state, and a locomotive startup response result is generated.

[0031] The dynamic comparison benchmark update submodule, based on the locomotive start response result, uses the restored driver facial feature vector as new cached facial feature data, stores it in a preset storage space, and sets it as the comparison benchmark data for the next cycle.

[0032] The permission reset control execution submodule monitors the real-time operating speed of the locomotive. When it detects that the real-time operating speed of the locomotive has returned to zero and the duration of the zero-speed state exceeds the preset downtime, it determines that the current operating cycle has ended, clears the locomotive single operating cycle verification identifier from the preset storage space, and generates the locomotive face recognition operating permission reset control result.

[0033] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0034] By measuring the vertical distance between the driver's key eyebrow and eye points and the baseline data deviation, the degree of longitudinal compression caused by the helmet on the driver's face is accurately quantified. The inverse weighted compensation and geometric longitudinal stretching algorithm are used to reconstruct the compressed deformation area at the pixel level, eliminating feature distortions caused by soft tissue displacement and eye closure. The resulting restored data that truly reflects physiological characteristics is generated for comparison, solving the problem of misidentification caused by wearing protective gear. The identity verification result is dynamically bound to the operation cycle. When the locomotive is detected to be stopped for a long time and the speed is zero, the verification mark is automatically cleared, constructing a closed-loop mechanism for dynamic identity verification based on the status, ensuring the accuracy of identity verification and the safety of locomotive control while wearing protective gear. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a schematic diagram of the face recognition control system for mining locomotives provided in an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0038] Figure 3 This is a flowchart of the reference feature calibration module in this invention;

[0039] Figure 4 This is a flowchart of the structural deviation analysis module in this invention;

[0040] Figure 5 This is a flowchart of the reverse compensation modeling module in this invention;

[0041] Figure 6This is a flowchart of the periodic identity binding module in this invention;

[0042] Figure 7 This is a flowchart of the operation permission locking module in this invention. Detailed Implementation

[0043] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0044] This invention provides a face recognition control system for mining locomotives, such as... Figure 1 The diagram shown illustrates a facial recognition control system for mining locomotives. This system includes:

[0045] The baseline feature calibration module acquires the initial detection facial image of the current locomotive driver in front of the explosion-proof camera, determines whether the initial detection facial image appeared in the previous operating cycle of the locomotive, and acquires the current facial image of the driver to be verified.

[0046] The structural deviation analysis module extracts the key pixel coordinates in the current facial image of the driver to be verified, determines whether the facial structure of the driver to be verified is being squeezed by the helmet, and obtains the facial compression analysis results.

[0047] The inverse compensation modeling module, if the facial compression analysis results indicate that the facial structure of the driver to be verified is compressed by the helmet, corrects the facial geometry of the driver to be verified through affine transformation to obtain the restored facial feature vector of the driver.

[0048] The periodic identity binding module traverses the pre-set standard facial feature database of registered drivers, matches it with the facial feature vector of the current restored driver, binds the identity information of the current driver to be verified, and obtains the verification identifier for a single locomotive operation cycle.

[0049] The operation permission interlocking module adds a trigger request for the current driver to start the locomotive based on the locomotive's single operation cycle verification identifier, and obtains the locomotive face recognition operation permission reset control result.

[0050] The current facial image of the driver to be verified includes panoramic pixel data of the driver's face while wearing a helmet, as well as image data of the eye and eyebrow regions for key point extraction. The facial compression analysis result is specifically a facial structure compression state judgment instruction generated based on the numerical deviation between the current vertical pixel distance between the eyebrows and eyes and the driver's baseline eyebrow and eye distance, and the ratio of the deviation between the distances between the left and right eye regions. The restored driver's facial feature vector includes the compensated eyebrow and eye region pixel coordinate data calculated based on preset facial deformation recovery parameters, and the corrected eye geometric structure data generated based on the longitudinal stretching transformation model. The locomotive single operation cycle verification identifier includes a unique driver identity confirmation identifier generated after passing cosine similarity matching, and a locomotive operation cycle identifier that establishes a mapping relationship with the identity confirmation identifier. The locomotive face recognition operation permission reset control result is specifically a storage and clear operation instruction generated for the locomotive single operation cycle verification identifier when the locomotive's real-time operating speed returns to zero and the duration exceeds the preset stopping time.

[0051] Please see Figure 2 and Figure 3 The baseline feature calibration module includes:

[0052] The feature cache retrieval submodule, when the current locomotive driver is in front of the explosion-proof camera, calls the explosion-proof camera to collect the initial detection facial image of the current locomotive driver, searches the preset storage space to see if there is cached facial feature data that has established a mapping relationship with the previous operating cycle, converts the initial detection facial image into a feature vector to be compared, and calculates the similarity value between the feature vector to be compared and the cached facial feature data as the feature consistency verification coefficient.

[0053] When the built-in photoelectric sensor of the explosion-proof camera detects the presence of personnel in the locomotive cab, it captures raw image data, including facial areas, through a high-resolution optical lens. This raw image data undergoes photoelectric conversion and analog-to-digital conversion by an image signal processor to form a resolution of [resolution missing]. The digital image matrix, whose resolution parameter is derived from the physical pixel array specifications of the sensor's photosensitive element, is used by the feature cache retrieval submodule to access a specific sector in the preset storage space. This sector is specifically used to store the driver's facial feature data retained at the end of the previous operating cycle. This data is retained based on the correlation record between the timestamp and the locomotive engine status. If valid timestamp-correlated data is found, the feature extraction procedure is triggered, and the initial detected facial image is input into a preset convolutional neural network model. This model structure includes an input layer that receives... The image is a grayscale image matrix of pixels, with the input size derived from the standard normalized size set during model training. The image is sequentially passed through five convolutional layers, each equipped with... Feature mapping is performed using convolutional kernels of a specific size, and a linear rectified activation function is used to introduce non-linear features. Max pooling layers connect layers to reduce data dimensionality while preserving salient features. Finally, a fully connected layer outputs a 128-dimensional feature vector for comparison, the dimension of which is determined by the physical number of output nodes in the fully connected layer. Subsequently, feature consistency verification coefficients are calculated using a cosine similarity algorithm. The feature vector to be compared (denoted as vector A) and cached facial feature data (denoted as vector B) are obtained, and their dot product and magnitude are calculated respectively. The dot product is then divided by the product of the two magnitudes. The feature consistency check coefficient is a quantitative indicator that measures the cosine of the angle between two sets of feature vectors in vector space. Its value ranges from 0 to 1, and the possibilities include: a value close to 1, representing extremely high similarity; a value close to 0, representing no similarity; and a value in the middle, representing partial feature overlap. In this calculation, assuming the normalized feature value of vector A is 0.82 and the normalized feature value of vector B is 0.85 (this is a simplified one-dimensional illustration; the actual calculation involves 128 dimensions), the cosine of the angle between them is calculated as follows: The final result of 0.99 belongs to the first possibility, that is, high similarity. This value is the feature consistency verification coefficient.

[0054] The vertical reference distance calculation submodule determines that the locomotive is in a new operating cycle to be verified if the feature consistency verification coefficient is less than the preset continuity threshold. In the new operating cycle, the explosion-proof camera is called to collect the reference facial image of the locomotive driver wearing a safety helmet. The pixel coordinates of the pupil center point and the pixel coordinates of the lower edge center point of the eyebrow in the reference facial image are identified, and the Euclidean distance between the two in the vertical direction is calculated to obtain the driver's reference eyebrow-eye distance. If the feature consistency verification coefficient is greater than or equal to the continuity threshold, the locomotive is in a continuous operating cycle. The reference facial image and the driver's reference eyebrow-eye distance stored in the previous cycle are retrieved from the preset storage space.

[0055] The calculated feature consistency verification coefficient is compared with a preset continuity threshold. This preset continuity threshold is based on the statistical collection of facial feature natural decay data during shift changes of locomotive drivers in the mining area. By collecting same-person comparison sample datasets at different time intervals within the mining area, the arithmetic mean of sample similarity is calculated. with standard deviation The threshold is determined by subtracting three times the standard deviation from the arithmetic mean. For example, if the arithmetic mean of the sample set is calculated to be 0.95 and the standard deviation is 0.03, then the calculation process for setting the preset continuity threshold is as follows: In this comparison process, two possible states are involved: the first possibility is that the feature consistency verification coefficient is less than the preset continuity threshold of 0.86, indicating that the driver's identity may have changed or that the driver has been away from their post for a long time, and the locomotive is determined to be in a new operating cycle awaiting verification; the second possibility is that the feature consistency verification coefficient is greater than or equal to the preset continuity threshold of 0.86, in which case it is determined to be a continuous operating cycle. If the calculation result belongs to the first possibility, the explosion-proof camera is controlled to acquire a reference facial image of the driver wearing a safety helmet under the assistance of a supplementary light. The gradient orientation histogram algorithm is used to locate the key facial points in the image. This algorithm constructs features by calculating the gradient orientation histogram of the local region of the image. The vertical coordinate values ​​of the pupil center point pixel coordinates and the lower eyebrow lower edge center point pixel coordinates are read respectively. These two values ​​are derived from the row scan index value of the image coordinate system. These two vertical coordinate values ​​are extracted and used in the calculation to calculate the Euclidean distance between them in the vertical direction. For example, if the extracted pupil center point vertical coordinate is 460 pixels and the lower eyebrow lower edge center point vertical coordinate is 400 pixels, the calculation process of the driver's reference eyebrow-eye distance is as follows: The driver's baseline eyebrow-eye distance is obtained as a value of 60 pixels; if it belongs to the second possibility, the baseline facial image and the corresponding driver's baseline eyebrow-eye distance stored in the previous cycle are directly retrieved by addressing.

[0056] The eye fissure height reference extraction submodule locates the upper and lower edge pixels of the locomotive driver's left eyelid and the upper and lower edge pixels of the right eyelid in the reference facial image, measures the vertical span between the two and performs mean value processing to determine the driver's reference eye fissure height, integrates the driver's reference eyebrow-eye distance, and obtains the current facial image of the driver to be verified.

[0057] For the eye region in the baseline facial image, the Canny edge detection algorithm is used to extract the eyelid contour. This algorithm first smooths and denoises the image using a Gaussian filter to remove high-frequency interference. Then, it calculates the magnitude and direction of the image pixel gradient. After non-maximum suppression, it determines the ordinates of the upper and lower edges of the left and right eyelids, respectively. These coordinates are derived from the gray-level abrupt change points identified by the edge detection operator. Vertical span values ​​are measured and averaged. First, the difference between the ordinates of the lower and upper edges of the left eyelid is calculated to obtain the left palpebral fissure height. The difference between the ordinates of the lower and upper edges of the right eyelid is calculated to obtain the right palpebral fissure height. Finally, the arithmetic mean of the left and right palpebral fissure heights is calculated. For example, if the ordinate of the upper left eyelid is 465, the ordinate of the lower left eyelid is 485, the ordinate of the upper right eyelid is 466, and the ordinate of the lower right eyelid is 484, the calculation process for the driver's baseline palpebral fissure height is as follows: The driver's baseline eye fissure height was determined to be 19 pixels. This height data was then packaged and integrated with the driver's baseline eyebrow-eye distance as a baseline geometric parameter group. At the same time, the currently acquired image was retained as the current facial image of the driver to be verified.

[0058] Please see Figure 2 and Figure 4 The structural deviation analysis module includes:

[0059] The key point vertical distance extraction submodule extracts the vertical position of the current lower edge pixel coordinates of the eyebrow and the current pupil center pixel coordinates in the image coordinate system of the current facial image of the driver to be verified, and calculates the current eyebrow-eye vertical pixel distance between the current lower edge pixel coordinates of the eyebrow and the current pupil center pixel coordinates.

[0060] In the current facial image of the driver to be verified, the cascaded regression tree algorithm is used again to track and locate facial feature points. This algorithm approximates the true feature point positions through layer-by-layer regression, optimizes the key point coordinates using the residual minimization principle, and outputs the current lower edge pixel coordinates of the eyebrow and the current center point pixel coordinates of the pupil under the influence of the helmet. The vertical position data in the image coordinate system is extracted, namely the ordinate of the current lower edge of the eyebrow and the ordinate of the current center point of the pupil. These data are directly obtained from the leaf node coordinates output by the regression tree model. The vertical pixel distance between the eyebrow and eye is calculated by subtracting the ordinate of the lower edge of the eyebrow from the ordinate of the current center point of the pupil. This yields a distance metric that reflects the potential external force compression between the eyebrow and eye at the current moment. For example, in the current data, the ordinate of the current center point of the pupil is 458 pixels, and the ordinate of the current lower edge of the eyebrow is 410 pixels. The calculation process for the vertical pixel distance between the eyebrow and eye is as follows: The current vertical pixel distance between the eyebrows and eyes is 48 pixels.

[0061] The deformation deviation numerical calculation submodule calculates the difference between the current vertical pixel distance between the eyebrows and eyes and the driver's reference eyebrow and eye distance, determines the amount of change in the eyebrow and eye distance, compares it with the preset deformation fluctuation threshold, judges whether there is stress deformation in the eyebrow and eye area, and obtains the eyebrow and eye distance deviation analysis results.

[0062] Determining whether there is stress deformation in the eyebrow and eye area includes:

[0063] The numerical sign of the change is analyzed to determine whether the current vertical pixel distance between the eyebrows and eyes has been reduced compared to the driver's baseline eyebrow and eye distance. If the change is positive or zero, it is determined that the eyebrow and eye distance has not been shortened, the conclusion of no change is output, and the determination process is terminated.

[0064] If the change is negative, it indicates a vertical compression trend. Calculate the absolute value of the change and call the preset tolerance range.

[0065] If the absolute value of the change is less than or equal to the upper limit of the tolerance range, the stress determination will not be triggered.

[0066] If the absolute value of the change is greater than the upper limit of the tolerance range, the absolute value of the change is compared with the preset deformation fluctuation threshold. Only when the absolute value is strictly greater than the preset deformation fluctuation threshold is it confirmed that the eyebrow and eye area has undergone longitudinal compression caused by external physical pressure, and it is determined that the eyebrow and eye area has stress deformation.

[0067] First, calculate the difference between the current vertical pixel distance between the eyebrows and eyes and the driver's baseline eyebrow-eye distance. This is done by subtracting the driver's baseline eyebrow-eye distance from the current vertical pixel distance to obtain the change in eyebrow-eye distance. For example, calculate... The change is then compared with a preset deformation fluctuation threshold. This threshold is set based on a statistical analysis of the elastic deformation limit dataset of facial soft tissue under normal helmet wearing, calculating the upper limit of the 95% confidence interval of this dataset. For example, the upper limit of confidence for the mean elastic deformation limit is statistically determined to be 2.8 pixels, rounded down to 3 pixels. The process of determining whether there is stress deformation in the eyebrow and eye area covers the following possible branches: The first possibility is that the change is positive or zero (e.g., the change is positive or zero). The first possibility is that the distance between the eyebrows and eyes has not shortened, and the conclusion of no deformation is output. The second possibility is that the change is negative (e.g., -12), indicating a vertical compression trend. Further subdivisions are made for the second possibility: if the absolute value of the change is less than or equal to a preset tolerance range (set to 2 pixels, derived from the root mean square error value of sensor measurement error statistics), it is considered measurement noise and not triggered; if the absolute value of the change is greater than the preset tolerance range, the absolute value of the change is further compared with a preset deformation fluctuation threshold. In this example, the absolute value of the change is calculated... The comparison process shows and Therefore, the calculation result belongs to the case where the absolute value of the change is strictly greater than the preset deformation fluctuation threshold. It is confirmed that the eyebrow and eye area has undergone longitudinal compression caused by external physical pressure, and it is determined that there is stress deformation in the eyebrow and eye area. The output value is a deviation of 12 pixels as the result of the eyebrow and eye distance deviation analysis.

[0068] The symmetry ratio determination submodule, based on the eyebrow-eye distance deviation analysis results, calculates the distance deviation values ​​of the left eye region and the right eye region in the current facial image of the driver to be verified, as well as the quotient of the distance deviation values ​​of the left eye region and the right eye region, respectively, when the numerical deviation of the eyebrow-eye distance exceeds the preset deformation fluctuation threshold. If the quotient is within the preset symmetry tolerance range, it is determined that the current facial structure is squeezed by the helmet, and the facial compression analysis results are generated.

[0069] Based on the aforementioned result that the deviation in the eyebrow-eye distance exceeds the threshold, the deviation calculation process described above is performed independently for the left and right eye regions respectively. This yields the distance deviation values ​​for the left and right eye regions. These two values ​​are derived from independent keypoint calculations and baseline subtraction for the left and right eye ROI regions, respectively. A quotient operation is then performed, calculated by dividing the distance deviation value of the left eye region by the distance deviation value of the right eye region. For example, if the calculation yields a distance deviation value of 12 for the left eye region and 11.5 for the right eye region, the quotient calculation process is as follows: The quotient is then compared with a preset symmetry tolerance range, which is set as follows: The setting is based on finite element analysis data of the head stress model. When the force eccentricity is less than 10%, it is considered positive compression. This step involves two possibilities: the first possibility is that the calculation result falls within the preset symmetry tolerance range, which is judged as uniform compression; the second possibility is that the calculation result falls outside the preset symmetry tolerance range, which is judged as eccentric compression. In this example, the calculation result 1.04 belongs to the first possibility (because...). The system determines that the current facial structure is being uniformly compressed downwards by the helmet, and generates a definitive facial compression analysis result.

[0070] Please see Figure 2 and Figure 5 The reverse compensation modeling module includes:

[0071] The regional pixel weighted compensation submodule, if the facial compression analysis result indicates that the current facial structure is being compressed by the helmet, identifies the pixel coordinates of the eyebrow and eye region in the current facial image of the driver to be verified, performs inverse weighted compensation calculation on the pixel coordinates of the eyebrow and eye region, adjusts the spatial distribution position of the pixels in the compressed region, and obtains the compensated pixel coordinates of the eyebrow and eye region.

[0072] Performing inverse weighted compensation calculations on the pixel coordinates of the eyebrow and eye area includes:

[0073] Using the geometric centroid of the pixel coordinates of the eyebrow and eye area as the reference point, determine a deformation recovery radius that covers the pixel coordinates of the eyebrow and eye area.

[0074] Iterate through each target pixel within the deformation recovery radius and calculate the relative distance between the target pixel and the geometric centroid in the vertical direction;

[0075] Establish a numerical correspondence that makes the dynamic displacement value decrease non-linearly with the increase of relative distance. Based on the numerical correspondence, assign a dynamic displacement weight to each target pixel so that the pixel closer to the geometric centroid receives a larger dynamic displacement weight.

[0076] Obtain the preset maximum vertical compression deformation of the region, multiply the maximum vertical compression deformation of the region by the dynamic displacement weight, and calculate the reverse displacement distance that the target pixel needs to bounce back vertically upward.

[0077] The reverse displacement distance is superimposed on the original ordinate of the target pixel while keeping the abscissa of the target pixel unchanged, thus determining the mapped coordinate position of the target pixel after deformation recovery.

[0078] Based on the mapped coordinate position, calculate the corresponding grayscale or color value in the current facial image of the driver to be verified, and then refill the mapped coordinate position with the calculation result.

[0079] If the facial compression analysis indicates compression, the pixel reconstruction procedure is initiated. First, the set of pixel coordinates for the eyebrow and eye region in the current facial image of the driver to be verified is identified. The geometric centroid of this region is then calculated, which is the average of the horizontal and vertical coordinates of all pixels within the region. These coordinates are derived from the moment operation on the ROI mask. For example, the geometric centroid coordinates are... The deformation recovery radius covering the area is set, based on the statistical average width of the eyebrow and eye area (e.g., 80 pixels). Each target pixel within the deformation recovery radius is iterated over, and its relative distance to the geometric centroid in the vertical direction is calculated. This distance is derived from the absolute value of the difference between the pixel's ordinate and the centroid's ordinate. For example, if a point's ordinate is 410, the relative distance calculation process is as follows: A numerical correspondence between dynamic displacement weights is established, employing a Gaussian exponential decay model to ensure that points closer to the centroid have greater weights. The calculation process is as follows: Obtain the preset maximum longitudinal compression deformation of the region (i.e., the absolute value of the aforementioned change, 12), calculate the reverse displacement distance, which is the product of the maximum longitudinal compression deformation of the region and the dynamic displacement weight. The calculation process is as follows: Round the result to 12 pixels, add the reverse displacement distance to the original ordinate of the target pixel, keeping the abscissa unchanged. The process for calculating the new ordinate of the target pixel is as follows: The target pixel's mapped coordinate position after deformation recovery is determined as follows: The color values ​​at the corresponding positions in the original image are read using bilinear interpolation and filled into the new coordinates to complete the pixel distribution adjustment.

[0080] The eye geometry longitudinal correction submodule performs vertical geometric correction on the pixels in the orbital region of the driver’s current facial image by stretching the pixels in the reference eye fissure height. This corrects the eye fissure closure or deformation caused by the helmet compression and generates the corrected eye geometry structure.

[0081] The driver's baseline eye fissure height and the measured eye fissure height in the current image to be verified are read. The longitudinal stretching ratio is calculated by dividing the driver's baseline eye fissure height by the measured eye fissure height. These two height data are respectively obtained from historical baseline storage and current real-time detection. For example, if the driver's baseline eye fissure height is 19 and the measured eye fissure height is 14, the longitudinal stretching ratio is calculated as follows: Vertical geometric correction is performed on pixels within the eye socket region. Using the vertical coordinate of the eye fissure centerline as a reference, the corrected coordinates are calculated for any pixel's vertical coordinate within the region. The calculation rule is to add the difference between the pixel's vertical coordinate and the eye fissure centerline vertical coordinate to the product of the vertical stretching ratio. For example, if the eye fissure centerline vertical coordinate is 460 and a pixel's vertical coordinate is 455, the calculation process for the corrected coordinates is as follows: It corrects the closure or deformation of the eye fissure caused by the pressure of the helmet, and generates a corrected eye geometry.

[0082] The feature vector restoration and reconstruction submodule integrates the compensated eyebrow and eye region pixel coordinates with the corrected eye geometry, extracts key facial feature points, and reconstructs a multi-dimensional vector that reflects the current locomotive driver's facial physiological characteristics as the restored driver's facial feature vector.

[0083] The image data is integrated with the compensated pixel coordinates of the eyebrow and eye region and the corrected eye geometry. The processed image data is then input into a deep neural network. This network consists of an input layer, convolutional layers, pooling layers, and fully connected layers. Specifically, the input layer receives the pre-processed image, and the first convolutional layer uses 64... Convolutional kernels are used for feature extraction, and the activation function is a linear rectified function. The dimensionality is reduced by a max pooling layer, and finally mapped to a 128-dimensional vector through a fully connected layer. The feature vector restoration and reconstruction submodule extracts key facial feature points and reconstructs a multi-dimensional vector that reflects the current locomotive driver's facial physiological characteristics. This vector is used as the restored driver's facial feature vector, which comes from the final output layer value of the forward propagation operation of the deep network.

[0084] Please see Figure 2 and Figure 6 The periodic identity binding module includes:

[0085] The cosine similarity calculation submodule traverses the pre-set standard facial feature database of registered drivers, extracts the original facial feature data of drivers from the standard facial feature database of registered drivers one by one, calculates the cosine similarity between the restored driver facial feature vector and the original facial feature data and sorts them, and selects the matching items with the largest cosine similarity.

[0086] The pre-set database of standard facial features for registered drivers is traversed, and the original facial feature data (vector B) stored in the database is extracted one by one. This data is then compared with the restored driver facial feature vector (vector A), and the cosine similarity between the two is calculated. The calculation rule is to perform a dot product operation on the two vectors and then divide by the product of their magnitudes. The cosine similarity calculation submodule calculates the similarity of all registered individuals and sorts them in descending order of value, selecting the option with the highest value as the maximum cosine similarity match. This process aims to find the legal identity record that is closest to the currently restored facial features from a large-scale database. For example, if the dot product of the restored driver facial feature vector and a certain driver's data in the database is 0.92, and the product of their magnitudes is 1, the cosine similarity calculation process is as follows: The maximum cosine similarity match found was 0.92.

[0087] The unique identity generation submodule compares the cosine similarity of the maximum cosine similarity matching item with a preset identity verification threshold. If the preset identity verification threshold is met, the identity of the current locomotive driver is confirmed, the index information of the current locomotive driver in the registered driver standard facial feature database is locked, and a driver identity confirmation identifier is generated.

[0088] The selected maximum cosine similarity matches are compared with a preset identity verification threshold. This threshold is based on the Relationship of Consequences (ROC) curve, specifically the threshold corresponding to a false acceptance rate of one in a million. For example, by analyzing a large-scale test dataset, the similarity cutoff for this false acceptance rate is determined to be 0.88. The comparison process involves two possibilities: the first possibility is that the maximum cosine similarity match is greater than the preset identity verification threshold, thus satisfying the identity verification condition; the second possibility is that the maximum cosine similarity match is less than or equal to the preset identity verification threshold, thus not satisfying the identity verification condition. In this example, the maximum cosine similarity match is 0.92, and the preset identity verification threshold is 0.88. The comparison process is as follows: If the result falls under the first possibility, the identity verification is deemed successful, the identity of the current locomotive driver is confirmed, the index information (such as Index_007) in the standard facial feature database of registered drivers is locked, and a driver identity confirmation identifier is generated.

[0089] The cycle verification identifier combination submodule establishes a locomotive operation cycle identifier with a one-way mapping relationship in the preset storage space, and splices the locomotive operation cycle identifier with the driver identity confirmation identifier to form a locomotive single operation cycle verification identifier.

[0090] A locomotive operation cycle identifier, with a one-way mapping relationship established in the preset storage space, is derived from a unique code (e.g., Cycle_2026) generated based on the locomotive start timestamp and shift information. The cycle verification identifier combination submodule concatenates this locomotive operation cycle identifier with the previously generated driver identity verification identifier (Index_007) using binary bit concatenation to form a single locomotive operation cycle verification identifier. This identifier serves as an electronic pass for this locomotive operation task, binding the locomotive status for a specific cycle to the identity information of a specific driver. For example, the binary code stream of the locomotive operation cycle identifier is 1010, and the binary code stream of the driver identity verification identifier is 0011. The operation of the concatenated single locomotive operation cycle verification identifier is illustrated below. (This is a simplified illustration only; the actual data is long bytes, and the two sets of binary data are merged through bit shift operations.)

[0091] Please see Figure 2 and Figure 7 The execution permission locking module includes:

[0092] The startup request and response verification submodule receives the current locomotive driver's trigger request for locomotive startup, retrieves and verifies whether the locomotive single operation cycle verification identifier is stored in the preset storage space. If the verification is successful, the locomotive is allowed to enter the runnable state and a locomotive startup response result is generated.

[0093] The system receives a trigger request from the current locomotive driver to start the locomotive, such as the level signal generated when the ignition button is pressed. The start request response verification submodule immediately searches and verifies whether the locomotive single-run cycle verification identifier is stored in the preset storage space. The verification process includes checking the existence of the identifier, the integrity check code, and the standardization of the format. The determination process involves two possibilities: the first is that a valid identifier is found and the verification passes; the second is that no identifier is found or the identifier is inconsistent and the verification fails. If it is the first possibility, an allow control signal is output to the locomotive control unit, allowing the locomotive to enter the runnable state and generating a locomotive start response result.

[0094] The dynamic comparison benchmark update submodule, based on the locomotive start response result, uses the restored driver's facial feature vector as new cached facial feature data, stores it in the preset storage space, and sets it as the comparison benchmark data for the next cycle.

[0095] Based on the successful locomotive start-up response, in order to adapt to the physiological changes that may occur in the driver's face during long-term work, the generated restored driver facial feature vector is extracted. This vector comes from the output of the feature vector restoration and reconstruction submodule. It is stored in the cache area of ​​the preset storage space through the overwrite instruction and set as the comparison benchmark data for the next cycle. This data will be used as the reference object for the feature cache retrieval submodule to calculate the feature consistency verification coefficient when the driver enters the camera range again, thereby realizing the dynamic iterative update of the benchmark data and ensuring that the recognition architecture can adapt to minor facial changes.

[0096] The permission reset control execution submodule monitors the real-time operating speed of the locomotive. When it detects that the real-time operating speed of the locomotive has returned to zero and the duration of the zero-speed state exceeds the preset downtime, it determines that the current operating cycle has ended, clears the locomotive single operating cycle verification flag from the preset storage space, and generates the locomotive face recognition operating permission reset control result.

[0097] Clearing the locomotive single-run cycle verification identifier from the preset storage space includes:

[0098] The real-time value of the locomotive's real-time operating speed is obtained at a fixed data sampling frequency, and a zero-speed judgment interval is set to shield signal drift noise.

[0099] The locomotive's real-time operating speed is compared with the zero-speed judgment range. When the value of the locomotive's real-time operating speed falls back to within the zero-speed judgment range, a time accumulation counter is activated to start accumulating the duration.

[0100] During the accumulation period, the real-time operating speed of the locomotive is continuously checked. If the real-time operating speed of the locomotive jumps out of the zero speed judgment range before the accumulated time reaches the preset stopping time, the current accumulation process is immediately interrupted and the time accumulation counter is reset to zero.

[0101] Only when the cumulative duration of the time accumulator counter continuously meets the requirements and exceeds the preset downtime, is it confirmed that the locomotive is in a non-temporary long-term downtime state, and the logical address of the locomotive single-run cycle verification identifier stored in the preset storage space is located.

[0102] Write overlay data consisting of a random binary sequence into the logical address, and remove the locomotive single-run cycle verification identifier from the preset storage space by physical overwriting, thereby removing the association between the locomotive single-run cycle verification identifier and the current locomotive.

[0103] The system monitors the locomotive's real-time operating speed, which is calculated from the pulse frequency collected by wheel axle sensors. When the locomotive's real-time operating speed reaches zero and the zero-speed state lasts for more than a preset stopping time, the current operating cycle is considered complete. The locomotive's single-operation cycle verification flag is then cleared from the preset storage space, and a locomotive face recognition operating permission reset control result is generated. The process of clearing the locomotive's single-operation cycle verification flag from the preset storage space includes: acquiring the real-time value of the locomotive's real-time operating speed at a fixed data sampling frequency; setting a zero-speed determination interval to shield against signal drift noise; and setting this interval based on the statistical amplitude of the sensor's static noise. For example, if the maximum statistical static noise amplitude is 0.04 km / h, the set interval is negative 0. The speed ranges from 0.05 km / h to +0.05 km / h. The judgment process involves three possibilities: First, the locomotive's real-time speed falls back to within the zero-speed judgment range, at which point a time accumulator is activated to begin accumulating time. Second, the locomotive's real-time speed jumps out of the zero-speed judgment range before the accumulated time reaches the preset stopping time, at which point the current accumulation process is interrupted and the time accumulator is reset to zero. Third, the accumulated time of the time accumulator continues to meet the requirements and exceeds the preset stopping time. For example, if the preset stopping time is set to 300 seconds, the data sampling frequency is 100 Hz, the upper and lower limits of the zero-speed judgment range are 0.05 and -0.05 respectively, and the current sampling speed is 0.02, the judgment process is as follows: This situation falls under the first possibility, entering a timing state. If the timer reaches 301 seconds (i.e.... If the locomotive is confirmed to be in a non-temporary long-term shutdown state, the storage address of the locomotive single-run cycle verification identifier in the preset storage space is located, and overwriting data composed of random binary sequences is written to the storage address. The locomotive single-run cycle verification identifier is removed from the preset storage space by physical overwriting, and the association between the locomotive single-run cycle verification identifier and the current locomotive is terminated.

[0104] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A face recognition control system for mining locomotives, characterized in that, The system includes: The baseline feature calibration module acquires the initial detection facial image of the current locomotive driver in front of the explosion-proof camera, determines whether the initial detection facial image appeared in the previous operating cycle of the locomotive, and acquires the current facial image of the driver to be verified. The structural deviation analysis module extracts the key pixel coordinates in the current facial image of the driver to be verified, determines whether the facial structure of the driver to be verified is being squeezed by the helmet, and obtains the facial compression analysis results. The reverse compensation modeling module, if the facial compression analysis result indicates that the facial structure of the current driver to be verified is compressed by the helmet, corrects the facial geometry of the current driver to be verified through affine transformation to obtain the restored driver facial feature vector. The periodic identity binding module traverses the pre-set standard facial feature database of registered drivers, matches it with the currently restored facial feature vector of the driver, binds the identity information of the driver to be verified, and obtains the verification identifier for a single locomotive operation cycle. The operation permission interlocking module adds a trigger request for the current driver to start the locomotive based on the locomotive single operation cycle verification identifier, and obtains the locomotive face recognition operation permission reset control result.

2. The face recognition control system for mining locomotives according to claim 1, characterized in that: The current facial image of the driver to be verified includes panoramic pixel data of the driver's face while wearing a helmet, and image data of the eye and eyebrow regions for key point extraction. The facial compression analysis result is specifically a facial structure compression state judgment instruction generated based on the numerical deviation between the current vertical pixel distance between the eyebrows and eyes and the driver's baseline eyebrow-eye distance, as well as the ratio of the distance deviation between the left and right eye regions. The restored driver facial feature vector includes compensated eyebrow and eye region pixel coordinate data calculated based on preset facial deformation recovery parameters, and corrected eye geometric structure data generated based on a longitudinal stretching transformation model. The locomotive single-run cycle verification identifier includes a unique driver identity confirmation identifier generated after passing cosine similarity matching, and a locomotive run cycle identifier that establishes a mapping relationship with the identity confirmation identifier. The locomotive face recognition run permission reset control result is specifically a storage clear operation instruction generated for the locomotive single-run cycle verification identifier when the locomotive's real-time running speed returns to zero and the duration exceeds the preset stopping time.

3. The face recognition control system for mining locomotives according to claim 1, characterized in that: The reference feature calibration module includes: The feature cache retrieval submodule, when the current locomotive driver is in front of the explosion-proof camera, calls the explosion-proof camera to collect the initial detection facial image of the current locomotive driver, searches the preset storage space to see if there is cached facial feature data that has established a mapping relationship with the previous operating cycle, converts the initial detection facial image into a feature vector to be compared, and calculates the similarity value between the feature vector to be compared and the cached facial feature data as the feature consistency verification coefficient. The vertical reference distance calculation submodule determines that the locomotive is in a new operating cycle to be verified if the feature consistency verification coefficient is less than the preset continuity threshold. In the new operating cycle, the explosion-proof camera is used to collect the reference facial image of the locomotive driver wearing a safety helmet. The pixel coordinates of the pupil center point and the pixel coordinates of the lower edge center point of the eyebrow in the reference facial image are identified, and the Euclidean distance between the two in the vertical direction is calculated to obtain the driver's reference eyebrow-eye distance. If the feature consistency verification coefficient is greater than or equal to the continuity threshold, the locomotive is in a continuous operating cycle. The reference facial image and the driver's reference eyebrow-eye distance stored in the previous cycle are retrieved from the preset storage space. The eye fissure height reference extraction submodule locates the upper and lower edge pixels of the locomotive driver's left eyelid and the upper and lower edge pixels of the right eyelid in the reference facial image, measures the vertical span between the two and performs mean value processing to determine the driver's reference eye fissure height, integrates the driver's reference eyebrow-eye distance, and obtains the current facial image of the driver to be verified.

4. The face recognition control system for mining locomotives according to claim 1, characterized in that: The structural deviation analysis module includes: The key point vertical distance extraction submodule extracts the vertical position of the current lower edge pixel coordinates of the eyebrow and the current center point pixel coordinates of the pupil in the current facial image of the driver to be verified in the image coordinate system, and calculates the current eyebrow-eye vertical pixel distance between the current lower edge pixel coordinates of the eyebrow and the current center point pixel coordinates of the pupil. The deformation deviation numerical calculation submodule calculates the difference between the current vertical pixel distance between the eyebrows and eyes and the driver's reference eyebrow and eye distance, determines the amount of change in the eyebrow and eye distance, compares it with a preset deformation fluctuation threshold, determines whether there is stress deformation in the eyebrow and eye area, and obtains the eyebrow and eye distance deviation analysis result. The symmetry ratio determination submodule, based on the eyebrow-eye distance deviation analysis results, calculates the distance deviation values ​​of the left eye region and the right eye region in the current facial image of the driver to be verified, as well as the quotient of the distance deviation values ​​of the left eye region and the right eye region, respectively, when the numerical deviation of the eyebrow-eye distance exceeds the preset deformation fluctuation threshold. If the quotient is within the preset symmetry tolerance range, it is determined that the current facial structure is squeezed by the helmet, and a facial compression analysis result is generated.

5. The face recognition control system for mining locomotives according to claim 1, characterized in that: The reverse compensation modeling module includes: The regional pixel weighted compensation submodule, if the facial compression analysis result indicates that the current facial structure is being compressed by a safety helmet, identifies the pixel coordinates of the eyebrow and eye region in the current facial image of the driver to be verified, performs inverse weighted compensation calculation on the pixel coordinates of the eyebrow and eye region, adjusts the spatial distribution position of the pixels in the compressed region, and obtains the compensated pixel coordinates of the eyebrow and eye region. The eye geometry longitudinal correction submodule performs vertical geometric correction on the pixels in the eye socket region of the current facial image of the driver to be verified by longitudinal stretching, based on the driver's reference eye fissure height, to correct the eye fissure closure or deformation caused by the helmet compression, and generate the corrected eye geometry structure. The feature vector restoration and reconstruction submodule integrates the compensated eyebrow and eye region pixel coordinates with the corrected eye geometry, extracts key facial feature points, and reconstructs a multi-dimensional vector reflecting the current locomotive driver's facial physiological characteristics as the restored driver's facial feature vector.

6. The face recognition control system for mining locomotives according to claim 1, characterized in that: The periodic identity binding module includes: The cosine similarity calculation submodule traverses the preset standard facial feature database of registered drivers, extracts the original facial feature data of drivers in the standard facial feature database of registered drivers one by one, calculates the cosine similarity between the restored driver facial feature vector and the original facial feature data and sorts them, and filters the matching items with the largest cosine similarity. The unique identity generation submodule compares the cosine similarity of the maximum cosine similarity matching item with a preset identity verification threshold. If the preset identity verification threshold is met, the identity of the current locomotive driver is confirmed, the index information of the current locomotive driver in the registered driver standard facial feature database is locked, and a driver identity confirmation identifier is generated. The cycle verification identifier combination submodule establishes a one-way mapping relationship for locomotive operation cycle identifiers in a preset storage space, and splices the locomotive operation cycle identifiers with the driver identity confirmation identifiers to form a locomotive single operation cycle verification identifier.

7. The face recognition control system for mining locomotives according to claim 1, characterized in that: The execution permission locking module includes: The startup request response verification submodule receives the current locomotive driver's trigger request for locomotive startup, retrieves and verifies whether the locomotive single operation cycle verification identifier is stored in the preset storage space. If the verification is successful, the locomotive is allowed to enter the runnable state, and a locomotive startup response result is generated. The dynamic comparison benchmark update submodule, based on the locomotive start response result, uses the restored driver facial feature vector as new cached facial feature data, stores it in a preset storage space, and sets it as the comparison benchmark data for the next cycle. The permission reset control execution submodule monitors the real-time operating speed of the locomotive. When it detects that the real-time operating speed of the locomotive has returned to zero and the duration of the zero-speed state exceeds the preset downtime, it determines that the current operating cycle has ended, clears the locomotive single operating cycle verification identifier from the preset storage space, and generates the locomotive face recognition operating permission reset control result.

8. The face recognition control system for mining locomotives according to claim 4, characterized in that: The determination of whether there is stress deformation in the eyebrow and eye area includes: The numerical sign of the change is analyzed to determine whether the current vertical pixel distance between the eyebrows and eyes has been reduced compared to the driver's baseline eyebrow and eye distance. If the change is positive or zero, it is determined that the eyebrow and eye distance has not been shortened, the conclusion of no change is output, and the determination process is terminated. If the change is negative, it indicates a vertical compression trend. Calculate the absolute value of the change and call the preset tolerance range. If the absolute value of the change is less than or equal to the upper limit of the tolerance range, the stress determination will not be triggered. If the absolute value of the change is greater than the upper limit of the tolerance range, the absolute value of the change is compared with the preset deformation fluctuation threshold. Only when the absolute value is strictly greater than the preset deformation fluctuation threshold is it confirmed that the eyebrow and eye area has undergone longitudinal compression caused by external physical pressure, and it is determined that the eyebrow and eye area has stress deformation.

9. The face recognition control system for mining locomotives according to claim 5, characterized in that: The inverse weighted compensation calculation for the pixel coordinates of the eyebrow and eye area includes: Using the geometric centroid of the pixel coordinates of the eyebrow and eye area as the reference point, determine a deformation recovery radius that covers the pixel coordinates of the eyebrow and eye area. Iterate through each target pixel within the deformation recovery radius and calculate the relative distance between the target pixel and the geometric centroid in the vertical direction; Establish a numerical correspondence that makes the dynamic displacement value decrease non-linearly with the increase of relative distance. Based on the numerical correspondence, assign a dynamic displacement weight to each target pixel so that the pixel closer to the geometric centroid receives a larger dynamic displacement weight. Obtain the preset maximum vertical compression deformation of the region, multiply the maximum vertical compression deformation of the region by the dynamic displacement weight, and calculate the reverse displacement distance that the target pixel needs to bounce back vertically upward. The reverse displacement distance is superimposed on the original ordinate of the target pixel while keeping the abscissa of the target pixel unchanged, thus determining the mapped coordinate position of the target pixel after deformation recovery. Based on the mapped coordinate position, the corresponding grayscale or color value is calculated in the current facial image of the driver to be verified, and the calculation result is refilled into the mapped coordinate position.

10. The face recognition control system for mining locomotives according to claim 7, characterized in that: The step of clearing the locomotive single-run cycle verification identifier from the preset storage space includes: The real-time value of the locomotive's real-time operating speed is obtained at a fixed data sampling frequency, and a zero-speed judgment interval is set to shield signal drift noise. The locomotive's real-time operating speed is compared with the zero-speed judgment range. When the value of the locomotive's real-time operating speed falls back to within the zero-speed judgment range, a time accumulation counter is activated to start accumulating the duration. During the accumulation period, the real-time operating speed of the locomotive is continuously checked. If the real-time operating speed of the locomotive jumps out of the zero speed judgment range before the accumulated time reaches the preset stopping time, the current accumulation process is immediately interrupted and the time accumulation counter is reset to zero. Only when the cumulative duration of the time accumulator counter continuously meets the requirements and exceeds the preset downtime, is it confirmed that the locomotive is in a non-temporary long-term downtime state, and the logical address of the locomotive single-run cycle verification identifier stored in the preset storage space is located. Write overlay data consisting of a random binary sequence into the logical address, and remove the locomotive single-run cycle verification identifier from the preset storage space by physical overwriting, thereby deactivating the association between the locomotive single-run cycle verification identifier and the current locomotive.

Images