An image data processing method and system for face recognition of charging piles

Abstract

The application discloses an image data processing method and system for face recognition of charging piles, and relates to the technical field of regulation and control, comprising: generating different reference extraction regions, acquiring real-time image information and real-time shielding information; acquiring shielding efficiency, acquiring invalid regions, acquiring effective regions, and acquiring shielding correction weights; acquiring pre-stored image information of the effective regions of the users in a pre-stored user information database of the charging piles, acquiring a to-be-processed similarity according to the real-time image information of the effective regions and the pre-stored image information of the effective regions of the users, acquiring target similarities of the users, and acquiring a total similarity of the users according to the target similarities of the users; judging whether there is a user with a total similarity exceeding a consistency threshold in the user information database of the charging piles, and if so, outputting permission information of the user with the maximum total similarity. The application has the advantages of good image data processing effect, effectively reduced shielding influence, and high data processing efficiency.

Description

Technical Field

[0001] This invention relates to the field of adjustment and control technology, and specifically to an image data processing method and system for facial recognition in charging piles. Background Technology

[0002] In the actual application scenarios of facial recognition systems for charging stations, users often experience partial dynamic occlusion of their faces due to wearing protective equipment such as helmets, masks, and goggles, or due to environmental factors such as strong light, rain, fog, and especially the shadows of tree branches. Existing image data processing methods perform poorly in such situations, affecting the user's facial payment experience.

[0003] Specifically, existing image data processing methods for facial recognition at charging stations directly incorporate features of occluded areas into similarity calculations (e.g., forced matching even when the nose is obscured by a helmet), leading to contaminated recognition results. Furthermore, the quantification of occlusion degree is lacking, and no correlation mechanism between occlusion degree and recognition confidence has been established, making it impossible to dynamically adjust the contribution weights of different regions. Furthermore, feature points in different facial regions are correlated (e.g., the ratio between the width of the nostrils and the distance between the eyes), and occlusion in one region can disrupt the overall geometric constraints, causing cascading error propagation; for example, occlusion of the bridge of the nose can still amplify matching errors in other areas of the nose. Finally, charging station scenarios require recognition to be completed within a short time, and existing image processing methods lack the ability to respond quickly to sudden occlusions (e.g., a scarf covering the cheek when the user turns their head). Summary of the Invention

[0004] Existing technologies do not divide the face into independent reference regions and isolate occluded parts, causing local failures to spread to the whole. At the same time, similarity calculation uses fixed weights and does not dynamically adjust the region reliability according to the severity of occlusion. An image processing scheme with regional occlusion isolation and dynamic weight compensation capabilities is needed to solve the problem of sharp drop in recognition rate caused by local occlusion. Therefore, the present invention provides an image data processing method and system for face recognition of charging piles.

[0005] An image data processing method for facial recognition in charging piles includes: generating reference regions for the eyes, nose, mouth, left face, right face, and jaw based on the user's facial features, and using these as different reference extraction regions; acquiring real-time image information and real-time occlusion information of the user within each reference extraction region based on the charging pile image acquisition module; obtaining the occlusion effectiveness rate of the reference extraction regions based on the real-time occlusion information within the reference extraction regions; identifying reference extraction regions with an occlusion effectiveness rate exceeding a preset threshold as invalid regions; and identifying reference extraction regions with an occlusion effectiveness rate not exceeding a preset threshold as valid regions. The system obtains the occlusion correction weight based on the number of users; it obtains the pre-stored image information of the i-th effective region of the j-th user in the pre-stored user information database; it obtains the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user; it obtains the i-th target similarity of the j-th user based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user; it obtains the total similarity of the j-th user based on each target similarity of the j-th user; it determines whether there is a user in the charging pile user information database whose total similarity exceeds the consistency threshold; if so, it outputs the permission information of the user with the largest total similarity.

[0006] Optionally, obtaining the occlusion effectiveness of the benchmark extraction area based on real-time occlusion information within the benchmark extraction area includes: obtaining multiple points to be extracted within the benchmark extraction area; obtaining multiple points not extracted within the benchmark extraction area based on real-time occlusion information within the benchmark extraction area; and obtaining the occlusion effectiveness based on the proportion of the number of unextracted points to the number of points to be extracted.

[0007] Optionally, obtaining the occlusion correction weight based on the number of invalid regions includes: obtaining the unit correction ratio; and obtaining the occlusion correction weight based on the number of invalid regions and the unit correction ratio.

[0008] Optionally, the occlusion correction weight, obtained based on the number of invalid regions and the unit correction ratio, is expressed as follows: Among them, A c To correct the weights for occlusion, A u X is the unit correction ratio. in This represents the number of invalid regions.

[0009] Optionally, obtaining the i-th similarity to be processed for the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user includes: obtaining multiple points to be extracted located within the i-th effective region; obtaining real-time relative position ratio data of the multiple points to be extracted based on the real-time image information of the i-th effective region, and obtaining pre-stored relative position ratio data of the multiple points to be extracted based on the pre-stored image information of the i-th effective region of the j-th user; obtaining relative position similarity based on the real-time relative position ratio data and the pre-stored relative position ratio data, and using it as the i-th similarity to be processed for the j-th user.

[0010] Optionally, the target similarity of the j-th user is obtained based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user, represented as: S tji =A c ·S wji Among them, S tji Let A be the similarity of the j-th user to the i-th target. c To correct the weights for occlusion, S wji Let be the i-th similarity to be processed for the j-th user.

[0011] Optionally, the total similarity of the j-th user can be obtained based on the individual target similarities, as represented by: Among them, S j S represents the total similarity of the j-th user. tji The similarity of the j-th user to the i-th target, m j Let be the number of target similarities for the j-th user.

[0012] A system for image data processing for facial recognition in charging piles is also provided. The system includes: an image data acquisition module, used to generate reference regions for both eyes, nose, mouth, left face, right face, and jaw based on the user's facial features, and to use these as different reference extraction regions; and to acquire real-time image information and real-time occlusion information of the user within each reference extraction region based on the charging pile image acquisition module; and a first image data processing module, used to obtain the occlusion effectiveness rate of the reference extraction regions based on the real-time occlusion information within the reference extraction regions; to obtain reference extraction regions with an occlusion effectiveness rate exceeding a preset threshold and designate them as invalid regions; and to obtain reference extraction regions with an occlusion effectiveness rate not exceeding a preset threshold and designate them as valid regions; and to process the invalid regions based on the real-time occlusion information within the reference extraction regions. The system employs a first image processing module to obtain occlusion correction weights based on the number of regions. The second image processing module obtains pre-stored image information of the i-th valid region for the j-th user from the pre-stored user information database. It then obtains the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th valid region and the pre-stored image information of the i-th valid region of the j-th user. Finally, it obtains the i-th target similarity of the j-th user based on the occlusion correction weights and the i-th unprocessed similarity of the j-th user, and obtains the total similarity of the j-th user based on each target similarity. The third image processing module determines whether there are users in the charging pile user information database whose total similarity exceeds a consistency threshold. If so, it outputs the permission information of the user with the highest total similarity.

[0013] Optionally, the first image data processing module is further configured to: acquire multiple points to be extracted within the reference extraction area; acquire multiple unextracted points within the reference extraction area based on real-time occlusion information within the reference extraction area; and acquire the occlusion efficiency based on the proportion of the number of unextracted points to the number of points to be extracted.

[0014] Optionally, the first image data processing module is further configured to: obtain the unit correction ratio; and obtain the occlusion correction weight based on the number of invalid regions and the unit correction ratio.

[0015] The beneficial effects of this invention are reflected in:

[0016] In the entire image data processing method for facial recognition in charging piles, firstly, invalid regions are accurately determined by quantifying the missing feature point rate (e.g., the root of the nose point is not extracted due to helmet occlusion), ensuring complete isolation of occluded data; furthermore, occlusion correction weights are generated based on the number of invalid regions, actively amplifying the similarity contribution value of valid regions (e.g., the eye region that can still be identified under strong light); furthermore, a partitioned independent processing mechanism (e.g., comparing only the pre-stored data of the eye regions not covered by masks) combined with real-time occlusion analysis capabilities quickly responds to sudden occlusion events such as scarf fluttering, ensuring the timeliness requirements of charging scenarios. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 This is a schematic diagram illustrating the steps of the image data processing method for face recognition in charging piles according to the present invention;

[0019] Figure 2 This is a schematic diagram of a portion of the steps in the image data processing method S2 for face recognition in charging piles according to the present invention;

[0020] Figure 3 This is a schematic diagram of another part of the steps in the image data processing method S2 for face recognition of charging piles according to the present invention;

[0021] Figure 4 This is a schematic diagram showing the distribution of extraction points in six different reference extraction regions in the image data processing method for face recognition of charging piles according to the present invention.

[0022] Figure 5 This is a schematic diagram of some steps in the image data processing method S3 for face recognition of charging piles according to the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] like Figure 1and Figure 4 As shown, an image data processing method for facial recognition in charging stations is provided, including:

[0027] S1. Generate reference regions for both eyes, nose, mouth, left face, right face and jaw based on the user's facial features and use them as different reference extraction regions. Then, obtain real-time image information and real-time occlusion information of the user in each reference extraction region based on the charging pile image acquisition module.

[0028] S2. Obtain the occlusion efficiency of the benchmark extraction area based on the real-time occlusion information in the benchmark extraction area, obtain the benchmark extraction area with an occlusion efficiency exceeding a preset threshold and treat it as an invalid area, obtain the benchmark extraction area with an occlusion efficiency not exceeding a preset threshold and treat it as an effective area, and obtain the occlusion correction weight based on the number of invalid areas.

[0029] S3. Obtain the pre-stored image information of the i-th effective region of the j-th user in the pre-stored user information database of the charging pile. Obtain the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user. Obtain the i-th target similarity of the j-th user based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user. Obtain the total similarity of the j-th user based on each target similarity of the j-th user.

[0030] S4. Determine whether there are users in the charging pile user information database whose total similarity exceeds the consistency threshold. If so, output the permission information of the user with the highest total similarity.

[0031] In this embodiment, it should be noted that in S1, the user's face is divided into multiple independent reference extraction regions to achieve region-level processing and avoid the spread of local occlusion problems. This step first automatically generates and divides six reference regions (eyes, nose, mouth, left face, right face, and jaw) based on the user's inherent anatomical features (such as eye position, nose contour, mouth shape, etc.); each region is considered an independent semantic unit, including multiple extraction points; such as... Figure 4As shown, each region can accurately cover key facial structures. For example, the eye reference region includes extraction points 18 to 27 and 37 to 48, the nose reference region includes extraction points 28 to 36, the mouth reference region includes extraction points 49 to 68, the left face reference region includes extraction points 1 to 6, the right face reference region includes extraction points 12 to 17, and the jaw reference region includes extraction points 7 to 11. This method of partitioning and setting multiple extraction points essentially uses a spatial isolation mechanism to predict and mitigate the impact of occlusion, ensuring that effective information can be focused on in subsequent processing.

[0032] Furthermore, the S1 utilizes the image acquisition module integrated into the charging pile (such as a built-in high-definition camera) to capture real-time facial image data of the user. For each reference extraction area, it simultaneously extracts two key pieces of information: real-time image information and real-time occlusion information. Real-time image information includes image pixel data and geometric relative position data (distance ratio data between multiple extraction points) within the area. Real-time occlusion information is obtained by dynamically identifying coverings or interfering factors in each area through image analysis algorithms (for example, strong light may obscure the outline of the right facial reference area, or the helmet strap may obscure key points of the nose reference area). In practical applications, for example, when a user is riding while wearing a face mask, occlusion information is processed separately—the mouth reference area may experience real-time occlusion due to the face mask, but the real-time image information of the eye reference areas remains intact, thus providing independent support for subsequent weight correction and similarity calculation. This process ensures rapid response to sudden changes (such as temporary occlusion caused by the user's scarf fluttering), improving the overall adaptability to common interference in charging pile scenarios.

[0033] In S2, the effectiveness of regions and weighted classification are corrected by dynamically quantifying the impact of occlusion. First, the recognizability of each benchmark extraction region is analyzed based on real-time occlusion information: for the six benchmark extraction regions, multiple preset points to be extracted within each region (such as the nasal alar and nasal root in the nose region) are extracted. By identifying the number of unextracted points that cannot be obtained due to occlusion, the occlusion effectiveness rate (i.e., the proportion of unextracted points to all extractable points in the region) is calculated. For example, when a user wears a helmet with a nose shield, key points such as the nasal root in the nose benchmark region may not be collected due to physical occlusion, and in this case, the occlusion effectiveness rate is determined to be high; if rain or fog only causes slight blurring of the left face region, and some contour points can still be extracted, then the occlusion effectiveness rate of this region is low. This quantification method based on the feature point missing rate accurately reflects the degree of damage to the recognition data caused by local occlusion.

[0034] Furthermore, based on the occlusion effectiveness determination results, regions are classified into effective regions (occlusion effectiveness does not exceed a preset threshold, and feature points are basically complete) and invalid regions (occlusion effectiveness exceeds the standard, and feature points are severely missing); generally, the preset threshold for occlusion effectiveness is 30%. Further, the total number of invalid regions is counted, and a global occlusion correction weight is dynamically generated accordingly—this weight is positively correlated with the number of invalid regions. The core compensation logic is that when an invalid region appears, the weight coefficient increases proportionally (>1). For example, if a user's nose and jaw regions become invalid simultaneously due to wearing a full-face helmet (creating two invalid regions), the weight coefficient will increase significantly; while when only the mouth region becomes invalid due to a medical mask, the weight increase is relatively small. In summary, in subsequent calculations, the similarity contribution value of effective regions is actively amplified—for example, after a helmet occludes the nose, the weight coefficient is increased to enhance the similarity calculation results of effective regions such as the eyes and cheeks, so that the comprehensive score (total similarity) of the remaining effective regions still has a chance to reach the consistency threshold, thereby overcoming the recognition obstacle caused by local occlusion while ensuring safety.

[0035] In S3, accurate identity verification is achieved through region-independent matching and global compensation enhancement. First, all invalid regions marked by S2 (such as the nose area completely obscured by a helmet) are excluded, and operations are performed only on valid regions (such as the unobscured eyes and cheeks). When extracting pre-stored feature data from the charging station user information database, each valid region of the same user is independently compared—for example, the real-time iris texture and corner position ratio data of user A's eye baseline region are matched one by one with the baseline geometric features of the same region in the pre-stored database. Specifically, the relative position ratios of multiple key points within the region are extracted (such as the ratio of pupillary distance to palpebral fissure width in the eye baseline region). By comparing the deviation between the real-time collected data and the pre-stored data, the initial similarity score (i.e., the similarity to be processed) for that region is calculated. This region-independent calculation mode completely blocks the error propagation problem caused by local occlusion in traditional methods. For example, when a user wearing goggles causes abnormal data in the brow bone region, only the data in that region is discarded, without affecting the feature matching accuracy of the facial region.

[0036] Furthermore, an occlusion compensation mechanism is introduced to enhance the weight of the matching results for effective regions: the global occlusion correction weight (>1) generated in stage S2 is multiplied by the unprocessed similarity of each effective region to generate a weighted target similarity. For example, when a user's nose and jaw areas are invalid due to a full-face helmet, a higher weight coefficient (e.g., 1.3) is generated. If the unprocessed similarities of the remaining eyes and left face areas are 0.8 and 0.7 respectively, the weighted similarity increases to 1.04 and 0.91. Finally, the target similarities of all effective regions are summed to generate the user's total similarity. This design essentially overcomes the recognition threshold bottleneck caused by occlusion by compensating for the contribution value of effective regions. A typical example is when a user is wearing a mask; after the mouth area is removed, the similarity score of effective regions (such as the eyes) is increased by weighting, so that the total similarity may still exceed the consistency threshold, significantly improving the pass rate in complex occlusion scenarios while ensuring safety.

[0037] In S4, a weighted decision-making mechanism is used to complete the final identity verification. This step first iterates through all registered users in the user information database and calculates the total similarity (i.e., the sum of weighted target similarities for all valid regions) generated by S3 for each user, comparing it with a preset consistency threshold. For example, when a user is wearing a full-face helmet, causing multiple facial regions to become invalid, the weighted total similarity calculated based only on the unobstructed eyes, cheeks, and other valid regions may still reach 0.92. If this value exceeds the threshold (e.g., 0.9), the identity verification is considered successful. This design avoids interference from invalid regions, enabling reliable results even when there is severe partial occlusion of the face.

[0038] Furthermore, an optimal matching strategy is implemented: if the total similarity of multiple users exceeds a threshold (e.g., users A and B in the charging station user database simultaneously meet the criteria), the user with the highest total similarity is selected as the final identification target. A typical example is the twin user scenario: when user C (one of the twins) has their right cheek partially obscured due to tree branch shadows, their weighted total similarity might be 0.95, while user D (the other twin) has 0.93. In this case, user C's charging payment permissions are accurately output. If no user meets the threshold requirement (e.g., strong light causes more than half of the face area to fail), the process automatically ends to avoid misidentification. This dynamic threshold decision-making mechanism significantly improves the recognition success rate in sudden occlusion scenarios while ensuring safety.

[0039] In summary, the image data processing method for facial recognition in charging stations firstly quantifies the missing feature point rate (e.g., the nasal root point is not extracted due to helmet occlusion) to accurately identify invalid regions, ensuring complete isolation of occluded data. Secondly, occlusion correction weights are generated based on the number of invalid regions, actively amplifying the similarity contribution value of valid regions (e.g., the eye region still recognizable under strong light). Thirdly, a partitioned independent processing mechanism (e.g., comparing only pre-stored data of the eye regions not covered by masks) combined with real-time occlusion analysis capabilities allows for rapid response to sudden occlusion events such as scarf movement, ensuring the timeliness requirements of charging scenarios. Finally, geometric constraint decoupling eliminates correlation errors, and weight compensation addresses the feature attenuation caused by reduced valid regions. A dynamic correlation model between occlusion degree and confidence level is established, maintaining a high recognition rate even in complex environments such as tree shadows, rain, and fog, significantly improving the facial payment experience.

[0040] like Figure 2 and Figure 4 As shown, in one embodiment, S2, obtaining the occlusion efficiency of the reference extraction region based on real-time occlusion information within the reference extraction region includes:

[0041] S21. Obtain multiple points to be extracted located within the baseline extraction area;

[0042] S22. Obtain multiple unextracted points located within the benchmark extraction area based on real-time occlusion information within the benchmark extraction area;

[0043] S23. Obtain the occlusion efficiency based on the proportion of the number of unextracted points to the number of points that should be extracted.

[0044] In this embodiment, it should be noted that in S21, a feature point network corresponding to the key anatomical structures within each reference extraction region is predefined. These points to be extracted constitute the basic unit of facial recognition. Taking the nasal reference region as an example, based on biological characteristics such as the three-dimensional structure of the nasal bridge, the contour of the nasal wings, and the features of the nasal tip, a fixed number of points to be extracted (e.g., ...) are predefined. Figure 4 Points 28 to 36 in the model are used to construct a complete nasal biometric model. Similarly, the eye region will pre-define points to be extracted, such as eyebrows, corners of the eyes, and eye contours (e.g., points 28 to 36 in the model). Figure 4 (18 to 27 and 37 to 48 in the model) ensures comprehensive coverage of the geometric and textural information required for region recognition. This design makes each reference region an anatomically independent recognition unit, establishing a structured foundation for subsequent occlusion quantization.

[0045] In S22, the acquireability of feature points within a region is dynamically detected using real-time occlusion information. When physical occlusion or environmental interference occurs, extraction points that cannot be effectively acquired are marked. For example, when a helmet's nose guard covers the bridge of the nose, extraction points located on the bridge of the nose may not be captured by the image sensor due to metallic reflection or physical obstruction (e.g., causing...). Figure 4 If points 28 to 31 were not captured, then points 28 to 31 are considered uncaptured points. When strong light shines directly on the right side of the face, the outline of the right side is overexposed, causing loss of outline information, and will be identified as uncaptured points (e.g., ...). Figure 4 If points 2 to 4 were not obtained, then points 2 to 4 are considered unextracted. This process employs existing image analysis model algorithms: for mottled occlusion caused by tree branch shadows, it can distinguish between partially shadowed points and completely invisible points, ensuring quantization accuracy. Essentially, this mechanism constructs a real-time mapping relationship between "effective feature points and ineffective feature points," accurately reflecting the extent of damage caused by local occlusion.

[0046] In S23, a scientific quantitative index—the occlusion effectiveness rate—is generated based on the ratio of unextracted points to points that should be extracted. This value directly characterizes the degree of degradation in the recognizability of a region: if a helmet causes 4 out of 8 points that should be extracted in the nose region to fail, the occlusion effectiveness rate is 50%; while rain and fog only blur the contour points of 3 out of 6 points that should be extracted on the left cheek, the occlusion effectiveness rate is also 50%. This ratio algorithm breaks through the limitations of traditional binary judgment (occluded / unoccluded), and realizes a gradient classification of the severity of occlusion (e.g., 30% and below is an effective region, and above 30% is a time-sensitive region).

[0047] like Figure 3 As shown, in one embodiment, obtaining the occlusion correction weight based on the number of invalid regions in S2 includes:

[0048] S24. Obtain the unit correction ratio;

[0049] S25. Obtain the occlusion correction weight based on the number of invalid regions and the unit correction ratio.

[0050] In this embodiment, it should be noted that in S24, a benchmark unit correction ratio needs to be preset. This parameter serves as the standardized unit value for weight adjustment, and its value is less than and close to 1, typically set to 0.8. Its core significance lies in defining the influence intensity coefficient of a single invalid region on the overall recognition confidence. This ratio needs to be determined through big data training (e.g., statistically calculating the benchmark magnitude of the similarity of valid regions to be compensated in historical occlusion scenarios) to ensure a scientific correlation between the degree of occlusion and the compensation intensity.

[0051] In S25, a global compensatory weight coefficient needs to be dynamically generated based on the number of invalid regions. Specifically: First, the total number of invalid regions is counted (e.g., if a user's nose and right cheek are simultaneously invalid due to wearing a helmet and strong light, the invalidity count is 2). Then, the number of invalid regions is multiplied by a unit correction ratio, then multiplied by a standardized coefficient based on the total number of regions, and finally superimposed on a base value of 1 to generate a compensation weight greater than 1. In a typical case, if tree branch shadows and a mask jointly cause the mouth and left cheek to be invalid, the impact value of both invalid regions will be calculated, and the generated compensation weight will be significantly higher than in a single-region invalidity scenario where only a mask obscures the area. This design makes the weight coefficient strictly positively correlated with the severity of occlusion—the more invalid regions there are, the stronger the similarity amplification effect on the remaining valid regions.

[0052] In one implementation, the occlusion correction weight obtained in S25 based on the number of invalid regions and the unit correction ratio is expressed as follows:

[0053] in,

[0054] A c To correct the weights for occlusion, A u X is the unit correction ratio. in This represents the number of invalid regions.

[0055] In this embodiment, it should be noted that the weight base is set to a constant of 1 in the entire expression, which represents the theoretical baseline weight in the unoccluded state. This ensures that the compensation coefficient is strictly equal to 1 when there is no invalid region. This design maintains the weight balance of the original recognition system and avoids interference with unoccluded scenes.

[0056] Furthermore, This represents the standardized allocation of unit impact; dividing the unit correction ratio by the total number of reference areas (6 in this case) essentially achieves the standardized allocation of the impact of single-area failure. For example, when a region fails due to full-face helmet obstruction, its lost feature quantity theoretically accounts for nearly 1 / 6 of the global total, while this term precisely quantifies the base amount of compensation intensity required for a single-area failure. This division operation ensures compatibility with different numbers of regions—if expanded to 7 reference areas, the denominator is automatically adjusted to 7, ensuring that the calculation of the impact of a single-area failure always matches the scale.

[0057] Furthermore, the linear cumulative compensation for the failure area multiplies the standardized single-area compensation base by the actual number of failure areas to construct a linear proportional relationship between the failure area and the compensation intensity. For example, when the tree branch shadow and the mask together cause the failure of two areas, the expression is generated as follows: Compensation value = Standardized single-area influence × 2.

[0058] This design dynamically responds to the severity of occlusion—the more occluded the area, the stronger the amplification of the similarity of the remaining effective area needs to be to counteract the difficulty in breaking through the recognition threshold caused by the reduction in the total number of features.

[0059] like Figure 4 and Figure 5 As shown, in one embodiment, S3, obtaining the i-th similarity to be processed for the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user, includes:

[0060] S31. Obtain multiple points to be extracted located within the i-th valid region;

[0061] S32. Obtain real-time relative position ratio data of multiple points to be extracted based on the real-time image information of the i-th effective area, and obtain pre-stored relative position ratio data of multiple points to be extracted based on the pre-stored image information of the i-th effective area of ​​the j-th user.

[0062] S33. Obtain the relative position similarity based on the real-time relative position ratio data and the pre-stored relative position ratio data, and use it as the i-th similarity to be processed for the j-th user.

[0063] In this embodiment, it should be noted that, similar to S21, in S31, preset key extraction points are extracted for each effective area (such as the reference area of ​​both eyes not obscured by the helmet).

[0064] In S32, spatial ratio models for real-time and pre-stored data are extracted separately. Based on the feature point set defined in S31, a geometric relationship chain between points within the region is dynamically constructed: for real-time data, feature point positions are captured by the charging pile camera, and the ratio of the distance from the corner of the eye to the center of the pupil to the width of the eye fissure is calculated; for pre-stored data, a benchmark ratio template in the user library is called, such as the ratio of the pre-stored nose tip width extraction point to the nose wing width extraction point. The key breakthrough lies in completely replacing absolute coordinates with proportional data—even if strong light causes the entire right cheek area to shift, the relative distance ratio between the cheekbone point and the corner of the mouth point remains stable. This design allows matching accuracy to be maintained through the proportional relationship of unobstructed points even under dappled occlusion caused by tree branch shadows.

[0065] Furthermore, in S33, a geometric topological similarity measurement is performed. This involves comparing the fit between the real-time proportional model and the pre-stored template: for example, detecting whether there are deviations in the pupil distance ratio within the eye region, or whether the angular relationship of the nasal bridge curvature inflection points is consistent, ultimately generating a quantified similarity value (the similarity to be processed). In a typical case, when a user wears a semi-transparent mask, the mouth region successfully matches the pre-stored data based on the lip peak distance ratio, avoiding misjudgments caused by mask texture interference in conventional methods.

[0066] In one implementation, the target similarity of the j-th user obtained in S3 based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user is represented as follows:

[0067] S tji =A c ·S wji ;in,

[0068] S tji Let A be the similarity of the i-th target to the j-th user. c To correct the weights for occlusion, S wji Let be the i-th similarity to be processed for the j-th user.

[0069] In this embodiment, it should be noted that the occlusion correction weight A is used throughout the expression. c Multiplying by the similarity to be processed essentially transforms the region feature loss into a mathematical gain. For example, when a helmet causes nose failure, A c Value increases (e.g., A) c =1.3), so that the matching result S of the eye region tji =0.5 was promoted to S tji =0.65. In summary, to actively combat the decay of total feature value caused by discarding invalid regions (e.g., discarding 30% of the region will result in the loss of corresponding features), mathematical amplification is used to ensure that the confidence contribution value of the remaining region exceeds the recognition threshold bottleneck.

[0070] Furthermore, relying solely on relative positional ratio data within this region (such as the ratio of interpupillary distance to palpebral fissure width), and completely independent of other regions (such as nasal failure not affecting eye matching accuracy), regional decoupling eliminates the cascading amplification of errors caused by occlusion disrupting the overall geometric constraints of the face in traditional methods (such as the width of the nostrils affecting the judgment of interpupillary distance).

[0071] In one implementation, the total similarity of the j-th user obtained in S3 based on the individual target similarities of the j-th user is represented as follows:

[0072] in,

[0073] S j S represents the total similarity of the j-th user. tji The similarity of the j-th user to the i-th target, m j Let be the number of target similarities for the j-th user.

[0074] In this embodiment, it should be noted that the summation operation is performed throughout the entire expression. Dynamically adapts to changes in the number of effective regions: for example, when rain or fog causes three regions to become ineffective, only the S of the remaining three effective regions is accumulated. tjiIn scenarios involving sudden occlusion, the feature aggregation level is adjusted in real time to avoid false rejections caused by insufficient total features under a fixed-weight model.

[0075] Furthermore, the weights of each region have been normalized before accumulation: S tji It inherently incorporates dual calibration, including region-independent matching results (blocking propagation) and global occlusion compensation (adversarial attenuation). In highly similar scenarios such as twin users, even if tree branch shadows occlude part of the area, the accumulated value can still amplify the accuracy of the effective area (e.g., S after eye compensation). tji =0.65) to create a similarity gap and ensure the optimal matching decision.

[0076] An image data processing system for face recognition in charging piles is also provided. The system is characterized in that it is used to implement the image data processing method for face recognition in charging piles in any of the above embodiments. The system includes: an image data acquisition module, which is used to generate reference regions for both eyes, nose, mouth, left face, right face and jaw based on the user's facial features and use them as different reference extraction regions, and to acquire real-time image information and real-time occlusion information of the user in each reference extraction region based on the charging pile image acquisition module.

[0077] The first image data processing module is used to obtain the occlusion efficiency of the reference extraction area based on the real-time occlusion information in the reference extraction area, obtain the reference extraction area with the occlusion efficiency exceeding a preset threshold and regard it as an invalid area, obtain the reference extraction area with the occlusion efficiency not exceeding the preset threshold and regard it as an effective area, and obtain the occlusion correction weight based on the number of invalid areas.

[0078] The second image data processing module is used to obtain the pre-stored image information of the i-th effective region of the j-th user in the pre-stored user information database of the charging pile, obtain the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user, obtain the i-th target similarity of the j-th user based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user, and obtain the total similarity of the j-th user based on each target similarity of the j-th user.

[0079] The third image data processing module is used to determine whether there are users in the charging pile user information database whose total similarity exceeds the consistency threshold. If so, it outputs the permission information of the user with the highest total similarity.

[0080] In one embodiment, the first image data processing module is further configured to: acquire multiple points to be extracted located within the reference extraction area; acquire multiple unextracted points located within the reference extraction area based on real-time occlusion information within the reference extraction area; and acquire the occlusion efficiency based on the proportion of the number of unextracted points to the number of points to be extracted.

[0081] In one implementation, the first image data processing module is further configured to: obtain a unit correction ratio; and obtain an occlusion correction weight based on the number of invalid regions and the unit correction ratio.

[0082] In this embodiment, it should be noted that the specific method of performing the operation of the above-mentioned image data processing system for charging pile face recognition has been described in detail in the embodiment of the image data processing method for charging pile face recognition, and will not be elaborated here.

[0083] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0084] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

[0085] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. An image data processing method for facial recognition in charging piles, characterized in that, include: Based on the user's facial features, reference regions for both eyes, nose, mouth, left face, right face, and jaw are generated and used as different reference extraction regions. Real-time image information and real-time occlusion information of the user in each reference extraction region are obtained based on the charging pile image acquisition module. Obtain multiple points to be extracted within the baseline extraction area; obtain multiple points not extracted within the baseline extraction area based on real-time occlusion information within the baseline extraction area; The occlusion efficiency is obtained by determining the proportion of the number of unextracted points to the number of points that should be extracted. The baseline extraction region with an occlusion efficiency exceeding a preset threshold is obtained and designated as an invalid region, while the baseline extraction region with an occlusion efficiency not exceeding a preset threshold is obtained and designated as a valid region. Obtain the unit correction ratio; obtain the occlusion correction weight based on the number of invalid regions and the unit correction ratio; the occlusion correction weight based on the number of invalid regions and the unit correction ratio is expressed as follows: ;in, To correct the weights for occlusion, Adjustment ratio for units, This represents the number of invalid regions. Obtain the pre-stored image information of the i-th effective region of the j-th user in the pre-stored user information database of the charging pile. Obtain the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user. Obtain the i-th target similarity of the j-th user based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user. Obtain the total similarity of the j-th user based on each target similarity of the j-th user. Determine if there are users in the charging pile user information database whose total similarity exceeds the consistency threshold. If so, output the permission information of the user with the highest total similarity.

2. The image data processing method for face recognition in charging piles according to claim 1, characterized in that, The step of obtaining the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user includes: Obtain multiple points to be extracted located within the i-th valid region; Based on the real-time image information of the i-th effective region, obtain the real-time relative position ratio data of multiple points to be extracted, and based on the pre-stored image information of the i-th effective region of the j-th user, obtain the pre-stored relative position ratio data of multiple points to be extracted. The relative position similarity is obtained based on the real-time relative position ratio data and the pre-stored relative position ratio data, and is used as the i-th similarity to be processed for the j-th user.

3. The image data processing method for face recognition in charging piles according to claim 1, characterized in that, The method of obtaining the i-th target similarity of the j-th user based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user is expressed as follows: ;in, Let i be the similarity of the j-th user to the i-th target. To correct the weights for occlusion, Let be the i-th similarity to be processed for the j-th user.

4. The image data processing method for face recognition in charging piles according to claim 1, characterized in that, The method of obtaining the total similarity of the j-th user based on the various target similarities of the j-th user is expressed as follows: ;in, Let j be the total similarity of the j-th user. The similarity of the i-th target to the j-th user. Let be the number of target similarities for the j-th user.

5. An image data processing system for facial recognition in charging piles, characterized in that, The system is used to implement the image data processing method for face recognition in charging piles as described in any one of claims 1 to 4, the system comprising: The image data acquisition module is used to generate reference regions for both eyes, nose, mouth, left face, right face and jaw based on the user's facial features and use them as different reference extraction regions. It also uses the charging pile image acquisition module to acquire real-time image information and real-time occlusion information of the user in each reference extraction region. The first image data processing module is used to obtain the occlusion efficiency of the reference extraction area based on the real-time occlusion information in the reference extraction area, obtain the reference extraction area with the occlusion efficiency exceeding a preset threshold and regard it as an invalid area, obtain the reference extraction area with the occlusion efficiency not exceeding the preset threshold and regard it as an effective area, and obtain the occlusion correction weight based on the number of invalid areas. The second image data processing module is used to obtain the pre-stored image information of the i-th effective region of the j-th user in the pre-stored user information database of the charging pile, obtain the i-th unprocessed similarity of the j-th user based on the real-time image information of the i-th effective region and the pre-stored image information of the i-th effective region of the j-th user, obtain the i-th target similarity of the j-th user based on the occlusion correction weight and the i-th unprocessed similarity of the j-th user, and obtain the total similarity of the j-th user based on each target similarity of the j-th user. The third image data processing module is used to determine whether there are users in the charging pile user information database whose total similarity exceeds the consistency threshold. If so, it outputs the permission information of the user with the highest total similarity.

6. The image data processing system for face recognition in charging piles according to claim 5, characterized in that, The first image data processing module is also used for: Obtain multiple points to be extracted located within the baseline extraction area; Based on real-time occlusion information within the baseline extraction area, obtain multiple unextracted points located within the baseline extraction area; The occlusion efficiency is obtained by determining the proportion of unextracted points to the total number of points that should be extracted.

7. The image data processing system for face recognition in charging piles according to claim 5, characterized in that, The first image data processing module is also used for: Obtain the unit correction ratio; The occlusion correction weight is obtained based on the number of invalid regions and the unit correction ratio.

Images