Vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization

Abstract

The invention relates to a vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization, and innovatively provides a multi-stage collaborative verification mechanism based on a dynamic optimization algorithm aiming at the pain points of insufficient license plate-vehicle owner binding accuracy, high equipment modification cost and the like of an existing access control system in a multi-user concurrent scene. By constructing a multi-candidate temporary storage mechanism, a progressive identity confirmation process and a self-learning dynamic window optimization algorithm, a three-level authentication system including feature pre-screening, dynamic threshold verification and identity binding locking is formed, and the problem of high-precision matching of non-motor vehicles, motor vehicles and real users in complex access scenes is effectively solved. The scheme is particularly suitable for people and vehicle dense passing scenes such as peak and valley, the people and vehicle identification matching success rate and the system anti-interference capability under mixed traffic flow are improved while low-cost upgrading of a traditional access control facility is guaranteed, and an extensible identity verification solution is provided for subdivision scenes of a smart park.

Description

Technical Field

[0001] The present invention relates to the technical field of identity recognition and data fusion, and in particular to a vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization. Background Art

[0002] With the continuous advancement of the construction of smart parks, the intelligent management requirements for non-motor vehicle (including electric bicycles) and pedestrian mixed access gates are becoming increasingly prominent. The current mainstream access management systems expose multi-dimensional technical bottlenecks in practical applications, specifically manifested as follows:

[0003] I. Bottleneck in manual verification efficiency.

[0004] The traditional manual registration mode has a two-way efficiency decay problem: during peak hours, users need to actively stay and cooperate, resulting in a 40%-60% decrease in traffic efficiency; at the same time, manual entry causes an information error rate of 15%-25%, seriously affecting the integrity of access management data. This mode can no longer meet the daily traffic demand of tens of thousands of people in modern parks.

[0005] II. Dilemma in hardware upgrade costs.

[0006] The transformation solution based on RFID / barcode has systematic defects: more than 85% of the infrastructure (including readers, gate controllers, etc.) needs to be replaced, and the physical tags have a 20% periodic damage rate, and the risk of identity forgery caused by malicious damage is difficult to avoid. More critically, the existing solutions generally have a negative correlation between traffic efficiency and recognition efficiency, and it is impossible to achieve a double-effect improvement.

[0007] III. Paradox of rigid time window.

[0008] The traditional fixed-duration matching algorithm (such as a 90±5 second window) shows three contradictions in complex scenarios:

[0009] Lack of scene adaptability: No dynamic model of traffic density is established (300+ vehicle trips per hour during morning and evening rush hours, 50 vehicle trips per hour during off-peak periods), and environmental variables such as weather are not incorporated.

[0010] Insufficient feature extraction: Ignoring the spatio-temporal coupling features of user movement trajectories, the misbinding rate surges when multiple users concur.

[0011] Imbalanced computing power consumption: Full traversal of associated users within the time window results in an extremely high CPU peak occupancy rate.

[0012] IV. Defects in the evolution of the binding mechanism.

[0013] The existing system has an ineffective cycle of "calculation-storage-verification": the reuse rate of the verified binding data is insufficient, redundant computing power is consumed seriously, and a dynamically evolving identity association model has not been established.

[0014] V. Cold start recognition dilemma.

[0015] There is a problem of data sparsity in the first inspection scenario of new license plates: when the traffic density > 50 vehicle trips per minute, due to the lack of a behavioral feature baseline, the system's misbinding probability increases exponentially, and the actual effective binding rate only maintains in the range of 5% - 20%.

[0016] Chinese Patent Application Publication No. CN114550231A discloses a multi - identity authentication management system based on the combination of face, vehicle type, and license plate recognition, including an information pre - entry module, an OCR information extraction module, a threshold matching module, an image acquisition module, an intelligent recognition module, a coding module, a data storage module, an authentication comparison module, and a user management module, which improves the accuracy and flexibility of vehicle and personnel identity authentication, reduces the comparison and authentication time, enhances the privacy and confidentiality of user identity information, simplifies the process of user entering information, and reduces the need for manual review. However, it still has the problems of difficult vehicle and personnel recognition and matching under mixed traffic flow and difficulty in establishing the corresponding relationship between vehicles and personnel.

[0017] In summary, there is an urgent need for a vehicle - person binding method to solve or partially solve the above - mentioned problems. Summary of the Invention

[0018] The purpose of the present invention is to overcome the defects of the above - mentioned existing technologies and provide a vehicle - person progressive binding method based on multi - candidate temporary storage and self - learning optimization to solve or partially solve the problems of difficult vehicle - person binding, large computational overhead for binding, a lot of repeated calculations, and low binding accuracy.

[0019] The purpose of the present invention can be achieved through the following technical solutions:

[0020] One aspect of the present invention provides a vehicle - person progressive binding method based on multi - candidate temporary storage and self - learning optimization, including the following steps:

[0021] Step S0, initialize the size of the time window corresponding to multiple time periods of the channel;

[0022] Step S1, in response to a first - type recognition event, determine whether the recognized first object is a first - time recognition or the corresponding candidate pool is empty. If so, execute Step S2; if not, execute Step S3;

[0023] Step S2, obtain at least one second - type recognition event within the time window, establish a candidate license plate - identity pair and store it in the candidate pool corresponding to the first object, initialize the confidence of the license plate - identity pair, and execute Step S1;

[0024] Step S3: Determine whether the first object is in a bound state. If not, execute Step S4. If so, match the first type of recognition event with the bound second object, update the confidence level of the bound license plate-identity pair according to the matching result, and in response to meeting the preset time window update condition, update the time window for the current period through self-learning window optimization, and execute Step S1;

[0025] Step S4: Obtain the second type of recognition events within the time window before the current first type of recognition event, match them with the candidate license plate-identity pairs in the current candidate pool, update the confidence levels of the candidate license plate-identity pairs in the candidate pool according to the matching result through a progressive confirmation strategy, and when meeting the preset locking condition, bind the first object and the second object, and execute Step S1;

[0026] Among them,

[0027] The first type of recognition event is a license plate recognition event, the second type of recognition event is an identity recognition event, the first object is a license plate, the second object is an identity, or,

[0028] The first type of recognition event is an identity recognition event, the second type of recognition event is a license plate recognition event, the first object is an identity, and the second object is a license plate.

[0029] As a preferred technical solution, in Step S4, the process of updating the confidence levels of the candidate license plate-identity pairs in the candidate pool through a progressive confirmation strategy includes:

[0030] Increase the confidence levels of the candidate license plate-identity pairs that match successfully in the candidate pool, decrease the confidence levels of the candidate license plate-identity pairs that match unsuccessfully in the candidate pool, and delete the candidate license plate-identity pairs that have failed to match continuously multiple times in the candidate pool.

[0031] As a preferred technical solution, if the match is successful, then confidence level = MIN(current value × 1.5, 1.0); if the match fails, that is, the events are not related, then confidence level = MAX(current value × 0.7, 0.1). Generate new candidate license plate-identity pairs and add them to the candidate pool, where MIN() and MAX() are respectively the minimum value and the maximum value.

[0032] As a preferred technical solution, in Step S3, the process of self-learning optimization includes:

[0033] Step S301: After meeting the preset time window update condition, obtain at least one newly added historical bound license plate-identity pair that has been successfully matched since the last time the time window update condition was met;

[0034] Step S302: For the newly added bound license plate-identity pairs that have been successfully matched in history, obtain the time difference between their newly added first type of recognition event and the second type of recognition event.

[0035] Step S303: For the 1 to N bound license plate-identity pairs that have been successfully matched in history, update the time window size corresponding to the current period based on the time differences between their respective newly added first type of recognition event and the second type of recognition event, where N is an integer.

[0036] As a preferred technical solution, in step S3, the following formula is used to update the time window corresponding to the current period:

[0037] T user =μ + k×σ

[0038] where, T user is the updated time window corresponding to the current period, μ and σ are the mean and standard deviation of the time differences between the newly added first type of recognition event and the second type of recognition event for N bound license plate-identity pairs that have been successfully matched in history, k is a parameter, and N is an integer.

[0039] As a preferred technical solution, in step S3, the following formula is used to update the time window corresponding to the current period:

[0040] T user =P_85 + k*σ s

[0041] where, T user is the updated time window corresponding to the current period, σ is the standard deviation of the time differences between the newly added first type of recognition event and the second type of recognition event for the bound license plate-identity pairs that have been successfully matched in history, P_85 is the 85th percentile in the sliding window of the time differences between the newly added first type of recognition event and the second type of recognition event for N bound license plate-identity pairs that have been successfully matched in history, k is a parameter, and N is an integer.

[0042] As a preferred technical solution, the time window update condition is that the number of successful new matches of the first object in the bound state reaches the threshold, and the locking condition is that the confidence level of the candidate license plate-identity pairs in the candidate pool reaches the threshold, or the number of consecutive successful matches reaches the threshold.

[0043] As a preferred technical solution, in step S3, the process of updating the confidence level of the bound license plate-identity pairs according to the matching results includes:

[0044] Calculate the time difference between the first type of recognition event and the most recent bound second type of recognition event, and determine whether the time difference is within the time window corresponding to the current period. If so, the match is successful, otherwise, the match fails, and the confidence of the bound license plate-identity pair is updated using the following formula. When the confidence is lower than the threshold, the binding state of the first object is released:

[0045] θ ′ =0.7+A×ln(N success +1)

[0046] Among them, θ ′ is the updated confidence, A is the parameter, N success The number of successful matches of the bound license plate-identity pair.

[0047] As a preferred technical solution, when the candidate pool exceeds a preset size, the least recently used candidate license plate-identity pair is deleted through an LRU strategy.

[0048] Another aspect of the present invention provides a vehicle-person progressive binding system based on multiple candidate temporary storage and self-learning optimization, which is used to implement the aforementioned vehicle-person progressive binding method based on multiple candidate temporary storage and self-learning optimization, and the system includes:

[0049] Authentication devices;

[0050] License plate recognition equipment;

[0051] A data monitoring module, connected to the identity verification device and the license plate recognition device respectively, for generating a first type of recognition event and a second type of recognition event;

[0052] A candidate pool manager, used for storing a candidate pool corresponding to each first object;

[0053] A progressive confirmation engine is used to match the candidate license plate-identity pairs in the current candidate pool based on the second type of recognition event in the time window corresponding to the first object before the current first type of recognition event, and update the confidence of the candidate license plate-identity pairs in the candidate pool through a progressive confirmation strategy according to the matching result;

[0054] A self-learning optimizer, configured to update the time window corresponding to the first object through self-learning window optimization in response to satisfying a preset time window update condition;

[0055] A binding relationship database, used to record the license plate-identity pairs in binding status;

[0056] A real-time matching module is used to match the first type of recognition event with the bound second object for the already bound first object, and for the unbound first object, obtain the second type of recognition event within the time window corresponding to the first object before the current first type of recognition event, and match it with the candidate license plate-identity pairs in the current candidate pool;

[0057] An access control terminal.

[0058] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0059] (1) Realize vehicle-person binding: In view of the problem that in the entrance and exit scenarios where identity verification is performed first and then license plate recognition, it is difficult to correspond people to non-motor vehicles and there is a lack of an effective way to realize vehicle-person binding, the present invention provides a vehicle-person binding method, establishing a candidate license plate-identity pair for the license plate passing through for the first time and storing it in the candidate pool corresponding to the license plate. When the same license plate passes through again, the confidence of the candidate pair is updated, and when the confidence meets the conditions, vehicle-person binding is realized.

[0060] (2) Low hardware upgrade cost: The present invention does not require modification of existing hardware devices such as identity verification devices, license plate recognition devices, and access control terminals. Compared with the transformation scheme based on RFID / barcode, it has a lower hardware upgrade cost.

[0061] (3) Good robustness in complex scenarios: Considering complex conditional scenarios such as mixed traffic of people and vehicles, when the preset time window update conditions are met, the present invention updates the time window corresponding to the license plate through self-learning window optimization, dynamically updates the time window size of the already bound license plate-identity pair, and can fully consider user behavior differences and scenario impacts, having good robustness in complex scenarios.

[0062] (4) Small calculation and storage overhead: The present invention matches candidate license plate-identity pairs based on the time window size. At the same time, through a high-confidence binding mechanism, for the already bound license plate-identity pairs, there is no need to match them in the candidate pool, and the least recently used candidate license plate-identity pairs are deleted through the LRU strategy, effectively reducing the calculation overhead and storage overhead.

[0063] (5) Have self-correction ability: For the already bound license plate-identity pairs, the present invention determines whether they match by judging whether the time difference is within the time window corresponding to the license plate. When they do not match, the confidence is reduced, and when it is lower than the threshold, the binding is cancelled to achieve self-correction. Description of the Drawings

[0064] Figure 1 It is a flowchart of the vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization in the embodiment;

[0065] Figure 2Schematic diagram of the processing procedure for the license plate recognition time when it is first recognized in the embodiment or the corresponding candidate pool is empty;

[0066] Figure 3 Schematic diagram of the migration of the progressive confirmation status in the embodiment;

[0067] Figure 4 Schematic diagram of the self - learning window optimization in the embodiment;

[0068] Figure 5 Schematic diagram of the vehicle - person progressive binding system based on multi - candidate temporary storage and self - learning optimization in the embodiment;

[0069] Figure 6 Schematic diagram of the electronic device in the embodiment. Detailed implementation manners

[0070] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0071] The terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, "a plurality" means two or more unless otherwise specifically defined.

[0072] Embodiment 1

[0073] In view of the problems existing in the foregoing prior art, this embodiment provides a vehicle - person progressive binding method based on multi - candidate temporary storage and self - learning optimization, which is applicable to scenarios such as access control / channel equipped with identity verification devices and license plate recognition devices. By using existing or added acquisition devices such as license plate recognition and RFID vehicle tags, through the construction of a trinity matching of "multi - stage binding verification", "dynamic window optimization" and "confidence locking", the binding of people and non - motor vehicles is realized at low cost. Refer to Figure 1 and Figure 2 , the method includes the following steps:

[0074] Step S0, initialize the size of the time windows corresponding to multiple time periods of the channel.

[0075] Step S1, in response to a license plate recognition event (in this embodiment, the first type of recognition event), determines whether the recognized license plate (in this embodiment, the first object) is recognized for the first time or the corresponding candidate pool is empty. If so, execute step S2, if not, execute step S3.

[0076] Step S2, obtain at least one identity recognition event (in this embodiment, the second type of recognition event) within the time window [TW, T], establish a candidate license plate-identity pair and store it in the candidate pool corresponding to the license plate, initialize the confidence of the license plate-identity pair, and execute step S1. Where T is the time of the license plate recognition event, and W is the time window size corresponding to the license plate.

[0077] In this step, the identity recognition and license plate recognition data streams are processed in parallel, and candidate matching pairs are generated through spatiotemporal alignment. The initial confidence range can be 0.3-0.6. In addition, the specific value of the confidence of the license plate-identity pair can be initialized by combining the recognition confidence level of the identity authentication device and / or the license plate recognition device (including but not limited to the intersection of union and confidence of the detection, etc.).

[0078] Step S3, determine whether the license plate is in a bound state. If not, execute step S4. If so, match the license plate recognition event with the bound identity (in this embodiment, the second object), and update the confidence of the bound license plate-identity pair according to the matching result. In response to meeting the preset time window update condition, update the time window corresponding to the license plate through self-learning window optimization, and execute step S1.

[0079] The time window update condition is that the number of successful new matches of the license plate in the binding state reaches a threshold.

[0080] The process of updating the confidence of the bound license plate-identity pair according to the matching result includes:

[0081] Calculate the time difference between the license plate recognition event and the most recent bound identity recognition event, and determine whether the time difference is within the time window corresponding to the current period. If so, the match is successful, otherwise, the match fails, and the confidence of the bound license plate-identity pair is updated using the following formula. When the confidence is lower than the threshold, the license plate is unbound:

[0082] θ ′ =0.7+A×ln(N success +1)

[0083] Among them, θ ′ is the updated confidence, A is the parameter, N success The number of successful matches of the bound license plate-identity pair.

[0084] The self-learning optimization process includes:

[0085] Step S301: After meeting the preset time window update condition, obtain at least one newly added bound license plate-identity pair that has been successfully matched historically since the last time the time window update condition was met.

[0086] Step S302: For the newly added bound license plate-identity pairs that have been successfully matched historically, obtain the time difference between their newly added license plate recognition events and identity recognition events.

[0087] Step S303: For the 1 to N bound license plate-identity pairs that have been successfully matched historically, update the time window size corresponding to the current passing period based on the time differences between their respective newly added license plate recognition events and identity recognition events, where N is an integer.

[0088] See Figure 4 , in this embodiment, the following formula is used to update the time window corresponding to the current period:

[0089] T user = P_85 + k * σ s

[0090] where, T user is the time window corresponding to the updated current period, σ is the standard deviation of the time differences between the newly added license plate recognition events and identity recognition events of the bound license plate-identity pairs that have been successfully matched historically, P_85 is the 85th percentile in the sliding window of the time differences between the newly added license plate recognition events and identity recognition events of N bound license plate-identity pairs that have been successfully matched historically, k is a parameter, and here N can be set as needed. k ∈ [1.5, 2.5]. By using the passing data of the bound pairs to autonomously optimize the window parameters in real time, the adaptability to complex scenarios is improved. The time window is set with a lower limit value and an upper limit value to ensure the balance between calculation efficiency and matching success rate. By adopting the method of sliding window statistics and using the 85th percentile as the statistic, it can have a better performance under the influence of extreme values and a better performance in complex scenarios. This characteristic makes it more suitable for scenarios with dynamically changing behavior patterns, large outliers interference, and sensitive to mis-matching and requiring fine management. For example, transportation hubs, urban road checkpoints, logistics parks, etc.

[0091] Step S4: Obtain the identity recognition events within the time window before the current license plate recognition event, match them with the candidate license plate-identity pairs in the current candidate pool, and update the confidence levels of the candidate license plate-identity pairs in the candidate pool according to the matching results through a progressive confirmation strategy. When meeting the preset locking condition, the license plate is bound to the identity, the license plate is updated to the bound state, and step S1 is executed.

[0092] Among them, the locking condition is that the confidence level of the candidate license plate-identity pairs in the candidate pool reaches the threshold, or the number of consecutive successful matches reaches the threshold. By locking the permissions for the high-confidence binding pairs and directly performing one-to-one matching in the locked state, fast verification is achieved, reducing ineffective calculations.

[0093] See Figure 3 , the process of updating the confidence level of the candidate license plate-identity pairs in the candidate pool through the progressive confirmation strategy includes:

[0094] Increase the confidence level of the candidate license plate-identity pairs that match successfully in the candidate pool, decrease the confidence level of the candidate license plate-identity pairs that match fail in the candidate pool, and delete the candidate license plate-identity pairs that have failed to match continuously for multiple times in the candidate pool.

[0095] For example, if the match is successful, then the confidence level = MIN(current value × 1.5, 1.0); if the match fails, that is, the event is not related, then the confidence level = MAX(current value × 0.7, 0.1). Generate new candidate license plate-identity pairs and add them to the candidate pool, where MIN() and MAX() are used to take the minimum value and the maximum value respectively. It should be noted that the coefficients in the confidence level update formula are for reference only.

[0096] Through the above solution, three-stage binding verification can be achieved:

[0097] Primary screening: For the license plates that appear for the first time, temporarily store multiple candidate license plate-identity pairs and perform rough matching based on the dynamic time window. At this time, the confidence level of the license plate-identity pairs is basically between 0.3 and 0.6.

[0098] Secondary verification: Progressive dynamic confirmation, gradually eliminating mis-matches through subsequent events, and combining historical behavior patterns to achieve verification by updating the confidence level of the candidate license plate-identity pairs.

[0099] Final confirmation: Cross-cycle continuity verification, achieving binding when the locking condition is met, thereby effectively reducing the subsequent calculation amount.

[0100] This method reduces more than half of the repeated calculations through the binding locking mechanism, effectively improving the calculation efficiency. The mis-match rate is reduced to an extremely low level through three-stage verification, optimizing the matching accuracy. Through dynamic optimization of the window parameters, the response speed of detection is improved, and the adaptive ability is optimized. At the same time, this method can be widely compatible with a variety of access control devices.

[0101] To further illustrate the implementation process of this method, the following uses several examples to illustrate its processing process.

[0102] Example 1: Multi-user concurrent processing.

[0103] 1. Scenario process:

[0104] 08:00:00 User A swipes the card;

[0105] 08:00:15 User B swipes the card;

[0106] 08:00:40 License plate X is recognized;

[0107] Current window value of license plate X: 50 seconds

[0108] 2. Processing flow:

[0109] Generate candidate pairs: According to the initialized window of 60 seconds, within the window, A-X (time difference 40 seconds, confidence 0.4), B-X (time difference 25 seconds, confidence 0.6).

[0110] Subsequent events:

[0111] On a certain day later at 08:02:10, A swipes the card again, and X is recognized at 08:02:30 (time difference 20 seconds, confidence 0.6 → 0.9);

[0112] There is no subsequent related event for B (confidence 0.6 → 0.3);

[0113] The system confirms the binding of A-X and updates the latest window to 25 seconds (85th percentile of the time difference) + 5 = 30 seconds.

[0114] Example 2: Dynamic window optimization in the self-learning process.

[0115] 1. Scenario: The historical matching time differences of the already bound users A, B, C, and D at the south gate are [50s, 52s, 55s, 58s], and at this time N = 4..

[0116] 2. Calculation:

[0117] Mean calculation: μ = 53.75s,

[0118] Standard deviation calculation: σ = 3.3s

[0119] Optimize the new window of the bound license plate - identity pair corresponding to the south gate = 53.75 + 2 × 3.3 ≈ 60s.

[0120] Through dynamic window optimization in the self-learning process, while reducing the matching time difference, the gate mis-matching rate is reduced.

[0121] Example 3: Optimization of new license plate binding.

[0122] 1. The first passage of the new license plate Y:

[0123] Generate candidate pairs with 3 users (confidence 0.3 - 0.5);

[0124] There is no subsequent confirmation event.

[0125] 2. Second Pass:

[0126] 15 seconds behind the real user C (confidence level 0.6);

[0127] Subsequent consecutive 2 successful matches (confidence level 0.6 → 0.9);

[0128] The recommended window for completing the binding record is 15 + 5 = 20 seconds.

[0129] This method solves the long - standing contradiction between the accuracy and efficiency of vehicle - person matching through a dynamic feedback mechanism and a multi - stage verification system, and has significant industrial application value. The measured data shows that after deployment at the entrance of a certain university, within two weeks, more than 200,000 pieces of data were obtained, more than 10,000 vehicle - person trips were successfully bound, with accuracy, high efficiency, and low cost.

[0130] Example 2

[0131] Compared with Example 1, this example provides another vehicle - person progressive binding method based on multi - candidate caching and self - learning optimization. The difference is that in step S3, the following formula is used to update the time window corresponding to the current period:

[0132] T user = μ + k × σ

[0133] where, T user is the updated time window corresponding to the current period, μ and σ are the mean and standard deviation of the time differences between the newly added license plate recognition events and identity recognition events of N pairs of historically successfully matched license plate - identity pairs, k is a parameter, and k = 2.

[0134] Compared with the method of Example 1, this example does not require sliding window calculation, the parameters are fixed, and it is applicable to scenarios with stable historical behaviors, fewer outliers, and low real - time requirements, such as the entrance of a community parking lot, a closed park, etc.

[0135] Example 3

[0136] This example provides another vehicle - person progressive binding method based on multi - candidate caching and self - learning optimization. Compared with Example 1, the difference is that the first type of recognition event in this example is the identity recognition event, the second type of recognition event is the license plate recognition event, the first object is the identity, the second object is the license plate, that is, the first type of recognition time and the second type of recognition event are swapped, and the first identity and the second identity are swapped.

[0137] Example 4

[0138] This embodiment provides another vehicle-person progressive binding method based on multi-candidate caching and self-learning optimization. Compared with Embodiment 2, the difference is that the first type of recognition event in this embodiment is an identity recognition event, and the second type of recognition event is a license plate recognition event, that is, the first type of recognition time and the second type of recognition event are swapped with each other.

[0139] Embodiment 5

[0140] Based on the foregoing embodiments, refer to Figure 5 , this embodiment provides a vehicle-person progressive binding system based on multi-candidate caching and self-learning optimization, which is used to implement the vehicle-person progressive binding method based on multi-candidate caching and self-learning optimization in any of the foregoing embodiments. The system includes:

[0141] (1) Identity authentication device;

[0142] (2) License plate recognition device;

[0143] (3) Data monitoring module, which is respectively connected to the identity authentication device and the license plate recognition device, and is used to obtain identity and license plate recognition data in real time, and generate the first type of recognition event and the second type of recognition event;

[0144] (4) Candidate pool manager, which is used to store the candidate pool corresponding to each first object;

[0145] (5) Progressive confirmation engine, which is used to match the second type of recognition event within the time window corresponding to the first object before the current first type of recognition event with the candidate license plate-identity pairs in the current candidate pool, and update the confidence of the candidate license plate-identity pairs in the candidate pool through a progressive confirmation strategy according to the matching result;

[0146] (6) Self-learning optimizer, which is used to update the time window corresponding to the first object through self-learning window optimization in response to meeting the preset time window update condition;

[0147] (7) Binding relationship database, which is used to record the license plate-identity pairs in the binding state;

[0148] (8) Real-time matching module, which is used to match the first type of recognition event with the bound second object for the already bound first object, and for the unbound first object, obtain the second type of recognition event within the time window corresponding to the first object before the current first type of recognition event, and match it with the candidate license plate-identity pairs in the current candidate pool;

[0149] (9) Access control terminal.

[0150] The present invention belongs to the field of intelligent transportation and data fusion technology, focusing on the technical requirements for accurate identification of vehicles and personnel in scenarios of smart parks, communities and campuses. Aiming at the pain points of the existing access control system in multi-user concurrent scenarios, such as insufficient accuracy of license plate-owner binding and high equipment modification costs, a multi-stage collaborative verification mechanism based on a dynamic optimization algorithm is proposed. By constructing a multi-candidate temporary storage mechanism, a progressive identity confirmation process and a self-learning dynamic window optimization algorithm, a three-level authentication system including feature pre-screening, dynamic threshold verification and identity binding lock is formed, which effectively solves the problem of high-precision matching of non-motorized vehicles (including electric bicycles), motor vehicles and real users in complex entry and exit scenarios. This solution is particularly suitable for scenarios with dense traffic of people and vehicles such as peaks and valleys. While ensuring the low-cost upgrade of traditional access control facilities, it significantly improves the success rate of human-vehicle identification and matching under mixed traffic flows and the system's anti-interference ability, providing a scalable identity verification solution for smart park segmentation scenarios.

[0151] Example 4

[0152] Based on the above embodiments, see Figure 6 This embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization as described in Example 1 or Example 2.

[0153] like Figure 6 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to the software implementation, the present invention does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0154] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0155] A computer-readable medium includes permanent and non-permanent, removable and non-removable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette tapes, disk storage, or other magnetic storage devices, or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0156] As described above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or replacements, and these modifications or replacements should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization, characterized in that It includes the following steps: Step S0, initialize the size of the time window corresponding to multiple time periods of the channel; Step S1, in response to a first type of recognition event, determine whether the recognized first object is recognized for the first time or the corresponding candidate pool is empty. If so, execute Step S2; if not, execute Step S3; Step S2, obtain at least one second type of recognition event within the time window, establish a candidate license plate-identity pair and store it in the candidate pool corresponding to the first object, initialize the confidence level of the license plate-identity pair, and execute Step S1; Step S3, determine whether the first object is in a bound state. If not, execute Step S4; if so, match the first type of recognition event with the bound second object, update the confidence level of the bound license plate-identity pair according to the matching result, and in response to meeting the preset time window update condition, update the time window of the current time period through self-learning window optimization, and execute Step S1; Step S4, obtain the second type of recognition event within the time window before the current first type of recognition event, match it with the candidate license plate-identity pair in the current candidate pool, and update the confidence level of the candidate license plate-identity pair in the candidate pool according to the matching result through a progressive confirmation strategy. When meeting the preset locking condition, bind the first object and the second object, and execute Step S1; Wherein, the first type of recognition event is a license plate recognition event, the second type of recognition event is an identity recognition event, the first object is a license plate, the second object is an identity, or, the first type of recognition event is an identity recognition event, the second type of recognition event is a license plate recognition event, the first object is an identity, and the second object is a license plate.

2. The vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, wherein In the said Step S4, the process of updating the confidence level of the candidate license plate-identity pair in the candidate pool through a progressive confirmation strategy includes: Increase the confidence level of the candidate license plate-identity pair that has been successfully matched in the candidate pool, decrease the confidence level of the candidate license plate-identity pair that has failed to match in the candidate pool, and delete the candidate license plate-identity pair that has failed to match continuously for multiple times in the candidate pool.

3. The vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 2, wherein, If the match is successful, then confidence level = MIN(current value × 1.5, 1.0); if the match fails, that is, the event is not related, then confidence level = MAX(current value × 0.7, 0.1). Generate a new candidate license plate-identity pair and add it to the candidate pool, where MIN() and MAX() are respectively used to take the minimum value and the maximum value.

4. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, characterized in that, In the said Step S3, the process of self-learning optimization includes: Step S301, after meeting the preset time window update condition, obtain at least one newly added bound license plate-identity pair that has been successfully matched since the last time the time window update condition was met; Step S302, for the newly added bound license plate-identity pair that has been successfully matched, obtain the time difference between its newly added first type of recognition event and the second type of recognition event; Step S303, for the 1 to N bound license plate-identity pairs that have been successfully matched in history, update the size of the time window corresponding to the current time period based on the time difference between their respective newly added first type of recognition event and the second type of recognition event, where N is an integer.

5. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, characterized in that In the said Step S3, the following formula is used to update the time window corresponding to the current time period: T user = μ + k × σ where T user is the time window corresponding to the updated current time period, μ and σ are the mean and standard deviation of the time differences between the first type of recognition event and the second type of recognition event newly added for N pairs of license plate-identity pairs that have been successfully matched in history, k is a parameter, and N is an integer.

6. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, characterized in that In step S3, the time window corresponding to the current period is updated using the following formula: T user = P_85 + k * σ s Among them, T user is the time window corresponding to the updated current period, σ is the standard deviation of the time differences between the newly added first type of recognition events and the second type of recognition events for the license plate-identity pairs that have been successfully matched historically, P_85 is the 85th percentile in the sliding window of the time differences between the newly added first type of recognition events and the second type of recognition events for N pairs of license plate-identity pairs that have been successfully matched historically, k is a parameter, and N is an integer.

7. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, characterized in that The time window update condition is that the number of newly added successful matches of the first object in the bound state reaches a threshold, and the locking condition is that the confidence level of the candidate license plate-identity pair in the candidate pool reaches a threshold, or the number of consecutive successful matches reaches a threshold.

8. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, characterized in that, In step S3, the process of updating the confidence level of the bound license plate-identity pair according to the matching result includes: Calculating the time difference between the first type of recognition event and the most recent bound second type of recognition event, and determining whether the time difference is within the time window corresponding to the current period. If so, the match is successful; if not, the match is failed. The confidence level of the bound license plate-identity pair is updated using the following formula, and when the confidence level is lower than the threshold, the binding state of the first object is released: θ ′ = 0.7 + A × ln(N success + 1) where θ ′ is the updated confidence level, A is a parameter, and N success is the number of successful matches of the bound license plate-identity pairs.

9. A vehicle-person progressive binding method based on multi-candidate temporary storage and self-learning optimization according to claim 1, characterized in that When the candidate pool exceeds the preset size, the least recently used candidate license plate-identity pair is deleted through the LRU strategy.

10. A vehicle-person progressive binding system based on multi-candidate temporary storage and self-learning optimization, characterized in that For implementing the vehicle-person progressive binding method based on multi-candidate staging and self-learning optimization as described in any one of claims 1-9, the system includes: An identity verification device; A license plate recognition device; A data monitoring module, connected to the identity verification device and the license plate recognition device respectively, for generating the first type of recognition event and the second type of recognition event; A candidate pool manager for storing the candidate pool corresponding to each first object; A progressive confirmation engine for matching the second type of recognition event within the time window corresponding to the first object before the current first type of recognition event with the candidate license plate-identity pair in the current candidate pool, and updating the confidence level of the candidate license plate-identity pair in the candidate pool through a progressive confirmation strategy according to the matching result; A self-learning optimizer for updating the time window corresponding to the first object through self-learning window optimization in response to meeting the preset time window update condition; A binding relationship database for recording the license plate-identity pairs in the bound state; A real-time matching module for matching the first type of recognition event with the bound second object for the already bound first object, and for the unbound first object, obtaining the second type of recognition event within the time window corresponding to the first object before the current first type of recognition event, and matching it with the candidate license plate-identity pair in the current candidate pool; An access control terminal.

Images