A parking management and control method, system and device based on cognitive arbitration
By employing a cognitive arbitration-based approach, and utilizing multiple random forward inferences and dual-path recoding, the problem of misjudgment in parking management under complex scenarios is solved. This approach achieves efficient and interpretable vehicle identification and arbitration, reduces the misjudgment rate, and improves the robustness and efficiency of the system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 泰安市东信智联信息科技有限公司
- Filing Date
- 2026-04-15
- Publication Date
- 2026-08-04
AI Technical Summary
Existing parking management systems are prone to high-confidence misjudgments in complex scenarios such as obstruction, dirt, backlighting, heavy rain, following vehicles to break through checkpoints, and license plate swapping. In particular, when there are many vehicles of the same model and color, it is difficult to accurately distinguish their identities. Furthermore, there is a lack of hierarchical screening and interpretable arbitration mechanisms for multiple candidate competing scenarios. Cloud arbitration lacks semantic evidence, resulting in network overhead and computational delays.
By acquiring images of currently observed vehicles and performing multiple random forward inferences, a cognitive uncertainty score is calculated. Combined with spatiotemporal reachability filtering, memory strength value, and competition intensity index, an arbitration priority index is constructed. Dual-path recoding and semantic arbitration are performed to generate results for same vehicle, different vehicle, or pending results. Finally, the exit congestion index is used to select whether to allow passage, physically intercept, or virtually block the vehicle.
It can reliably identify vehicle identities in multi-candidate confusion scenarios, reduce the false judgment rate, improve the interpretability and robustness of arbitration, reduce invalid cloud calls, and balance traffic efficiency with the effectiveness of violation handling.
Smart Images
Figure QLYQS_3 
Figure QLYQS_26 
Figure QLYQS_33
Abstract
Description
Technical Field
[0001] This invention belongs to the field of parking management technology, specifically relating to a parking control method, system, and device based on cognitive arbitration. Background Technology
[0002] Existing parking management systems typically rely on license plate recognition or vehicle appearance comparison to manage vehicle entry, exit, and billing.
[0003] However, in actual parking lot operations, the following technical problems exist: In complex scenarios such as obstruction, dirt, backlighting, heavy rain, vehicles attempting to evade checkpoints, and license plate swapping, relying solely on license plates or conventional visual similarity comparisons can easily lead to high-confidence misjudgments. Especially when there are many vehicles of the same model and color in the parking lot, it is difficult to accurately distinguish the identities of different vehicles based solely on appearance features.
[0004] Furthermore, existing solutions mostly focus on supplementary judgments of outcomes, lacking tiered screening, unique candidate resolution, and interpretable arbitration mechanisms for scenarios with multiple competing candidates. When multiple historical vehicles are highly similar to the currently observed vehicle, existing solutions cannot converge a unique identity conclusion from multiple candidate identities, leading to a multi-candidate dilemma. Meanwhile, cloud-based arbitration lacks interpretability, fails to provide semantic evidence, and the uploading of a large number of low-value samples also causes network overhead and computational latency. Summary of the Invention
[0005] The purpose of this invention is to provide a parking management method based on cognitive arbitration, comprising the following steps: S1. Obtain the image of the currently observed vehicle, perform multiple random forward inferences on the image of the currently observed vehicle to obtain a set of feature vectors, calculate the mean of the feature vectors, and calculate the cognitive uncertainty score of the currently observed vehicle based on the difference between each feature vector and the mean of the feature vectors.
[0006] S2. Based on the observation time, observation location, and parking lot topology of the currently observed vehicle, the vehicles in the historical vehicle set are determined to be reachable. Vehicles that meet the reachability criteria are selected as candidate vehicles. The candidate vehicles are sorted from high to low according to the similarity of the coarse matching. The candidate vehicles with the highest similarity are selected to form the disputed candidate set. The memory strength value is calculated for each candidate vehicle in the disputed candidate set. The competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles.
[0007] The arbitration priority index is obtained by weighted summation of the standard deviations of the cognitive uncertainty score, the competition intensity index, and the memory intensity value. When the arbitration priority index exceeds the priority index threshold, S3 is executed.
[0008] S3. Construct disputed vehicle pairs for the current observed vehicle and each candidate vehicle in the disputed candidate set. Extract the stable common region and exclusive difference region of each disputed vehicle and map them as common mask and difference mask respectively. Perform dual-path recoding on the feature vectors of the current observed vehicle and each candidate vehicle based on the common mask and difference mask respectively. Calculate the common path drift and difference path drift corresponding to each candidate vehicle. Combine the memory strength value of each candidate vehicle to calculate the candidate decision score. Select the candidate vehicle with the highest candidate decision score as the final candidate vehicle. Based on the common path drift, difference path drift and candidate decision score of the final candidate vehicle, obtain the same vehicle result, different vehicle result or undetermined result.
[0009] S4. If the same vehicle is detected, issue a release command; If an abnormal vehicle is detected, an interception command is issued; If a pending result is obtained, a virtual blocking command is issued.
[0010] In S2, based on the observation time, observation location, and parking lot topology of the currently observed vehicle, reachability determination is performed on the vehicles in the historical vehicle set, specifically as follows: Vehicles in the historical vehicle collection reachability determination result Determined according to the following formula: , in, Vehicles in the historical vehicle collection The reachability determination result, Vehicles in the historical vehicle collection The length of the parking lot topological path from the last observation location to the current observation location. Vehicles in the historical vehicle collection The time interval between the last observation and the current observation. Set a reasonable maximum speed for vehicles to pass through the parking lot; Only when At that time, the corresponding historical vehicles will be included in the candidate set as candidate vehicles.
[0011] The memory strength value is calculated in S2 as follows: , in, This represents the memory strength value. and is the weighting coefficient for memory strength values, and , This is the time decay coefficient; For candidate vehicles The time interval; For candidate vehicles Trajectory continuity score, For candidate vehicles Historical appearance stability score; Candidate vehicles Trajectory continuity score Calculate according to the following formula: , in, For candidate vehicles The number of valid trajectory segments that satisfy topological adjacency and temporal continuity constraints within a preset time window; For candidate vehicles The total number of all trajectory segments within the preset time window; Candidate vehicles Historical appearance stability score Calculate according to the following formula: , in, For candidate vehicles The number of historical observations; For candidate vehicles No. The eigenvectors corresponding to each historical observation; For candidate vehicles The historical average eigenvector; This is the feature distance function.
[0012] In S3, based on the common path drift amount, different path drift amount, and candidate decision score of the final candidate vehicles, the results for the same vehicle, different vehicles, or pending results are obtained, specifically: When the final candidate vehicles meet the common path drift requirement Common path determination threshold, differential path drift The threshold for determining the difference path, and the candidate decision score. When the candidate decision threshold is reached, the result is determined to be the same vehicle. When all candidate vehicles do not meet the same vehicle determination criteria, and at least one of the candidate vehicles meets the difference path drift criteria. When the difference path determination threshold is reached, the result is determined to be an abnormal vehicle. When the highest score of a candidate decision is less than the candidate decision threshold, it is determined to be a pending result.
[0013] In S3, the common path drift and differential path drift of each candidate vehicle are calculated, and the candidate decision score is calculated by combining the memory strength value of each candidate vehicle. Specifically: Candidate decision score Calculate according to the following formula: , in, and The weighting coefficients for the candidate decision scores. The memory strength value of the i-th candidate vehicle, and the common path drift. and difference path drift Calculate according to the following formulas respectively: , , in, This is the feature vector of the currently observed vehicle. For the first Feature vectors of candidate vehicles and These are the current observed vehicle feature vectors enhanced by commonality masks and the first... Each candidate vehicle feature vector and These are the current observed vehicle feature vector and the i-th candidate vehicle feature vector, respectively, after difference masking enhancement. This is the feature distance function.
[0014] S1 calculates the cognitive uncertainty score of the currently observed vehicle, specifically as follows: Assume the current observed vehicle image is... The set of feature vectors is obtained through random forward inference. The mean of the eigenvectors is Cognitive uncertainty score Calculate according to the following formula: , in, For the number of random forward inferences, For the first The feature vector obtained by random forward inference. for The mean vector of the eigenvectors corresponding to each random forward inference. This indicates element-wise multiplication. Let U represent the 2-norm, and U be the cognitive uncertainty score.
[0015] In S2, the competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles. Specifically: , The highest matching score between the currently observed vehicle and the target number of candidate vehicles. The matching score between the currently observed vehicle and candidate vehicle i. To prevent a constant with a denominator of zero, Q is the competition intensity index.
[0016] If an undetermined result is obtained in S4, calculate the congestion index of the current exit. When the congestion index The parking lot exit is considered congested when the congestion threshold is reached. The parking lot exit is considered to be in a non-congested state when the congestion threshold is reached. When the vehicle is not in a congested state, an interception command is issued to perform a physical interception and further verify whether it is the same vehicle or a different vehicle. When the traffic is congested, virtual blocking is implemented; during the virtual blocking process, vehicles with pending results are physically released.
[0017] A parking management system based on cognitive arbitration, used to implement a parking management method based on cognitive arbitration, includes: The cognitive uncertainty score calculation module acquires the image of the currently observed vehicle, performs multiple random forward inferences on the image to obtain a set of feature vectors, calculates the mean of the feature vectors, and calculates the cognitive uncertainty score of the currently observed vehicle based on the difference between each feature vector and the mean of the feature vectors.
[0018] The competition intensity index calculation module performs reachability determination on vehicles in the historical vehicle set based on the observation time, observation location, and parking lot topology of the currently observed vehicle. Vehicles that meet the reachability determination are selected as candidate vehicles. The candidate vehicles are sorted from high to low according to the similarity of the coarse matching. The candidate vehicles with the highest similarity are selected to form the disputed candidate set. The memory intensity value is calculated for each candidate vehicle in the disputed candidate set. The competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles.
[0019] The arbitration priority index is obtained by weighted summation of the standard deviations of cognitive uncertainty score, competition intensity index and memory intensity value. When the arbitration priority index exceeds the priority index threshold, the execution result is determined.
[0020] The result determination module constructs disputed vehicle pairs for the currently observed vehicle and each candidate vehicle in the disputed candidate set. It extracts the stable common region and exclusive difference region of each disputed vehicle and maps them to commonality masks and difference masks, respectively. Based on the commonality mask and difference mask, it performs dual-path recoding on the feature vectors of the currently observed vehicle and each candidate vehicle, calculates the common path drift and difference path drift corresponding to each candidate vehicle, and calculates the candidate decision score by combining the memory strength value of each candidate vehicle. The candidate vehicle with the highest candidate decision score is selected as the final candidate vehicle. Based on the common path drift, difference path drift and candidate decision score of the final candidate vehicle, it obtains the same vehicle result, different vehicle result, or pending result.
[0021] The processing module issues a release command if it receives a result indicating the vehicle is in the same vehicle; an intercept command if it receives a result indicating the vehicle is in a different vehicle; and a virtual block command if it receives a pending result.
[0022] A parking management device based on cognitive arbitration includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement a parking management method based on cognitive arbitration. Compared with the prior art, the present invention has the following beneficial effects: 1. By using spatiotemporal reachability filtering, candidate truncation, memory strength value, and competition strength index, it can more stably identify identity disputes in multi-candidate confusion scenarios, effectively solve the problem of misidentification in scenarios with the same vehicle model, the same color, and partial occlusion, and reduce the false judgment rate.
[0023] 2. By using the arbitration priority index, cloud semantic arbitration is triggered only when the uncertainty on the edge side is high and the competition among candidates is obvious, which can reduce invalid calls and improve efficiency.
[0024] 3. By using a dual-hint mechanism of common semantic cue words and differential semantic cue words, as well as a joint decision mechanism of common path drift, differential path drift and candidate decision score, two types of evidence supporting the same vehicle and different vehicles can be verified simultaneously, and a unique final candidate can be determined among multiple candidate vehicles, thereby improving the interpretability and robustness of arbitration in complex scenarios.
[0025] 4. The final arbitration result of S3 can be directly used as the trigger condition for S4 graded intervention. Combined with the exit congestion index, the decision can be made to allow passage, physically block or virtually block passage, which can balance traffic efficiency and the effectiveness of handling violations. Detailed Implementation
[0026] To further understand the content of this invention, the invention will be described in detail with reference to the embodiments.
[0027] This invention relates to a parking management method based on cognitive arbitration, which includes the following steps: S1. Obtain the image of the currently observed vehicle, perform multiple random forward inferences on the image of the currently observed vehicle to obtain a set of feature vectors, calculate the mean of the feature vectors, and calculate the cognitive uncertainty score of the currently observed vehicle based on the difference between each feature vector and the mean of the feature vectors.
[0028] Specifically, the system acquires images of the currently observed vehicles, extracts vehicle representation features, and uses stochastic forward inference to obtain cognitive uncertainty scores.
[0029] Suppose that the same vehicle image is subjected to T random forward inferences to obtain a set of feature vectors. , feature mean vector Calculate according to the following formula: , Cognitive uncertainty score Calculate according to the following formula: , in, For the number of random forward inferences, For the first The feature vector obtained by random forward inference. for The mean vector of the eigenvectors corresponding to each random forward inference. This indicates element-wise multiplication. Let U represent the 2-norm, and U be the cognitive uncertainty score.
[0030] When the cognitive uncertainty score is less than the deterministic threshold, the local judgment process is directly initiated, and the current vehicle is allowed to pass. Specifically, local judgment refers to the determination of vehicle identity independently by the edge perception agent and the spatiotemporal memory agent at the edge, without triggering the intervention of the cloud semantic audit agent. When the cognitive uncertainty score is greater than or equal to the deterministic threshold, S2 is executed. By performing multiple random forward inferences on the currently observed vehicle image and statistically analyzing the feature dispersion (cognitive uncertainty score), the stability of the currently observed vehicle image can be measured in an edge manner. When U is large, it indicates that the current sample may have occlusion, blurring, extreme lighting, or out-of-distribution conditions, and should proceed to the subsequent disputed identity arbitration process.
[0031] S2. Based on the observation time, observation location, and parking lot topology of the currently observed vehicle, reachability determination is performed on the vehicles in the historical vehicle set. Vehicles that meet the reachability determination are selected as candidate vehicles. The candidate vehicles are sorted from high to low according to the similarity of the coarse matching. The candidate vehicles with the highest similarity are selected to form the dispute candidate set. The memory strength value is calculated for each candidate vehicle in the dispute candidate set. The competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles. The arbitration priority index is obtained by weighted summation based on the standard deviation of the cognitive uncertainty score, the competition intensity index, and the memory strength value. When the arbitration priority index exceeds the priority index threshold, S3 is executed.
[0032] Specifically, based on the observation time, observation location, and parking lot topology of the current observed vehicle, candidate vehicles that are physically reachable from the historical vehicle set are selected to form a candidate set; further, the current observation location and observation time of the current observed vehicle are obtained, and vehicle nodes that have appeared within a preset time window are retrieved from the historical vehicle database as the historical vehicle set.
[0033] For vehicles in the historical vehicle collection reachability determination result Determined according to the following formula: , in, Vehicles in the historical vehicle collection The reachability determination result, Vehicles in the historical vehicle collection The length of the parking lot topological path from the last observation location to the current observation location. Vehicles in the historical vehicle collection The time interval between the last observation and the current observation. Set a reasonable maximum speed for vehicles to pass through the parking lot; Only when At that time, the corresponding historical vehicles will be included in the candidate set as candidate vehicles.
[0034] This formula utilizes the condition that vehicle movement within a parking lot must satisfy topological path and maximum safe speed constraints to perform reachability screening on historical vehicles, thereby eliminating candidate vehicles that are physically impossible to reach the current observation point.
[0035] Subsequently, the candidate vehicles are initially sorted based on the coarse matching similarity, and only the candidate vehicles with the highest similarity to the target number are retained as the dispute candidate set to reduce computational complexity and improve processing efficiency. Then, the memory strength value is calculated for each candidate vehicle of the target number, and the competition intensity index is calculated based on the matching score between the currently observed vehicle and the candidate vehicles of the target number.
[0036] Specifically, the memory strength value of candidate vehicle i Calculate according to the following formula: , in, This represents the memory strength value. and is the weighting coefficient for memory strength values, and , This is the time decay coefficient; For candidate vehicles The time interval; For candidate vehicles Trajectory continuity score, For candidate vehicles Historical appearance stability score; Candidate vehicles Trajectory continuity score Calculate according to the following formula: , in, For candidate vehicles The number of valid trajectory segments that satisfy topological adjacency and temporal continuity constraints within a preset time window; For candidate vehicles The total number of all trajectory segments within the preset time window; Candidate vehicles Historical appearance stability score Calculate according to the following formula: , in, For candidate vehicles The number of historical observations; For candidate vehicles No. The eigenvectors corresponding to each historical observation; For candidate vehicles The historical average eigenvector; This is the feature distance function.
[0037] By combining time decay term, trajectory continuity score and historical appearance stability score, the historical vehicle identity can be transformed from static record to dynamic credibility, and in multi-candidate scenarios, candidate vehicles that have recently appeared, have coherent trajectories and stable appearances can be prioritized.
[0038] The competition intensity index is calculated based on the matching score between the currently observed vehicles and the target number of candidate vehicles, specifically: , The highest matching score between the currently observed vehicle and the target number of candidate vehicles. The matching score between the currently observed vehicle and candidate vehicle i. To prevent a constant with a denominator of zero, Q is the competition intensity index.
[0039] When the highest matching score is close to the second highest matching score, it indicates that multiple candidate vehicles are similar to the currently observed vehicle, and the identity competition is obvious, so Q increases; when the difference between the two is large, it indicates that the best candidate is more prominent, so Q decreases. This formula is used to quantify the degree of competition among multiple candidates and avoid the system directly confirming the identity based solely on the single highest similarity.
[0040] By combining the cognitive uncertainty score, the competition intensity index, and the standard deviation of the target number of candidate vehicle memory intensity values, an arbitration priority index is generated. Calculate according to the following formula: , in, The standard deviation of the memory strength values of the target number of candidate vehicles. and These are the weighting coefficients.
[0041] This formula integrates the uncertainty of the edge model, the degree of candidate competition, and the distribution of candidate historical credibility into a unified index. When P is large, it indicates that the edge side is unable to make a stable identity determination, and cloud semantic arbitration should be triggered.
[0042] When P is not greater than the priority index threshold, the edge side has high confidence in recognizing the currently observed vehicle, the competition is not fierce, and the candidate differences are obvious. It has the conditions to make independent identity determinations. The local similarity ranking result is directly used as the identity determination result of the currently observed vehicle. That is, the one with the highest similarity is the currently observed vehicle. The local determination output is not a three-class result of same vehicle / different vehicle / pending, but directly specifies the identity of the specific candidate vehicle. Specifically, the candidate vehicle with the highest similarity is identified as the identity of the currently observed vehicle.
[0043] When the arbitration priority index exceeds the priority index threshold, an arbitration package is generated and sent to the cloud semantic audit agent to execute S3.
[0044] S3. Construct disputed vehicle pairs for the current observed vehicle and each candidate vehicle in the disputed candidate set. Extract the stable common region and exclusive difference region of each disputed vehicle and map them as common mask and difference mask respectively. Perform dual-path recoding on the feature vectors of the current observed vehicle and each candidate vehicle based on the common mask and difference mask respectively. Calculate the common path drift and difference path drift corresponding to each candidate vehicle. Combine the memory strength value of each candidate vehicle to calculate the candidate decision score. Select the candidate vehicle with the highest candidate decision score as the final candidate vehicle. Based on the common path drift, difference path drift and candidate decision score of the final candidate vehicle, obtain the same vehicle result, different vehicle result or undetermined result.
[0045] Specifically, after receiving the arbitration packet, the cloud-based semantic auditing agent constructs disputed vehicle pairs for the currently observed vehicle and the target number of candidate vehicles in the disputed candidate set. For each disputed vehicle pair, common semantic prompts and differential semantic prompts are constructed to simultaneously extract stable common regions that support that the two vehicles are the same vehicle and exclusive differential regions that support that the two vehicles are different vehicles.
[0046] The common semantic prompts use the following template: "Please compare image A and image B. If they belong to the same vehicle, please point out the area whose appearance remains stable under different shooting angles, lighting and occlusion conditions, and output the location description and feature description of the area." The semantic prompts for differences use the following template: "Please compare image A and image B. If they do not belong to the same vehicle, please indicate the appearance difference region that best excludes them from being the same vehicle, and output the location description and difference description of that region." Stable common regions are mapped to common masks, and exclusive difference regions are mapped to difference masks, which are then fed back to the edge-aware agent. The edge-aware agent performs dual-path recoding on the original features of the currently observed vehicle and each candidate vehicle based on the common and difference masks, calculates the common path drift and difference path drift for each candidate vehicle, and combines this with the memory strength values obtained in step S2 to calculate the candidate decision score. Specifically: Candidate decision score Calculate according to the following formula: , in, and The weighting coefficients for the candidate decision scores. The memory strength value of the i-th candidate vehicle, and the common path drift. and difference path drift Calculate according to the following formulas respectively: , , in, This is the feature vector of the currently observed vehicle. For the first Feature vectors of candidate vehicles and These are the current observed vehicle feature vectors enhanced by commonality masks and the first... Each candidate vehicle feature vector and These are the current observed vehicle feature vector and the i-th candidate vehicle feature vector, respectively, after difference masking enhancement. This is the feature distance function.
[0047] The candidate vehicle with the highest candidate decision score from the previous target number of candidate vehicles is selected as the final candidate vehicle, and this final candidate vehicle is used as the sole comparison object of the currently observed vehicle to output the same vehicle result, different vehicle result, or pending result. Specifically: When the final candidate vehicles meet the common path drift requirement Common path determination threshold, differential path drift The threshold for determining the difference path, and the candidate decision score. When the candidate decision threshold is reached, the result is determined to be the same vehicle. When all candidate vehicles do not meet the same vehicle determination criteria, and at least one of the candidate vehicles meets the difference path drift criteria. When the difference path determination threshold is reached, the result is determined to be an abnormal vehicle. When the highest score of a candidate decision is less than the candidate decision threshold, it is determined to be a pending result.
[0048] The method described in this application can identify a unique and optimal candidate from a target number of candidate vehicles, thus avoiding the situation where multiple scattered arbitration results are obtained and a final disposal conclusion cannot be reached.
[0049] S4. If the same vehicle is detected, issue a release command; If an abnormal vehicle is detected, an interception command is issued; If a pending result is obtained, a virtual blocking command is issued.
[0050] When S3 issues a result indicating that the vehicle is in the same vehicle, it issues a release command and does not enter the abnormal handling process.
[0051] When S3 sends out the abnormal vehicle result, an interception command is issued.
[0052] When S3 issues a pending result, the current exit congestion index is calculated. Specifically, the parking lot exit congestion status is calculated according to the following formula: , Where H represents the exit congestion index. The number of grids to divide the parking lot exit area. For the current moment, the first The probability of vehicle occupancy within a grid cell Preset congestion threshold; when When it is determined that the parking lot exit is congested, The parking lot exit is determined to be in a non-congested state.
[0053] This formula reflects the level of congestion in the passage by statistically analyzing the dispersion of vehicle occupancy distribution in the exit area, and can quickly determine whether the exit is suitable for implementing rigid interception.
[0054] When S3 issues a pending result and the parking lot exit is not congested, an interception command is issued to execute a physical interception and further verify whether it is the same vehicle or a different vehicle. When S3 issues a pending result and the parking lot exit is congested, virtual blocking is executed. During the virtual blocking process, vehicles with pending results are physically released. At the same time, vehicle representation features, semantic tag summaries, and temporal and spatial indexes are extracted to generate multimodal digital fingerprints. Hash values are calculated for the corresponding video segments. The multimodal digital fingerprints and hash values are written into the default evidence chain, and blacklist information is synchronized.
[0055] This approach avoids exacerbating exit congestion due to rigid interception during peak periods, while also preserving the evidentiary basis needed for subsequent recovery and joint punishment.
[0056] Thus, S3 completes the selection of the sole target from the candidate vehicles of the target number, and S4 executes the corresponding on-site control actions based on the sole arbitration conclusion, thereby forming a clear closed-loop process of candidate screening, sole decision, and on-site handling.
[0057] A parking management system based on cognitive arbitration, used to implement a parking management method based on cognitive arbitration, includes: The cognitive uncertainty score calculation module acquires the image of the currently observed vehicle, performs multiple random forward inferences on the image to obtain a set of feature vectors, calculates the mean of the feature vectors, and calculates the cognitive uncertainty score of the currently observed vehicle based on the difference between each feature vector and the mean of the feature vectors.
[0058] The competition intensity index calculation module performs reachability determination on vehicles in the historical vehicle set based on the observation time, observation location, and parking lot topology of the currently observed vehicle. Vehicles that meet the reachability determination are selected as candidate vehicles. The candidate vehicles are sorted from high to low according to the similarity of the coarse matching. The candidate vehicles with the highest similarity are selected to form the disputed candidate set. The memory intensity value is calculated for each candidate vehicle in the disputed candidate set. The competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles.
[0059] The arbitration priority index is obtained by weighted summation of the standard deviations of cognitive uncertainty score, competition intensity index and memory intensity value. When the arbitration priority index exceeds the priority index threshold, the execution result is determined.
[0060] The result determination module constructs disputed vehicle pairs for the currently observed vehicle and each candidate vehicle in the disputed candidate set. It extracts the stable common region and exclusive difference region of each disputed vehicle and maps them to commonality masks and difference masks, respectively. Based on the commonality mask and difference mask, it performs dual-path recoding on the feature vectors of the currently observed vehicle and each candidate vehicle, calculates the common path drift and difference path drift corresponding to each candidate vehicle, and calculates the candidate decision score by combining the memory strength value of each candidate vehicle. The candidate vehicle with the highest candidate decision score is selected as the final candidate vehicle. Based on the common path drift, difference path drift and candidate decision score of the final candidate vehicle, it obtains the same vehicle result, different vehicle result, or pending result.
[0061] The processing module issues a release command if it receives a result indicating the vehicle is in the same vehicle; an intercept command if it receives a result indicating the vehicle is in a different vehicle; and a virtual block command if it receives a pending result.
[0062] A parking management device based on cognitive arbitration includes a processor and a memory, wherein the processor executes a computer program stored in the memory to implement a parking management method based on cognitive arbitration.
Claims
1. A parking management method based on cognitive arbitration, characterized in that, Includes the following steps: S1. Obtain the image of the currently observed vehicle, perform multiple random forward inferences on the image of the currently observed vehicle to obtain a set of feature vectors, calculate the mean of the feature vectors, and calculate the cognitive uncertainty score of the currently observed vehicle based on the difference between each feature vector and the mean of the feature vectors. S2. Based on the observation time, observation location and parking lot topology of the current observed vehicle, the vehicles in the historical vehicle set are made reachable. The vehicles that meet the reachability criteria are selected as candidate vehicles. The candidate vehicles are sorted from high to low according to the similarity of the coarse matching. The candidate vehicles with the highest similarity are selected to form the disputed candidate set. The memory strength value is calculated for each candidate vehicle in the disputed candidate set. The competition intensity index is calculated based on the matching score between the current observed vehicle and the target number of candidate vehicles. The arbitration priority index is obtained by weighted summation of the standard deviations of the cognitive uncertainty score, the competition intensity index, and the memory intensity value. When the arbitration priority index exceeds the priority index threshold, S3 is executed. S3. Construct disputed vehicle pairs for the current observed vehicle and each candidate vehicle in the disputed candidate set. Extract the stable common region and exclusive difference region of each disputed vehicle and map them as common mask and difference mask respectively. Perform dual-path recoding on the feature vectors of the current observed vehicle and each candidate vehicle according to the common mask and difference mask respectively. Calculate the common path drift and difference path drift corresponding to each candidate vehicle. Calculate the candidate decision score by combining the memory strength value of each candidate vehicle. Select the candidate vehicle with the highest candidate decision score as the final candidate vehicle. Obtain the same vehicle result, different vehicle result, or undetermined result based on the common path drift, difference path drift, and candidate decision score of the final candidate vehicle. S4. If the same vehicle is detected, issue a release command; If an abnormal vehicle is detected, an interception command is issued; If a pending result is obtained, a virtual blocking command is issued.
2. The parking management method based on cognitive arbitration according to claim 1, characterized in that, In S2, based on the observation time, observation location, and parking lot topology of the currently observed vehicle, reachability determination is performed on the vehicles in the historical vehicle set, specifically as follows: vehicles in the historical vehicle set reachability determination result is determined according to the following formula: ， in, Vehicles in the historical vehicle collection The reachability determination result, Vehicles in the historical vehicle collection The length of the parking lot topological path from the last observation location to the current observation location. Vehicles in the historical vehicle collection The time interval between the last observation and the current observation. Set a reasonable maximum speed for vehicles to pass through the parking lot; only when the corresponding historical vehicle is included in the candidate set as a candidate vehicle.
3. The parking management method based on cognitive arbitration according to claim 1, characterized in that, In S2, the memory strength value is calculated, specifically: , in, This represents the memory strength value. and is the weighting coefficient for memory strength values, and , This is the time decay coefficient; For candidate vehicles The time interval; For candidate vehicles Trajectory continuity score, For candidate vehicles Historical appearance stability score; Candidate vehicles Trajectory continuity score Calculate according to the following formula: ， wherein, the candidate vehicle the number of valid trajectory segments satisfying the topological adjacency and time continuity constraints within the preset time window; the candidate vehicle the number of all trajectory segments within the preset time window candidate vehicle a historical appearance stability score is calculated according to the following formula: ， wherein, is a number of historical observations of the candidate vehicle ; is a historical average feature vector of the candidate vehicle ; is a feature vector corresponding to the nth historical observation; is a historical average feature vector of the candidate vehicle ; is a feature distance function.
4. The parking management method based on cognitive arbitration according to claim 1, characterized in that, In S3, based on the common path drift amount, different path drift amount, and candidate decision score of the final candidate vehicles, the results for the same vehicle, different vehicles, or pending results are obtained, specifically: when the final candidate vehicle satisfies a common path drift amount a common path determination threshold, a difference path drift amount a difference path determination threshold, and a candidate resolution score a candidate resolution threshold, the same vehicle result is determined; When all candidate vehicles do not satisfy the same vehicle determination condition, and at least one of the candidate vehicles satisfies the difference path drift amount difference path determination threshold, the result is determined as different vehicles. When the highest score of a candidate decision is less than the candidate decision threshold, it is determined to be a pending result.
5. The parking management method based on cognitive arbitration according to claim 1, characterized in that, In S3, the common path drift and differential path drift of each candidate vehicle are calculated, and the candidate decision score is calculated by combining the memory strength value of each candidate vehicle. Specifically: candidate resolution score is calculated according to the following formula: ， wherein, and is a weight coefficient for the candidate resolution score, is a memory strength value for the ith candidate vehicle, common path drift amount and difference path drift amount are calculated according to the following equations, respectively: ， ， in, This is the feature vector of the currently observed vehicle. For the first Feature vectors of candidate vehicles and These are the current observed vehicle feature vectors enhanced by commonality masks and the first... Each candidate vehicle feature vector and These are the current observed vehicle feature vector and the i-th candidate vehicle feature vector, respectively, after difference masking enhancement. This is the feature distance function.
6. The parking management method based on cognitive arbitration according to claim 1, characterized in that, S1 calculates the cognitive uncertainty score of the currently observed vehicle, specifically as follows: Let the current observed vehicle image be denoted by The set of feature vectors is obtained by the second random forward inference The mean of the feature vectors is The cognitive uncertainty score is computed as where ， where, is the number of random forward inferences, is the feature vector obtained from the th random forward inference, is the th random forward inference, denotes element-wise multiplication, denotes the 2-norm, and U is the cognitive uncertainty score.
7. The parking management method based on cognitive arbitration according to claim 1, characterized in that, In S2, the competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles. Specifically: ， is the highest matching score between the current observed vehicle and the target number of candidate vehicles, is the matching score between the current observed vehicle and candidate vehicle i, Q is a constant to prevent the denominator from being zero, and is the competition strength index.
8. The parking management method based on cognitive arbitration according to claim 1, characterized in that, If an undetermined result is obtained in S4, calculate the congestion index of the current exit. When the congestion index The parking lot exit is considered congested when the congestion threshold is reached. The parking lot exit is considered to be in a non-congested state when the congestion threshold is reached. When the vehicle is not in a congested state, an interception command is issued to perform a physical interception and further verify whether it is the same vehicle or a different vehicle. When the traffic is congested, virtual blocking is implemented; during the virtual blocking process, vehicles with pending results are physically released.
9. A parking management system based on cognitive arbitration, for implementing the method of any one of claims 1-8. include: The cognitive uncertainty score calculation module acquires the image of the currently observed vehicle, performs multiple random forward inferences on the image to obtain a set of feature vectors, calculates the mean of the feature vectors, and calculates the cognitive uncertainty score of the currently observed vehicle based on the difference between each feature vector and the mean of the feature vectors. The competition intensity index calculation module performs reachability determination on vehicles in the historical vehicle set based on the observation time, observation location, and parking lot topology of the currently observed vehicle. Vehicles that meet the reachability determination are selected as candidate vehicles. The candidate vehicles are sorted from high to low according to the similarity of the coarse matching. The candidate vehicles with the highest similarity are selected to form the disputed candidate set. The memory intensity value is calculated for each candidate vehicle in the disputed candidate set. The competition intensity index is calculated based on the matching score between the currently observed vehicle and the target number of candidate vehicles. The arbitration priority index is obtained by weighted summation of the standard deviations of cognitive uncertainty score, competition intensity index and memory intensity value. When the arbitration priority index exceeds the priority index threshold, the execution result is determined. The result determination module constructs disputed vehicle pairs for the currently observed vehicle and each candidate vehicle in the disputed candidate set. It extracts the stable common region and exclusive difference region of each disputed vehicle and maps them to common and difference masks, respectively. Based on the common and difference masks, it performs dual-path recoding on the feature vectors of the currently observed vehicle and each candidate vehicle, calculates the common path drift and difference path drift corresponding to each candidate vehicle, calculates the candidate decision score by combining the memory strength value of each candidate vehicle, selects the candidate vehicle with the highest candidate decision score as the final candidate vehicle, and obtains the same vehicle result, different vehicle result, or pending result based on the common path drift, difference path drift, and candidate decision score of the final candidate vehicle. The processing module issues a release command if it receives a result indicating the vehicle is in the same vehicle; an intercept command if it receives a result indicating the vehicle is in a different vehicle; and a virtual block command if it receives a pending result.
10. A parking management and control device based on cognitive arbitration, characterized in that, It includes a processor and a memory, wherein the processor implements a parking control method based on cognitive arbitration as described in any one of claims 1-8 when executing a computer program stored in the memory.