Intelligent parking space management method and system based on image recognition

Abstract

The invention relates to the technical field of object recognition, in particular to an intelligent parking space management method and system based on image recognition, and the method comprises the following steps: based on the image data of a parking lot, calling sparse convolution operation, eliminating a zero pixel region, screening non-zero pixel boundary feature points, extracting boundary feature point distribution, and detecting geometric characteristics; and separating the parking space boundary contour from the background area. According to the method, the feature matrix is constructed by combining pixel density distribution, so that parking space area feature capture is more accurate, the recognition capability of a complex scene is enhanced, the arrangement mode and shielding change are extracted through global and local feature weighted operation, the parking space state analysis capability is improved, abnormal data processing is optimized through abnormal area labeling and boundary adjustment, and the accuracy of parking space identification is improved. The method improves the recognition precision of a complex environment, supplements contour points and geometric symmetry fitting to achieve parking space contour reconstruction, still guarantees the recognition accuracy under the shielding condition, integrates and optimizes the parking space information output in a multi-resolution manner, enables the result to be coordinated and unified, and improves the parking management efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of object recognition, and in particular to an intelligent parking space management method and system based on image recognition. Background Art

[0002] The field of object recognition technology refers to the technology of automatically identifying and extracting specific objects from digital images or videos through computer vision and image processing technology. This technology usually uses algorithms to analyze the feature information in the image to determine the category, location and other relevant attributes of the target object. Object recognition is widely used in scenarios such as autonomous driving, security monitoring, face recognition, industrial inspection, etc. Its core is to achieve accurate recognition and real-time response to target objects through model training and optimization, which is efficient and intelligent.

[0003] Among them, the intelligent parking space management method based on image recognition refers to the real-time detection and analysis of parking space occupancy in the parking lot through object recognition technology, helping users to quickly find vacant parking spaces, while improving the management efficiency and resource utilization of the parking lot. This method usually combines the parking space image data collected by the camera and uses the object recognition algorithm to accurately locate the vehicle position, thereby realizing intelligent parking space allocation and management.

[0004] The existing technology is not capable of processing boundary separation in complex scenes, and is easily affected by occlusion or low light, resulting in errors in the recognition of parking space contours. There is a lack of effective filtering of background noise, resulting in redundant information affecting the recognition results. Feature extraction and matching are weak when the parking spaces are arranged in a complex manner or there are occlusions, and it is difficult to handle abnormal areas. The multi-resolution data integration capability is insufficient, making it difficult to ensure the uniformity and global coordination of the recognition results. These deficiencies make it difficult for parking lots to achieve real-time and accurate analysis of parking space status in complex environments, reducing management efficiency. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent parking space management method and system based on image recognition.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an intelligent parking space management method based on image recognition, comprising the following steps:

[0007] S1: Based on the image data of the parking lot, the sparse convolution operation is called to exclude the zero pixel area and filter the non-zero pixel boundary feature points, extract the distribution of boundary feature points and detect the geometric characteristics, separate the parking space boundary contour and the background area, obtain the feature area based on the pixel density distribution, and generate the regional feature matrix;

[0008] S2: Based on the regional feature matrix, extract the parking space boundary contour coordinate set, call the arrangement parameters and spacing parameters to perform feature extraction, extract the parking space arrangement pattern and the boundary change of the occluded area, calculate the weighted operation results of the global features and the local features, and generate the global and local fusion feature values;

[0009] S3: Based on the global and local fusion feature values, the distribution parameters of the boundary coordinate set are extracted, matching operations are performed in combination with the parking space rectangular constraint, the abnormal parking area is marked and the order of the boundary points of the abnormal area is adjusted, the geometric and position parameter data of the abnormal area is extracted, and the shape constraint prediction value is generated;

[0010] S4: based on the shape constraint prediction value, extracting the boundary coordinate points of the occluded area, calling the offset parameters to adjust the coordinate point positions and calculating the supplementary contour points, performing the complete shape fitting operation using the supplementary boundary point set, reconstructing the complete parking space contour in combination with the geometric symmetry relationship, and generating the offset dynamic prediction result;

[0011] S5: Based on the dynamic prediction result of the offset, call the multi-resolution image feature to extract the parking space arrangement and boundary contour data, integrate the multi-resolution arrangement and boundary data, unify the coordinate mapping and output the complete parking space information of the parking lot, and generate a parking space identification data table.

[0012] The regional feature matrix includes boundary feature point distribution, geometric characteristics, and pixel density distribution; the global-local fusion feature value includes global features, local features, and weighted calculation results; the shape constraint prediction value includes abnormal area geometric parameters, position parameters, and boundary point sequence; the offset dynamic prediction result includes boundary coordinate points after offset parameter adjustment, a set of supplementary contour points, and a complete shape fitting result; the parking space recognition data table includes parking space arrangement, boundary contour data, and coordinate mapping information.

[0013] As a further solution of the present invention, the step of obtaining the regional feature matrix is ​​specifically as follows:

[0014] S111: extracting non-zero pixel areas in the parking lot image, removing zero pixel areas by sparse convolution operation, and then screening non-zero pixel boundary feature points to generate a parking lot boundary feature point set;

[0015] S112: using the parking lot boundary feature point set, detecting the geometric distribution characteristics of the points and calculating the pixel density distribution, separating the parking space boundary contour and the background area, and generating parking space boundary area data;

[0016] S113: Based on the parking space boundary area data, the corresponding feature area is extracted in combination with the density distribution and the boundary characteristics, and a feature area matrix is ​​established according to the geometric shape and distribution law, using the formula:

[0017]

[0018] Generate regional feature matrix;

[0019] Among them, M represents the regional feature matrix, d i Represents the density distribution value of the boundary points of the feature area, x i ,y i They represent the horizontal and vertical distribution coordinates of the boundary points respectively, and n is the total number of feature points in the feature area.

[0020] As a further solution of the present invention, the step of obtaining the global-local fusion feature value is specifically as follows:

[0021] S211: extracting all coordinate sets of the parking space boundary contour based on the regional feature matrix, calling the arrangement parameters to analyze the distribution pattern of the boundary points, and generating a parking space boundary contour coordinate set;

[0022] S212: using the parking space boundary contour coordinate set and combining the spacing parameter to detect the boundary change of the occluded area, analyzing the parking space arrangement pattern and calculating the global features and local feature values ​​of multiple regions, and generating regional global feature values ​​and regional local feature values;

[0023] S213: Based on the regional global eigenvalue and the regional local eigenvalue, weighted fusion is performed through weight parameters, using the formula:

[0024]

[0025] Calculate and generate global and local fusion eigenvalues;

[0026] Among them, F represents the global and local fusion feature value, G i represents the global eigenvalue of the ith region, L i represents the local eigenvalue of the ith region, w i is the weight parameter of the ith area, and n is the number of areas where parking spaces are distributed.

[0027] As a further solution of the present invention, the step of obtaining the shape constraint prediction value is specifically as follows:

[0028] S311: extracting distribution parameters of a boundary coordinate set based on the global-local fusion feature value, analyzing the spacing and arrangement characteristics of boundary points in combination with the distribution parameters, and generating a boundary coordinate distribution characteristic set;

[0029] S312: using the boundary coordinate distribution characteristic set and combining the parking space rectangle constraint to perform a matching operation on the boundary points, detecting the coordinate change of the abnormal area boundary and adjusting the order of the abnormal area boundary points, and generating an adjusted abnormal area boundary coordinate set;

[0030] S313: extract the adjusted abnormal area boundary coordinate set, combine the geometric characteristics and position distribution of the boundary points, and use the formula:

[0031]

[0032] Calculate and generate shape constraint prediction values;

[0033] Among them, P represents the shape constraint prediction value, x i Represents the horizontal coordinate of the i-th boundary point, y i Represents the longitudinal coordinate of the i-th boundary point, x c Represents the horizontal coordinate of the center point of the abnormal area, y c Represents the vertical coordinate of the center point of the abnormal area, l i represents the lateral spacing parameter of the i-th boundary point in the abnormal area, w i represents the longitudinal spacing parameter of the i-th boundary point in the abnormal area, d i is the distance from the ith boundary point to the center point, σ is the distribution adjustment parameter, and n is the number of boundary points.

[0034] As a further solution of the present invention, the step of obtaining the offset dynamic prediction result is specifically:

[0035] S411: extracting boundary coordinate points of the occluded area based on the shape constraint prediction value, adjusting the positions of the boundary coordinate points using offset parameters, and generating an adjusted boundary coordinate point set;

[0036] S412: using the adjusted boundary coordinate point set, calculating supplementary contour points to fill the gaps between the coordinate points, and using the supplementary boundary point set to perform a shape fitting operation to generate a supplementary contour shape set;

[0037] S413: extract the supplementary contour shape set, combine the geometric symmetry relationship, and use linear and nonlinear transformations to adopt the formula:

[0038]

[0039] Reconstruct the complete parking space outline and generate dynamic prediction results of the offset;

[0040] Among them, R represents the offset dynamic prediction result, α i is the weight factor of the i-th supplementary contour point, S i is the area of ​​the i-th supplementary contour point, n is the number of supplementary contour points, d oi is the original distance of the i-th supplementary contour point, d ci is the adjusted distance, d max is the maximum estimated offset distance.

[0041] As a further solution of the present invention, the steps of obtaining the parking space identification data table are specifically as follows:

[0042] S511: Based on the offset dynamic prediction result, a multi-resolution image feature extraction technology is used to obtain parking space arrangement and boundary contour data, and a parking space boundary is extracted through detail enhancement and edge sharpening technology to generate a parking space boundary contour feature set;

[0043] S512: Based on the parking space boundary contour feature set, the multi-resolution arrangement data is integrated to unify the coordinate mapping, and the corresponding parking space position information is adjusted to generate a unified coordinate mapping data set;

[0044] S513: extract the unified coordinate mapping data set, and perform data fusion operation using the formula:

[0045]

[0046] Integrate and output complete parking space information to generate a parking space identification data table;

[0047] Among them, T represents the parking space identification data table, β i is the matching coefficient of the i-th parking space, P i is the location data of the i-th parking space, n is the number of parking spaces, O j is the observation data of the jth parking space, and m is the total number of observation data points.

[0048] An intelligent parking space management system based on image recognition, the intelligent parking space management system based on image recognition is used to execute the above-mentioned intelligent parking space management method based on image recognition, the system comprises:

[0049] The parking space boundary extraction module extracts the distribution of non-zero pixel areas based on the image data of the parking lot, screens the geometric parameters of boundary feature points, calculates the distribution value of the boundary contour, removes the background area, extracts the density and eigenvalue of the characteristic area distribution, and establishes the regional feature matrix;

[0050] The parking space feature calculation module extracts the boundary contour coordinate point set based on the regional feature matrix and calculates the arrangement mode change value, analyzes the gradient distribution of the boundary point change, extracts the change characteristics of the occluded area, and combines the global and local weighted fusion operations to obtain the global and local fusion feature value;

[0051] The parking space abnormality matching module extracts the boundary coordinate set distribution parameters based on the global and local fusion feature values, calculates the matching value of the abnormal area boundary, adjusts the order of the abnormal area boundary points, extracts the geometric and position change parameters, generates the boundary features of the abnormal area, and establishes the shape constraint prediction value;

[0052] The parking space shape reconstruction module extracts the boundary coordinate points of the occluded area based on the shape constraint prediction value, supplements the missing coordinate points with the offset, regenerates the boundary point set, calculates the geometric symmetry parameters of the contour, reconstructs the complete boundary distribution of the parking space contour, and obtains the offset dynamic prediction result;

[0053] The parking space information integration module extracts the arrangement data and boundary contours in the multi-resolution image based on the offset dynamic prediction results, integrates the parking space arrangement and boundary data, completes coordinate mapping, generates complete data on the parking space arrangement and boundary, and establishes a parking space identification data table.

[0054] Compared with the prior art, the advantages and positive effects of the present invention are:

[0055] In the present invention, sparse convolution is used to exclude zero pixel areas and extract boundary feature points, thereby reducing invalid data interference and improving the accuracy of boundary separation. The feature matrix is ​​constructed in combination with the pixel density distribution to make the parking area feature capture more accurate and enhance the recognition ability of complex scenes. Global and local feature weighted operations are used to extract arrangement patterns and occlusion changes, thereby improving the ability to analyze parking status. Abnormal area labeling and boundary adjustment optimize abnormal data processing and improve recognition accuracy in complex environments. Supplementary contour points and geometric symmetry fitting are used to achieve parking space contour reconstruction, and recognition accuracy is still guaranteed under occlusion. Multi-resolution integration optimizes parking space information output, making the results coordinated and unified, and improving parking management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0057] Figure 2 A flow chart of the steps for obtaining the regional feature matrix of the present invention;

[0058] Figure 3 This is a flow chart of the steps for obtaining the global and local fusion feature values ​​of the present invention;

[0059] Figure 4 A flow chart of the steps for obtaining the shape constraint prediction value of the present invention;

[0060] Figure 5 A flow chart of the steps for obtaining the offset dynamic prediction result of the present invention;

[0061] Figure 6 This is a flow chart of the steps for obtaining the parking space identification data table of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0064] Embodiment 1

[0065] See also Figure 1 The present invention provides a technical solution: an intelligent parking space management method based on image recognition, comprising the following steps:

[0066] S1: Based on the image data of the parking lot, the sparse convolution operation is called to exclude the zero pixel area and filter the non-zero pixel boundary feature points, extract the distribution of boundary feature points and detect the geometric characteristics, separate the parking space boundary contour and the background area, obtain the feature area based on the pixel density distribution, and generate the regional feature matrix;

[0067] S2: Based on the regional feature matrix, extract the parking space boundary contour coordinate set, call the arrangement parameters and spacing parameters for feature extraction, extract the parking space arrangement pattern and the boundary changes of the occluded area, calculate the weighted operation results of the global features and local features, and generate the global and local fusion feature values;

[0068] S3: Based on the global and local fusion feature values, the distribution parameters of the boundary coordinate set are extracted, and matching operations are performed in combination with the parking space rectangle constraint. The abnormal parking area is marked and the order of the boundary points of the abnormal area is adjusted. The geometric and position parameter data of the abnormal area are extracted to generate the shape constraint prediction value;

[0069] S4: Based on the shape constraint prediction value, the boundary coordinate points of the occluded area are extracted, the offset parameters are called to adjust the coordinate point positions and calculate the supplementary contour points, the complete shape fitting operation is performed using the supplementary boundary point set, the complete parking space contour is reconstructed in combination with the geometric symmetry relationship, and the offset dynamic prediction result is generated;

[0070] S5: Based on the offset dynamic prediction results, call the multi-resolution image features to extract the parking space layout and boundary contour data, combine the multi-resolution layout and boundary data for integration, unify the coordinate mapping and output the complete parking space information, and generate a parking space identification data table.

[0071] The regional feature matrix includes the distribution of boundary feature points, geometric characteristics, and pixel density distribution. The global-local fusion feature values ​​include global features, local features, and weighted calculation results. The shape constraint prediction values ​​include the geometric parameters of the abnormal area, position parameters, and boundary point sequence. The offset dynamic prediction results include the boundary coordinate points after the offset parameters are adjusted, the supplementary contour point set, and the complete shape fitting results. The parking space recognition data table includes the parking space arrangement, boundary contour data, and coordinate mapping information.

[0072] See also Figure 2 , the steps to obtain the regional feature matrix are as follows:

[0073] S111: extracting non-zero pixel areas in the parking lot image, removing zero pixel areas by sparse convolution operation, and then screening non-zero pixel boundary feature points to generate a parking lot boundary feature point set;

[0074] The purpose of preliminary processing of parking lot image data is to reduce data redundancy and background noise, and improve the accuracy and efficiency of subsequent processing. The process includes inputting the parking lot image data into a sparse convolutional network. The network learns to identify and separate useful feature points and background noise. The output boundary feature point set is the key basic data for subsequent steps to analyze parking spaces. It determines the approximate outline and position of the parking space, and provides the necessary preliminary data processing for the next step of parking space identification and geometric characteristics analysis. Through this series of operations, clear parking lot image data with boundary markings can be obtained, which improves the overall performance of the system and user experience.

[0075] S112: Using the parking lot boundary feature point set, by detecting the geometric distribution characteristics of the points and calculating the pixel density distribution, the parking space boundary contour and the background area are separated to generate parking space boundary area data;

[0076] First, call the parking lot boundary feature point set, and use advanced image processing technology to analyze the geometric characteristics of these points, such as shape, size and relative position. Based on these characteristics, use the image segmentation algorithm to separate the parking space boundary and background area. Then, perform density analysis on the separated parking space area and calculate the pixel distribution density in the area. High-density areas often indicate that the parking space outline is more obvious. The output of this step is the parking space boundary area data, which provides accurate input data for the next step of feature area extraction.

[0077] S113: Based on the parking space boundary area data, the corresponding feature area is extracted in combination with the density distribution and boundary characteristics, and a feature area matrix is ​​established according to the geometric shape and distribution law, using the formula:

[0078]

[0079] Generate regional feature matrix;

[0080] Among them, M represents the regional feature matrix, d i Represents the density distribution value of the boundary points of the feature area, x i ,y i They represent the horizontal and vertical distribution coordinates of the boundary points respectively, and n is the total number of feature points in the feature area.

[0081] formula:

[0082]

[0083] The benefit of the formula is that, by combining geometric shape and pixel density, the formula can comprehensively consider boundary characteristics and pixel distribution to more accurately determine and describe the feature matrix of the parking area.

[0084] Detailed explanation of the formula and the process of formula calculation and derivation:

[0085] Suppose there is a parking lot image area, which contains n boundary feature points, and the position coordinates of each point are (x i ,y i ), and the density distribution value of each point is d i , then the regional feature matrix M can be obtained by calculating the weighted average relative position and the root mean square of the relative distance of these points. For example, if there are 3 points, their coordinates are (1,2), (3,4), (5,6), and the density distribution values ​​are 10, 20, and 30 respectively, then:

[0086]

[0087] The result shows that the calculated feature matrix M value is 17.84, which means that the key geometry and density information about the parking area is successfully extracted from the original boundary feature points. This value reflects the comprehensive characteristics of the distribution and geometric properties of the feature points in the area, which is helpful for further image processing and parking space detection.

[0088] See also Figure 3 , the specific steps for obtaining the global and local fusion feature values ​​are:

[0089] S211: extracting all coordinate sets of the parking space boundary contour based on the regional feature matrix, calling the arrangement parameters to analyze the distribution pattern of the boundary points, and generating a parking space boundary contour coordinate set;

[0090] Image processing technology is used to analyze the physical boundaries of parking spaces and identify the boundaries between parking spaces and the surrounding environment. This involves grayscale conversion, edge detection, and contour tracking of images. Boundary detection algorithms such as Canny or Sobel algorithms are used to identify straight lines and curves in images. These algorithms determine the location of boundaries by calculating the gradient of image pixels. Through these technologies, the geometric shape of parking spaces can be accurately distinguished from adjacent parking spaces or other obstacles, ensuring the accurate application of arrangement parameters and spacing parameters in subsequent steps. Through accurate parking space boundary contour recognition, the parking space arrangement pattern can be better analyzed, which is crucial for optimizing parking lot space utilization and navigation systems, and a set of parking space boundary contour coordinates can be generated.

[0091] S212: using the parking space boundary contour coordinate set and the spacing parameter to detect the boundary change of the occluded area, analyzing the parking space arrangement pattern and calculating the global features and local feature values ​​of multiple regions, and generating the regional global feature value and the regional local feature value;

[0092] The process involves advanced image analysis and pattern recognition technology. By analyzing the distribution of the coordinates of the parking space boundary contours, the arrangement of the parking spaces can be identified, such as whether they are arranged diagonally, vertically or horizontally. This information is critical to understanding the layout of the parking lot, especially in a complex parking lot environment. In addition, the calculation of the spacing parameters is based on the distance between the parking space boundaries, which requires precise measurement and calculation to ensure the safe parking and effective flow of vehicles. Through these analyses, the parking lot space can be better managed, the vehicle parking and extraction process can be optimized, and regional global eigenvalues ​​and regional local eigenvalues ​​can be generated.

[0093] S213: Based on the regional global eigenvalue and the regional local eigenvalue, weighted fusion is performed through weight parameters, using the formula:

[0094]

[0095] Calculate and generate global and local fusion eigenvalues;

[0096] Among them, F represents the global and local fusion feature value, G i represents the global eigenvalue of the ith region, L i represents the local eigenvalue of the ith region, w i is the weight parameter of the ith area, and n is the number of areas where parking spaces are distributed.

[0097] formula:

[0098]

[0099] The benefit of the formula is that it combines global and local eigenvalues ​​in a weighted manner, which allows adjustments to specific conditions in different parking areas, making the results more refined and adapted to specific scenarios.

[0100] Detailed explanation of the formula and the process of formula calculation and derivation:

[0101] Set the number of parking areas n = 5, and the weight parameter w is the importance of each area feature. The specific value is as follows

[0102] {0.2, 0.3, 0.1, 0.3, 0.1}, the global eigenvalue G and the local eigenvalue L are {200, 150, 180, 160, 170} and {50, 60, 55, 65, 45} respectively, the specific calculation is:

[0103]

[0104] F = 460.2;

[0105] The results show that the weighted global-local fusion feature value is 460.2, which reflects the comprehensive impact of different regional characteristics and provides a comprehensive evaluation index for parking management systems to adjust parking area design or management strategies.

[0106] See also Figure 4 , the steps for obtaining the shape constraint prediction value are as follows:

[0107] S311: extracting distribution parameters of the boundary coordinate set based on the global and local fusion feature values, analyzing the interval and arrangement characteristics of the boundary points in combination with the distribution parameters, and generating a boundary coordinate distribution characteristic set;

[0108] By calling the distribution parameter analysis of the boundary coordinate set, the parameter is mainly calculated based on the spatial distribution and density of the boundary points in the image. Combined with the rectangular constraint of the parking space, this step mainly includes calculating the distance from each boundary point to the center of the parking space, and the angle difference between the boundary points to determine whether each point conforms to the rectangular characteristics of the parking space. The acquisition of distribution parameters is mainly achieved through the edge detection algorithm in the image processing software. The algorithm enhances the contrast of the image boundary to make the parking space boundary more obvious. The algorithm further sorts these boundary points to ensure that they are arranged in order from one end to the other, which helps the accuracy of subsequent matching operations. The adjusted boundary point order is particularly critical for the subsequent abnormal area extraction and matching operations, because only the correct order can correctly match the difference between the template and the actual parking space image.

[0109] S312: using the boundary coordinate distribution characteristic set and combining the parking space rectangle constraint to perform a matching operation on the boundary points, detecting the coordinate change of the abnormal area boundary and adjusting the order of the abnormal area boundary points, and generating an adjusted abnormal area boundary coordinate set;

[0110] The first thing to do is to detect anomalies at the boundary points. This detection is mainly based on the standard rectangular model of the parking space. Each actually measured boundary point position is evaluated to determine whether these points fall within the predetermined rectangular boundary. In addition, the distance between each point and the theoretical boundary of the parking space needs to be calculated to determine whether the boundary point is abnormal. The calculation of these distances depends on the relative position between the boundary point and the center point of the parking space. Through these calculations, the order of the boundary points can be accurately adjusted to make it more consistent with the rectangular constraints of the parking space. This adjustment process is achieved through a geometric algorithm to ensure that each point can be optimized and adjusted according to the actual shape of the parking space, thereby improving the effectiveness and safety of the parking space use.

[0111] S313: Extract the adjusted abnormal area boundary coordinate set, combine the geometric characteristics and position distribution of the boundary points, and use the formula:

[0112]

[0113] Calculate and generate shape constraint prediction values;

[0114] Where P represents the shape constraint prediction value, x i Represents the horizontal coordinate of the i-th boundary point, y i Represents the longitudinal coordinate of the i-th boundary point, x c Represents the horizontal coordinate of the center point of the abnormal area, y c Represents the vertical coordinate of the center point of the abnormal area, l i represents the lateral spacing parameter of the i-th boundary point in the abnormal area, w i represents the longitudinal spacing parameter of the i-th boundary point in the abnormal area, d i is the distance from the ith boundary point to the center point, σ is the distribution adjustment parameter, and n is the number of boundary points.

[0115] formula:

[0116]

[0117] The benefit of the formula is that it can effectively evaluate and predict the geometry of abnormal parking spaces by comprehensively considering the position deviation of boundary points and the distance attenuation function, so that areas that do not meet parking space standards can be accurately identified and adjusted, thereby optimizing the layout of the parking lot and the use of parking spaces.

[0118] Detailed explanation of the formula and the process of formula calculation and derivation:

[0119] Set the center coordinates of the parking space to (x c ,y c ), the coordinates of each boundary point are (x i ,y i ), the ideal length and width of the parking space are l i and w i , the actual measured distance from the boundary point to the center point is d i , adjust the parameter σ to control the influence of distance on the result. For example, for a specific parking space, let x c =50,y c =50, boundary point (x1, y1) = (45, 55), l1 = 10, w1 = 10, d1 = 7, σ = 5, the contribution of the first boundary point is calculated as:

[0120]

[0121] Repeat this calculation for all boundary points and sum to get P.

[0122] The results show that the calculated P value can be used directly to judge the compliance of parking spaces. If the P value is too high, it means that the shape of the parking space is very different from the preset rectangular model and needs to be adjusted. This numerical result is directly related to the management and layout optimization of parking spaces, helping managers to quickly identify and deal with irregular parking spaces.

[0123] See also Figure 5 , the specific steps for obtaining the offset dynamic prediction results are:

[0124] S411: extracting boundary coordinate points of the occluded area based on the shape constraint prediction value, adjusting the positions of the boundary coordinate points using the offset parameters, and generating an adjusted boundary coordinate point set;

[0125] The boundary coordinate point adjustment process based on shape constraint prediction uses computer vision algorithms for boundary detection. First, the occluded area in the image is identified through advanced image processing technology, and then the initial boundary coordinate points of the occluded area are extracted through the algorithm. These coordinate points are fine-tuned using offset parameters to ensure that the actual shape of the occluded object is more accurately reflected. The calculation of the offset parameters is optimized by comparing the predicted model of the occluded object with the actual observation data. This process involves parameter tuning technology in machine learning and spatial geometry algorithms in computer graphics processing to generate a set of accurate adjusted boundary coordinate points.

[0126] S412: using the adjusted boundary coordinate point set, calculating supplementary contour points to fill the gaps between the coordinate points, and using the supplementary boundary point set to perform a shape fitting operation to generate a supplementary contour shape set;

[0127] Calculations are performed using a set of adjusted boundary coordinate points, which are obtained through specific surveying techniques and precise data from geographic information systems, including data from satellite image analysis and actual ground measurements. These coordinate points are combined to perform calculations that include using interpolation algorithms and smoothing techniques to generate missing coordinate points. This algorithm calculates the approximate location of unknown points based on the distribution characteristics of known points and the geographical location relationship of adjacent points. At the same time, smoothing techniques are used to optimize the distribution of these points so that the contour is closer to the actual terrain. This process strictly adjusts interpolation parameters and smoothing parameters based on terrain continuity and regional characteristics. The specific values ​​of the generated supplementary contour points are obtained through computer simulation and field verification to ensure that each supplementary point can truly reflect the details and changes of the terrain.

[0128] S413: Extract the complementary contour shape set, combine the geometric symmetry relationship, and use linear and nonlinear transformations, using the formula:

[0129]

[0130] Reconstruct the complete parking space outline and generate dynamic prediction results of the offset;

[0131] Among them, R represents the offset dynamic prediction result, α i is the weight factor of the i-th supplementary contour point, S i is the area of ​​the i-th supplementary contour point, n is the number of supplementary contour points, d oi is the original distance of the i-th supplementary contour point, d ci is the adjusted distance, d max is the maximum estimated offset distance.

[0132] formula:

[0133]

[0134] The benefit of the formula is that by considering the area ratio of each supplementary contour point and the normalization of the offset distance, the weight of each point can be dynamically adjusted, so that the reconstructed parking space contour is closer to the actual physical form, thereby improving the accuracy and reliability of the parking space detection system.

[0135] Detailed explanation of the formula and the process of formula calculation and derivation:

[0136] Assume that the parking space has 5 supplementary contour points, and the area of ​​each point is S i They are 5, 3, 4, 2, and 6 square meters respectively, with a total area of square meters, the original and adjusted distances are d oi and d ci , maximum offset distance d max= 5 meters, assuming that the adjusted distances are 1, 0, 2, 1, and 3 meters, and the original distances are 3, 0, 2, 3, and 1 meters, respectively, and the weight factor α i Assuming 1:

[0137]

[0138] After calculation, the R value is 0.76. The result shows that by accurately adjusting and weighting each boundary point, the accurate outline of the parking space can be effectively reconstructed, providing an efficient and reliable method for the parking space management system to ensure the correct use and efficient management of the parking space.

[0139] See also Figure 6 , the specific steps for obtaining the parking space identification data table are:

[0140] S511: Based on the offset dynamic prediction result, the parking space arrangement and boundary contour data are obtained by using the multi-resolution image feature extraction technology, the parking space boundary is extracted by using the detail enhancement and edge sharpening technology, and a parking space boundary contour feature set is generated;

[0141] This technology can analyze images at different resolution levels and identify boundaries and details of different scales, so as to more accurately extract the geometric contours of parking spaces. Through detail enhancement and edge sharpening techniques, the contour features of parking spaces can be effectively separated from the background, which not only improves the reliability of the data, but also provides a solid foundation for subsequent parking space identification and analysis. The generated parking space boundary contour feature set contains all the boundary information of the parking space, which is obtained by analyzing and enhancing the image layer by layer. This process involves systematic processing of the original image data, including the comprehensive application of multiple steps such as image segmentation, edge detection, and morphological operations. Each operation is adjusted based on the standard algorithm in image processing technology and optimized for the specific scene of the parking lot.

[0142] S512: Based on the parking space boundary contour feature set, the multi-resolution arrangement data is integrated, the coordinate mapping is unified, and the corresponding parking space position information is adjusted to generate a unified coordinate mapping data set;

[0143] These technologies can synthesize image data from different sources and with different resolutions into a unified data map, so that the accurate location information of parking spaces is retained. In this process, special emphasis is placed on precise matching of coordinates and seamless stitching of data to ensure that all parking space data are aligned without errors in the same coordinate system. The generated unified coordinate mapping data set not only includes the location information of each parking space, but also integrates the relative relationship and relative size between parking spaces. Each data point is obtained through precise calculation and meticulous adjustment to ensure that the parking space layout of the entire parking lot is visually consistent with the actual situation, providing an accurate reference for subsequent parking space monitoring and management.

[0144] S513: Extract the unified coordinate mapping data set, and use the formula:

[0145]

[0146] Integrate and output complete parking space information to generate a parking space identification data table;

[0147] Among them, T represents the parking space identification data table, β i is the matching coefficient of the i-th parking space, P i is the location data of the i-th parking space, n is the number of parking spaces, O j is the observation data of the jth parking space, and m is the total number of observation data points.

[0148] formula:

[0149]

[0150] The benefit of the formula is that it combines the location data of each parking space with the matching coefficient and averages the observed data of all parking spaces, thereby achieving efficient integration and accurate output of parking lot parking space information. This method not only improves the efficiency of data processing, but also enhances the accuracy and availability of the data.

[0151] Detailed explanation of the formula and the process of formula calculation and derivation:

[0152] Assume that a parking lot has 5 parking spaces, and the matching coefficients β of each parking space are 0.8, 0.95, 0.75, 0.9, and 0.85 respectively. The position data P are 20, 25, 15, 22, and 18 respectively, and the observation data O are 200, 240, 180, 220, and 210 respectively. The total number of observation data points is m = 5. First, calculate the β of each parking space. i ·P i :

[0153] 0.8×20=16;

[0154] 0.95×25=23.75;

[0155] 0.75×15=11.25;

[0156] 0.9×22=19.8;

[0157] 0.85×18=15.3;

[0158] Then calculate the mean of the observations:

[0159]

[0160] Finally, substitute all the calculated results into the formula:

[0161]

[0162] The results show that through comprehensive processing and weighted averaging of parking space data, a comprehensive value representing the parking space information of the entire parking lot can be obtained. This value can be used for further data analysis and decision support, such as resource allocation and optimization of parking management strategies.

[0163] An intelligent parking space management system based on image recognition, the intelligent parking space management system based on image recognition is used to execute the above-mentioned intelligent parking space management method based on image recognition, and the system includes:

[0164] The parking space boundary extraction module extracts the distribution of non-zero pixel areas based on the image data of the parking lot, screens the geometric parameters of boundary feature points, calculates the distribution value of the boundary contour, removes the background area, extracts the density and eigenvalue of the characteristic area distribution, and establishes the regional feature matrix;

[0165] The parking space feature calculation module extracts the boundary contour coordinate point set based on the regional feature matrix and calculates the arrangement pattern change value, analyzes the gradient distribution of boundary point changes, extracts the change characteristics of the occluded area, and combines global and local weighted fusion operations to obtain global and local fusion feature values;

[0166] The parking space anomaly matching module extracts the boundary coordinate set distribution parameters based on the global and local fusion feature values, calculates the matching value of the abnormal area boundary, adjusts the order of the abnormal area boundary points, extracts the geometric and position change parameters, generates the boundary features of the abnormal area, and establishes the shape constraint prediction value;

[0167] The parking space shape reconstruction module extracts the boundary coordinate points of the occluded area based on the shape constraint prediction value, supplements the missing coordinate points with the offset, regenerates the boundary point set, calculates the geometric symmetry parameters of the contour, reconstructs the complete boundary distribution of the parking space contour, and obtains the offset dynamic prediction result;

[0168] The parking space information integration module extracts the arrangement data and boundary contours in the multi-resolution image based on the offset dynamic prediction results, integrates the parking space arrangement and boundary data, completes the coordinate mapping, generates complete data on the parking space arrangement and boundary, and establishes a parking space identification data table.

[0169] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent parking space management method based on image recognition, characterized in that: The following steps are involved: Based on the image data of the parking lot, the sparse convolution operation is called to exclude the zero pixel area and filter the non-zero pixel boundary feature points, extract the distribution of boundary feature points and detect the geometric characteristics, separate the parking space boundary contour and the background area, and obtain the feature area based on the pixel density distribution to generate the regional feature matrix; Based on the regional feature matrix, a parking space boundary contour coordinate set is extracted, arrangement parameters and spacing parameters are called to perform feature extraction, parking space arrangement mode and occlusion area boundary changes are extracted, and weighted operation results of global features and local features are calculated to generate global and local fusion feature values; Based on the global and local fusion feature values, the distribution parameters of the boundary coordinate set are extracted, matching operations are performed in combination with the parking space rectangular constraints, the abnormal parking area is marked and the order of the boundary points of the abnormal area is adjusted, the geometric and position parameter data of the abnormal area are extracted, and the shape constraint prediction value is generated; Based on the shape constraint prediction value, the boundary coordinate points of the occluded area are extracted, the offset parameters are called to adjust the coordinate point positions and calculate the supplementary contour points, the complete shape fitting operation is performed using the supplementary boundary point set, the complete parking space contour is reconstructed in combination with the geometric symmetry relationship, and the offset dynamic prediction result is generated; Based on the dynamic prediction result of the offset, the multi-resolution image features are called to extract the parking space arrangement and boundary contour data, and the multi-resolution arrangement and boundary data are integrated to unify the coordinate mapping and output the complete parking space information of the parking lot to generate a parking space identification data table.

2. The intelligent parking space management method based on image recognition according to claim 1 is characterized in that: The regional feature matrix includes boundary feature point distribution, geometric characteristics, and pixel density distribution; the global-local fusion feature value includes global features, local features, and weighted calculation results; the shape constraint prediction value includes abnormal area geometric parameters, position parameters, and boundary point sequence; the offset dynamic prediction result includes boundary coordinate points after offset parameter adjustment, a set of supplementary contour points, and a complete shape fitting result; the parking space recognition data table includes parking space arrangement, boundary contour data, and coordinate mapping information.

3. The intelligent parking space management method based on image recognition according to claim 2 is characterized in that: The steps of obtaining the regional feature matrix are specifically as follows: Extract the non-zero pixel area in the parking lot image, remove the zero pixel area by sparse convolution operation, and then filter the non-zero pixel boundary feature points to generate a set of parking lot boundary feature points; By using the parking lot boundary feature point set, the parking space boundary contour and the background area are separated by detecting the geometric distribution characteristics of the points and calculating the pixel density distribution, thereby generating parking space boundary area data; Based on the parking space boundary area data, the corresponding feature area is extracted in combination with the density distribution and boundary characteristics, and a feature area matrix is ​​established according to the geometric shape and distribution law, using the formula: Generate regional feature matrix; Among them, M represents the regional feature matrix, d i Represents the density distribution value of the boundary points of the feature area, x i ,y i They represent the horizontal and vertical distribution coordinates of the boundary points respectively, and n is the total number of feature points in the feature area.

4. The intelligent parking space management method based on image recognition according to claim 3 is characterized in that: The steps for obtaining the global and local fusion feature values ​​are specifically as follows: Based on the regional feature matrix, all coordinate sets of the parking space boundary contour are extracted, and the distribution law of the boundary points is analyzed by calling the arrangement parameters to generate the parking space boundary contour coordinate set; Using the parking space boundary contour coordinate set and combining the spacing parameter to detect the boundary change of the occluded area, analyzing the parking space arrangement pattern and calculating the global features and local feature values ​​of multiple regions, generating regional global feature values ​​and regional local feature values; Based on the regional global eigenvalue and the regional local eigenvalue, weighted fusion is performed through weight parameters, using the formula: Calculate and generate global and local fusion eigenvalues; Among them, F represents the global and local fusion feature value, G i represents the global eigenvalue of the ith region, L i represents the local eigenvalue of the ith region, w i is the weight parameter of the ith area, and n is the number of areas where parking spaces are distributed.

5. The intelligent parking space management method based on image recognition according to claim 4 is characterized in that: The steps for obtaining the shape constraint prediction value are specifically as follows: Based on the global and local fusion feature values, the distribution parameters of the boundary coordinate set are extracted, and the interval and arrangement characteristics of the boundary points are analyzed in combination with the distribution parameters to generate a boundary coordinate distribution characteristic set; Using the boundary coordinate distribution characteristic set and combining the parking space rectangle constraint, the boundary points are matched, the coordinate changes of the abnormal area boundary are detected, and the order of the abnormal area boundary points is adjusted to generate an adjusted abnormal area boundary coordinate set; Extract the adjusted abnormal area boundary coordinate set, combine the geometric characteristics and position distribution of the boundary points, and use the formula: Calculate and generate shape constraint prediction values; Among them, P represents the shape constraint prediction value, x i Represents the horizontal coordinate of the i-th boundary point, y i Represents the longitudinal coordinate of the i-th boundary point, x c Represents the horizontal coordinate of the center point of the abnormal area, y c Represents the vertical coordinate of the center point of the abnormal area, l i represents the lateral spacing parameter of the i-th boundary point in the abnormal area, w i represents the longitudinal spacing parameter of the i-th boundary point in the abnormal area, d i is the distance from the ith boundary point to the center point, σ is the distribution adjustment parameter, and n is the number of boundary points.

6. The intelligent parking space management method based on image recognition according to claim 5 is characterized in that: The steps for obtaining the offset dynamic prediction result are specifically as follows: Based on the shape constraint prediction value, extracting the boundary coordinate points of the occluded area, adjusting the positions of the boundary coordinate points using the offset parameters, and generating an adjusted boundary coordinate point set; Using the adjusted boundary coordinate point set, calculating supplementary contour points to fill the gaps between the coordinate points, and using the supplementary boundary point set to perform a shape fitting operation to generate a supplementary contour shape set; The complementary contour shape set is extracted, combined with geometric symmetry, through linear and nonlinear transformations, using the formula: Reconstruct the complete parking space outline and generate dynamic prediction results of the offset; Among them, R represents the offset dynamic prediction result, α i is the weight factor of the i-th supplementary contour point, S i is the area of ​​the i-th supplementary contour point, n is the number of supplementary contour points, d oi is the original distance of the i-th supplementary contour point, d ci is the adjusted distance, d max is the maximum estimated offset distance.

7. The intelligent parking space management method based on image recognition according to claim 6 is characterized in that: The steps for obtaining the parking space identification data table are specifically as follows: Based on the dynamic prediction result of the offset, the parking space arrangement and boundary contour data are obtained by using a multi-resolution image feature extraction technology, the parking space boundary is extracted by using detail enhancement and edge sharpening technology, and a parking space boundary contour feature set is generated; Based on the parking space boundary contour feature set, the multi-resolution arrangement data is integrated, the coordinate mapping is unified, and the corresponding parking space position information is adjusted to generate a unified coordinate mapping data set; Extract the unified coordinate mapping data set, and perform data fusion operation using the formula: Integrate and output complete parking space information to generate a parking space identification data table; Among them, T represents the parking space identification data table, β i is the matching coefficient of the i-th parking space, P i is the location data of the i-th parking space, n is the number of parking spaces, O j is the observation data of the jth parking space, and m is the total number of observation data points.

8. An intelligent parking space management system based on image recognition, characterized in that: According to the intelligent parking space management method based on image recognition according to any one of claims 1 to 7, the system comprises: The parking space boundary extraction module extracts the distribution of non-zero pixel areas based on the image data of the parking lot, screens the geometric parameters of boundary feature points, calculates the distribution value of the boundary contour, removes the background area, extracts the density and eigenvalue of the characteristic area distribution, and establishes the regional feature matrix; The parking space feature calculation module extracts the boundary contour coordinate point set based on the regional feature matrix and calculates the arrangement mode change value, analyzes the gradient distribution of the boundary point change, extracts the change characteristics of the occluded area, and combines the global and local weighted fusion operations to obtain the global and local fusion feature value; The parking space abnormality matching module extracts the boundary coordinate set distribution parameters based on the global and local fusion feature values, calculates the matching value of the abnormal area boundary, adjusts the order of the abnormal area boundary points, extracts the geometric and position change parameters, generates the boundary features of the abnormal area, and establishes the shape constraint prediction value; The parking space shape reconstruction module extracts the boundary coordinate points of the occluded area based on the shape constraint prediction value, supplements the missing coordinate points with the offset, regenerates the boundary point set, calculates the geometric symmetry parameters of the contour, reconstructs the complete boundary distribution of the parking space contour, and obtains the offset dynamic prediction result; The parking space information integration module extracts the arrangement data and boundary contours in the multi-resolution image based on the offset dynamic prediction results, integrates the parking space arrangement and boundary data, completes coordinate mapping, generates complete data on the parking space arrangement and boundary, and establishes a parking space identification data table.

Images