A smart parking management system and method based on the Internet of Things

Abstract

The present invention relates to the field of parking lot management technology, and in particular to a smart parking management system and method based on the Internet of Things. The present invention proposes the following scheme: a parking space image is obtained by an image sensor, the parking space occupancy status is determined by using a neural network technology, and the best parking space is recommended according to the parking preference of the owner and the real-time parking space status. At the same time, the system adjusts the parking fee in real time according to the parking space occupancy rate, traffic flow and parking time of the owner through a dynamic pricing strategy. The monitoring module can remotely monitor the parking lot, detect and handle abnormal situations in a timely manner, and improve the efficiency and safety of parking management.

Description

Technical Field

[0001] The present invention relates to the technical field of parking lot management, and in particular to an intelligent parking management system and method based on the Internet of Things. Background Art

[0002] With the acceleration of urbanization, the number of vehicles in cities has increased rapidly, and parking has become an urgent problem to be solved. Traditional parking management systems mostly use manual management or automatic control based on a single sensor, which is difficult to meet the current urban parking needs. The development of Internet of Things technology has provided a new solution for smart parking management. By interconnecting the equipment, sensors, and communication modules in the parking lot, it can achieve real-time monitoring of vehicles, accurate perception of parking space status, and remote intelligent management.

[0003] The existing technologies are mainly deficient in the following aspects: First, the accuracy of parking space detection is limited, and it is impossible to achieve real-time and efficient parking space occupancy status detection. Second, parking space recommendations mainly rely on the car owner's own selection, lack of intelligent parking space allocation strategy, and it is difficult to meet personalized needs. Third, traditional parking fees are mostly fixed or simply charged on time, and cannot be flexibly adjusted according to supply and demand and parking space location, resulting in low parking lot resource utilization. Finally, the parking lot's monitoring system usually only provides basic monitoring functions and cannot effectively integrate the parking management system for abnormal detection and remote processing.

[0004] For example, the Chinese patent with the authorization announcement number CN109979233B discloses a parking lot planning system based on narrowband Internet of Things, which involves the technical field of parking lot information management and solves the problem that in the actual application process, when car users need to park, there are many parking lots around, but car users cannot effectively distinguish which parking lot takes the shortest time to park. It includes: a parking space vacancy detection device; a parking lot situation database, which stores the parking lot number, the location of the corresponding numbered parking lot, and the current number of vacancies in the corresponding numbered parking lot in real time; a car owner's car positioning device, which is used to locate the location of the car owner's car; a path planning device, which plans a route from the location of the car owner's car as the starting point to the location of the parking lot; a car owner's mobile phone number database, which stores the car owner's mobile phone number; and a text message sending device, which is used to send the car owner a corresponding reminder text message. The invention has the effect of helping car owners to choose the most suitable parking lot in a timely manner, so that the car owners can park in time.

[0005] The above patents all have the problems raised by this background technology: lack of accurate monitoring of parking space occupancy status and inability to achieve intelligent allocation of parking spaces. To solve the above problems, the present invention designs an intelligent parking management system and method based on the Internet of Things. Summary of the invention

[0006] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide an intelligent parking management system and method based on the Internet of Things. The parking space image is obtained through an image sensor, the parking space occupancy status is determined using neural network technology, and the best parking space is recommended based on the owner's parking preference and real-time parking space status. At the same time, the system uses a dynamic pricing strategy to adjust the parking fee in real time based on the parking space occupancy rate, traffic flow, and parking time of the owner. The monitoring module can remotely monitor the parking lot, detect and handle abnormal situations in a timely manner, and improve the efficiency and safety of parking management.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A smart parking management system based on the Internet of Things, the system includes a parking space detection module, a parking recommendation module, a navigation guidance module and a dynamic pricing module;

[0009] The parking space detection module is configured with a parking space status detection strategy, which acquires a parking space image through an image sensor, processes the parking space image, and generates occupancy status data;

[0010] The parking recommendation module is configured with a parking space allocation strategy, which analyzes the parking preference and parking time of the car owner and recommends a parking space to the car owner according to the occupancy status data;

[0011] The navigation guidance module is connected to the parking recommendation module, generates a parking navigation path according to the parking space, and guides the car owner to the parking space;

[0012] The dynamic pricing module is configured with an intelligent pricing strategy, which is used to dynamically adjust the parking fee according to the real-time status in the parking lot.

[0013] The parking space detection module comprises:

[0014] An image acquisition unit, used to acquire real-time images of parking spaces in the parking lot through an image sensor;

[0015] A target detection unit, used to process the real-time image of the parking space by image processing technology to detect whether there is a vehicle in the parking space area;

[0016] A parking space status analysis unit, used to analyze the detection result output by the target detection unit to generate occupancy status data of each parking space;

[0017] The parking space status detection strategy includes image processing logic and status judgment logic;

[0018] The image processing logic is used to perform edge detection on the real-time image of the parking space, crop the target area image, and identify the target area image through neural network technology to determine whether there is a vehicle in the area. The image processing logic is configured in the target detection unit;

[0019] The state judgment logic is used to analyze the recognition results of the image processing logic and determine the occupancy status of the parking space according to the detection situation of the vehicle. At the same time, when the detection result is unclear, the abnormal state mark is triggered and the image is requested to be re-captured. The state judgment logic is configured in the parking space state analysis unit.

[0020] The image processing logic comprises:

[0021] Performing weighted sampling on the real-time image of the parking space through an upsampling function, identifying the real-time image of the parking space after weighted sampling through an edge detection operator, and extracting a regional contour map;

[0022] The regional contour map is divided into a foreground area and a background area according to a mixed Gaussian model, and the foreground area is expanded and eroded to calculate the contour feature, wherein the calculation formula of the contour feature is:

[0023] ,

[0024] in, Represents the contour features of the foreground area, i represents a single contour point in the foreground area, and n represents the total number of contour points in the foreground area. Represents the weighting coefficient, which is related to the distance between the i-th contour point and the contour center. represents the edge strength of the i-th contour point, X represents the foreground area of ​​the regional contour map, represents a single curvature point in the foreground area, m represents the total number of curvature points in the foreground area, represents the curvature value of the jth curvature point, represents the normalized value of the local edge strength of the jth curvature point, represents the curvature adjustment parameter of the j-th curvature point, represents the corrosion operation, represents the expansion operation;

[0025] A vehicle recognition network is constructed, and the contour features are used as input parameters of the vehicle recognition network. The vehicle recognition network is trained to determine whether there is a vehicle in the parking area, and a recognition result is output.

[0026] The vehicle identification network comprises:

[0027] The sampling layer is used to expand the plane dimension of the input parameters according to the bilinear interpolation, and transform the contour features of different levels and scales into a unified scale;

[0028] The convolution layer is used to expand the channel of the output of the sampling layer and fuse the output of the sampling layer using a 1×1 convolution fusion channel;

[0029] The detection layer is used to activate the output of the convolution layer, perform error back propagation on the output of the convolution layer after the activation processing according to the loss function, and output the recognition result after reaching the convergence condition, wherein the activation processing includes plane dimension activation mapping and stereo dimension activation mapping.

[0030] The parking recommendation module includes:

[0031] Parking data analysis unit, used to analyze the parking preferences of car owners and predict parking time;

[0032] An occupancy status matching unit, configured to output a parking space to be matched according to the output result of the parking data analysis unit and the occupancy status data;

[0033] A parking space priority calculation unit, used to perform priority scoring on the parking spaces to be matched by using a weighted algorithm;

[0034] A personalized recommendation unit, used to recommend the parking space with the highest priority as the best parking space to the car owner according to the parking space priority calculation result;

[0035] The parking space allocation strategy includes data analysis logic, free parking space matching logic and parking space recommendation logic;

[0036] The data analysis logic is configured in the parking data analysis unit, and is used to analyze the parking habits of the vehicle owner based on the vehicle owner's historical parking data, and predict the vehicle owner's parking time;

[0037] The free parking space matching logic is configured in the occupancy status matching unit, and is used to filter out free parking spaces from the occupancy status data, and further match the free parking spaces to obtain parking spaces to be matched;

[0038] The parking space recommendation logic is configured in the parking space priority calculation unit and is used to perform priority scoring on the parking spaces to be matched.

[0039] The parking space recommendation logic includes:

[0040] Performing occupancy status detection on both sides of the to-be-matched parking space and calculating parking difficulty;

[0041] Generate the shortest path from the entrance to the parking space to be matched according to the path planning algorithm, and calculate the accessibility of the parking space to be matched according to the shortest path;

[0042] Calculate parking safety based on the monitoring coverage and lighting conditions of the parking spaces to be matched;

[0043] The parking space priority is calculated by the priority calculation formula, and the priority calculation formula is:

[0044] ,

[0045] in, Indicates the parking space priority of the parking space to be matched, K indicates the parking suggestion index, if there are no vehicles or obstacles on both sides, K=3, if there are vehicles or obstacles on one side, K=2, if there are vehicles or obstacles on both sides, K=1, represents the impact coefficient of parking difficulty, Indicates the width of the area to be matched. Indicates the distance from the left line of the parking space to be matched to the nearest obstacle. Indicates the distance from the right line of the parking space to be matched to the nearest obstacle. Indicates the angle correction coefficient of the adjacent vehicle. If there is no adjacent vehicle, =1, represents the influence coefficient of accessibility, Indicates the shortest path length from the entrance to the parking space to be matched, represents the number of turns in the shortest path, represents the turning weight coefficient, Indicates the impact coefficient of parking safety, Indicates the coverage of the surveillance camera. Indicates the parking space lighting intensity of the parking space to be matched. Represents historical risk records, Represents a constant greater than 0.

[0046] The dynamic pricing module includes:

[0047] Real-time data collection unit, used to collect real-time data from various sensors and systems in the parking lot, including parking space occupancy, vehicle flow, parking duration and parking space area distribution information;

[0048] The pricing unit is used to analyze the supply and demand of parking spaces in the parking lot and dynamically adjust the parking fees in different time periods and areas;

[0049] A settlement unit, used to perform real-time settlement based on the output of the pricing unit and provide parking fee information when the car owner leaves the parking space;

[0050] The intelligent pricing strategy includes a dynamic pricing logic, which is configured in the pricing unit and is used to dynamically adjust the parking fee according to the real-time situation of the parking lot.

[0051] The dynamic pricing logic includes:

[0052] By collecting the parking space occupancy rate and vehicle flow of the parking lot, the current supply and demand relationship of the parking lot is analyzed, and the parking fee is adjusted according to the analysis results;

[0053] adjusting parking fees according to the location of the parking space in the parking lot;

[0054] Through the tiered pricing method, the parking fee is adjusted according to the parking time of the car owner.

[0055] The system also includes a monitoring module, which is used to monitor the situation in the parking lot and provide a remote monitoring function to the parking lot manager, specifically including:

[0056] An anomaly detection unit, which is used to determine whether an abnormal situation occurs based on real-time data collected from various sensors and systems in the parking lot;

[0057] A notification unit for immediately sending an alert to the parking lot manager when an abnormal situation is detected;

[0058] The remote monitoring unit is used to provide the functions of remotely monitoring parking lot conditions and viewing real-time monitoring video parking spaces through smart devices.

[0059] A smart parking management method based on the Internet of Things, the specific steps of the method are as follows:

[0060] S1: Collect parking space images in the parking lot through image sensors to obtain real-time image data of each parking space;

[0061] S2: Process the real-time image data, identify the target area image through neural network technology, and generate the occupancy status data of each parking space;

[0062] S3: Based on the occupancy status data, combined with the owner's parking preferences and historical parking time, the best parking space is recommended to the owner through the parking space allocation strategy;

[0063] S4: generating a navigation path for the vehicle owner to the recommended parking space and providing real-time guidance information;

[0064] S5: adjusting parking fees based on real-time parking space occupancy, parking duration, and parking space location, and providing updated charging information to the car owner;

[0065] S6: The parking lot is monitored in real time through the remote monitoring system. If any abnormal situation is detected, an alarm will be issued and the management staff will be notified.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] 1. The present invention uses image sensors and neural network technology to achieve high-precision detection of the occupancy status of parking spaces in parking lots, and can determine in real time whether a parking space is occupied, thereby improving the accuracy and efficiency of parking space detection;

[0068] 2. The present invention recommends the best parking space to the car owner based on the car owner's parking preference and historical parking data, combined with the parking space occupancy status and parking difficulty through a weighted algorithm. The personalized recommendation improves the convenience of parking. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0070] Figure 1 This is a flow chart of a smart parking management method based on the Internet of Things according to Embodiment 1 of the present invention;

[0071] Figure 2 This is a flow chart of parking space occupancy status detection in Example 1 of the present invention;

[0072] Figure 3 This is a diagram of a vehicle identification network structure in accordance with Embodiment 1 of the present invention;

[0073] Figure 4 This is a flow chart of the parking space allocation strategy in Example 1 of the present invention;

[0074] Figure 5 This is a structural diagram of a weighted scoring model according to Embodiment 1 of the present invention;

[0075] Figure 6 This is a schematic diagram of an intelligent parking management system based on the Internet of Things according to Example 2 of the present invention. DETAILED DESCRIPTION

[0076] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0077] Example 1

[0078] See also Figure 1 The present invention provides an embodiment: a smart parking management method based on the Internet of Things, the specific steps of the method are as follows:

[0079] S1: Collect parking space images in the parking lot through image sensors to obtain real-time image data of each parking space;

[0080] In this step, the real-time image data of each parking space is collected by the image sensor installed in the parking lot. In order to improve the image quality, it may be necessary to perform pre-processing operations such as denoising and white balance correction on the image. At the same time, the frequency of image acquisition can be set to increase the acquisition frequency during peak hours to achieve more accurate monitoring;

[0081] S2: Process the real-time image data, identify the target area image through neural network technology, and generate the occupancy status data of each parking space;

[0082] In this step, the collected real-time image data is processed. First, the target area is extracted using edge detection and image cropping techniques. Then, the target area image is recognized through neural network technology (such as convolutional neural network) to determine whether there is a vehicle in the parking area. The recognition results include status data such as "occupied", "free" or "abnormal";

[0083] S3: Based on the occupancy status data, combined with the owner's parking preferences and historical parking time, the best parking space is recommended to the owner through the parking space allocation strategy;

[0084] In this step, the parking space allocation strategy analyzes the owner's historical parking behavior and preferences (such as the usual parking area, parking duration, etc.), and makes intelligent recommendations based on the real-time occupancy status of the parking space. In addition, the owner's destination location (such as close to the exit, shopping area, etc.) can be prioritized to recommend suitable parking spaces, and short-term or long-term parking spaces can be optimized based on the parking duration. If multiple parking spaces meet the recommendation criteria, a weighted algorithm is used to comprehensively consider factors such as the distance to the parking space, the surrounding traffic density, and safety to recommend the best parking space for the owner.

[0085] S4: generating a navigation path for the vehicle owner to the recommended parking space and providing real-time guidance information;

[0086] S5: adjusting parking fees based on real-time parking space occupancy, parking duration, and parking space location, and providing updated charging information to the car owner;

[0087] In this step, the pricing strategy is adjusted based on a variety of factors, such as parking space location, parking time period, vehicle type, and real-time parking demand. For example, parking spaces near entrances or busy areas can be charged higher, while parking spaces in remote locations can be charged lower. In addition, a time-sharing pricing mechanism can be introduced, with higher charges during peak hours and discounts during off-peak hours. For parking spaces dedicated to electric vehicle charging, the fees can also be adjusted based on the vehicle's charging status and remaining charging time;

[0088] S6: When the vehicle arrives at or leaves the parking space, the parking space image is recaptured and the state detection process of S2 is executed to update the occupancy state data of the parking space in real time;

[0089] S7: Real-time monitoring of the parking lot is carried out through the remote monitoring system. If any abnormal situation is detected, an alarm is issued and the management personnel are notified;

[0090] In this step, abnormal situations include:

[0091] Illegal occupation of parking spaces: parking spaces are occupied by unauthorized vehicles;

[0092] Vehicle collision: Physical collision between vehicles is detected by sensors;

[0093] Overtime parking: The vehicle occupies a parking space for a long time, exceeding the scheduled time limit;

[0094] Security anomalies: such as abnormal activities (intrusion, vandalism, etc.) occurring in the parking lot.

[0095] See also Figure 2 , the parking space occupancy status detection flow chart of the embodiment of the present invention, the specific steps of S2 are as follows:

[0096] S2.1: Background modeling is performed through a mixed Gaussian model. The model can be trained with a large number of images of empty parking spaces to establish a background model of parking spaces, and adaptively update according to the dynamic changes of lighting, weather, and vehicles. When there are large differences between the newly collected image and the background model, the model marks these differences as foreground, which may represent the presence of vehicles;

[0097] The mixed Gaussian model fits the background with a multivariate Gaussian distribution and can capture changes in the parking lot background, such as changes in lighting and weather. For each pixel, the model treats it as coming from a different Gaussian distribution and determines whether it is part of the background or foreground by estimating the Gaussian distribution to which each pixel belongs. When a pixel deviates significantly from a known Gaussian distribution, the pixel will be marked as foreground. By continuously updating and learning new background distributions, the model can continue to adapt to new environmental changes.

[0098] S2.2: After detecting the foreground area, use the Canny edge detection algorithm to detect the edge of the vehicle. The algorithm calculates the grayscale gradient of the image, identifies the area with significant grayscale value changes, and extracts the contour edge of the vehicle;

[0099] The Canny algorithm is a classic and effective edge detection algorithm with strong noise immunity. In parking lot scenes, there is a clear grayscale difference between the edge of the vehicle and the background information such as the parking space lines. The Canny algorithm can accurately extract the vehicle outline and remove noise interference at the same time.

[0100] Canny edge detection first calculates the grayscale gradient of the image in the x-axis and y-axis directions through the Sobel operator, and then uses non-maximum suppression technology to retain only pixels with significant gradient changes and remove unnecessary edges. By setting low and high thresholds, the Canny algorithm can balance noise removal and edge detail retention, filtering out false edges while retaining key vehicle edge information;

[0101] S2.3: After the preliminary contour is obtained by Canny edge detection, the contour is further processed using morphological operations. The dilation operation is used to expand the foreground contour to make it more complete, while the erosion operation removes small noise points to ensure the smoothness of the edge;

[0102] Especially in the complex lighting environment of parking lots, morphological processing can keep the vehicle outline clear and complete under complex backgrounds and light interference, effectively solving problems such as light reflection in parking lot environments;

[0103] Morphological processing mainly relies on pixel neighborhood operations in the image. The dilation operation expands the boundary points in the target area to its neighboring pixels, making the edge more obvious and filling the gaps inside the edge; while the erosion operation is the opposite, by cutting the pixels in the boundary area to remove isolated noise or impurities. The combination of the two can effectively retain the outline of the foreground object and remove irrelevant noise;

[0104] S2.4: After the vehicle outline is detected, the outline feature is calculated according to the foreground area of ​​each parking space. The calculation formula of the outline feature is:

[0105] ,

[0106] in, Represents the contour features of the foreground area, mainly including the strength and curvature of the vehicle edge. i represents a single contour point in the foreground area, and n represents the total number of contour points in the foreground area. It represents the weighting coefficient, which is related to the distance between the i-th contour point and the contour center. It is used to adjust the influence of each edge point in the overall contour feature. Usually, the edge points of the vehicle contour have a large weight, while the edge points of the parking space line have a small weight. represents the edge strength of the i-th contour point, specifically the grayscale change of the contour point relative to the surrounding pixels. The edge strength is calculated based on the gradient information of the contour point pixels. X represents the foreground area of ​​the regional contour map. represents a single curvature point in the foreground area, m represents the total number of curvature points in the foreground area, Indicates the curvature value of the jth curvature point, describing the degree of curvature at the contour point. The curvature is calculated by analyzing the change in the normal direction of the contour point. The larger the curvature, the more obvious the curvature of the area. It is particularly suitable for identifying complex parts such as wheels. Represents the normalized value of the local edge strength of the jth curvature point, which prevents the difference in eigenvalues ​​caused by the different image scales of different parking spaces. Normalization ensures the consistency of the model in different environments. represents the curvature adjustment parameter of the jth curvature point to avoid unstable fluctuations in curvature due to noise or edge detection errors. Represents the erosion operation, which is used to enhance the integrity of the foreground area contour and make the detected vehicle edge smoother. represents the dilation operation, which is used to suppress the influence of the background on the foreground area and ensure that feature extraction is performed only on the vehicle contour;

[0107] S2.5: construct a vehicle recognition network, and use the contour features as input parameters of the vehicle recognition network, train the vehicle recognition network, determine whether there is a vehicle in the parking area, and output a recognition result;

[0108] Construct a vehicle recognition model based on a convolutional neural network, and use the previously extracted contour features as the input parameters of the network. The vehicle recognition network learns how to determine whether a parking space is occupied based on the contour features through training data. The use of convolutional neural networks can effectively extract features in images. The depth and complexity of the network layer can enhance the recognition ability of the system. Through supervised learning with a large amount of annotated data, the model can maintain high accuracy under different light and weather conditions. Convolutional neural networks are a multi-layer neural network that is good at processing image-related problems. Through convolution operations, CNN can extract features at different levels from the input image. The first few layers of the network extract lower-level features (such as edges, textures, etc.), while the latter few layers extract more complex high-level features (such as shapes, contours, etc.). Through training with a large amount of annotated data, CNN can learn how to accurately identify vehicles based on the input contour features and output the parking space occupancy status.

[0109] See also Figure 3 , a vehicle identification network structure diagram of an embodiment of the present invention, wherein the vehicle identification network comprises:

[0110] The sampling layer is used to describe the contour features through the elliptical Fourier descriptor, expand the plane dimension of the input parameters according to the bilinear difference, and transform the contour features of different levels and scales into a unified scale;

[0111] The main task of this layer is to process the contour features of the vehicle, unify their scale and adapt to subsequent convolution processing. The elliptical Fourier descriptor is a common tool for contour analysis. By decomposing the contour curve into a series of Fourier coefficients, it can accurately describe the edge shape of the vehicle, effectively capture the global and local features of the vehicle contour, and has robustness to geometric transformations such as scaling, rotation, and translation. The elliptical Fourier descriptor is invariant, and even if the perspective or scale of the vehicle in the image changes, the descriptor can still maintain a good representation of the contour. This is crucial for the robustness of vehicle recognition, especially in parking lot scenarios where vehicles may be at different perspectives;

[0112] After the Fourier descriptor calculation is completed, the contour features may have the problem of different scales. In order to solve the scale differences between different vehicle contours, the bilinear interpolation algorithm is used to interpolate the input feature map. The bilinear interpolation algorithm estimates the new pixel value through the linear combination of four neighboring pixels. It can map the pixel values ​​in the original image to a plane of uniform scale, ensuring that the subsequent network can process vehicle features of different sizes.

[0113] The convolution layer is used to further process the output of the sampling layer. Its main task is to perform channel expansion and fusion of features. Through the 1×1 convolution operation, it can efficiently fuse the contour features from the sampling layer and generate feature maps suitable for the subsequent detection layer;

[0114] The convolution layer will first expand the channels of the sampling layer output, that is, increase the number of feature channels, so as to capture more contour details and local features. This process is similar to the dimension increase of feature maps. By increasing the representation ability of features, the network can better distinguish the subtle differences between vehicles, such as the shape of different models, license plate positions, headlight shapes, etc.

[0115] The convolution layer uses a 1×1 convolution kernel. The advantage of this operation is that different channels can be fused and reduced in dimension without changing the spatial size of the feature map. For vehicle recognition tasks, 1×1 convolution can combine the features extracted by different convolution kernels and learn the importance of different features through weights to achieve the purpose of information integration;

[0116] For example, when the sampling layer extracts different contour features of the vehicle (such as the edge of the body, headlights, roof, etc.), the convolution layer can fuse these different features through 1×1 convolution to generate a comprehensive feature map with higher expressive power. Combining the global information of the contour with the local detail information enables the network to learn more discriminative features on different parts of the vehicle.

[0117] The detection layer is used to perform activation processing on the output of the convolution layer, perform error back propagation on the output of the convolution layer after the activation processing according to the loss function, and output the recognition result after reaching the convergence condition, wherein the activation processing includes plane dimension activation mapping and stereo dimension activation mapping;

[0118] First, the detection layer activates the output feature map of the convolution layer on a two-dimensional plane, and uses the activation function to perform nonlinear mapping on each pixel of the feature map. Through this operation, the network can capture the nonlinear relationship of the vehicle contour features and enhance the ability to represent complex shapes;

[0119] In addition to activating in the plane dimension, the detection layer also processes in the three-dimensional dimension of the feature map, that is, it fuses the features of different channels. Through three-dimensional activation mapping, the detection layer can capture the spatial structure information of the vehicle, such as the relative position relationship between the roof, body and wheels, further improving the recognition ability of the network;

[0120] The detection layer uses the classification loss function to evaluate the gap between the network output and the true label. Through the error back propagation mechanism, the network can gradually adjust the parameters of the sampling layer and the convolution layer so that the final recognition result converges to the optimal solution. In vehicle recognition applications, the choice of loss function can be combined with the accuracy and precision of detection. For example, license plate recognition requires higher precision, while overall vehicle classification focuses on the correct identification of categories.

[0121] The detection layer enhances the network's ability to identify the overall contour and local features of the vehicle through activation mapping in the plane and three-dimensional dimensions. Through the error back-propagation mechanism, the network can continuously optimize parameters, so that the accuracy of vehicle recognition is gradually improved. Especially in traffic scenarios, the detection layer can output the category, location and contour information of the vehicle based on the input feature map, providing accurate data support for applications such as autonomous driving and smart parking.

[0122] The calculation formula of the loss function is:

[0123] ,

[0124] Among them, Loss represents the loss function, represents the coordinate loss of the convolutional layer output after activation processing, represents the coordinate loss weight, represents the confidence loss of the convolutional layer output after activation processing, represents the confidence loss weight, represents the category loss of the convolutional layer output after activation processing, represents the category loss weight;

[0125] See also Figure 4, a flow chart of a parking space allocation strategy according to an embodiment of the present invention, wherein the parking space allocation strategy provides personalized parking space recommendations for the car owner by analyzing the car owner's historical parking data, real-time parking space status, and multiple attributes of the parking space. The specific steps of the parking space allocation strategy are as follows:

[0126] S3.1: Establish a parking behavior database based on the owner's historical parking records (such as the parking areas, parking duration, and parking frequency that the owner often chooses). By collecting the owner's parking history data over a long period of time, including the owner's parking behavior in different scenarios (such as parking preferences on weekdays, weekends, and shopping malls), the estimated parking time of the owner is predicted based on the time series analysis model (such as short parking on weekdays and long parking on weekends);

[0127] For example, if it is detected that the parking time of car owners on weekdays is usually 1-2 hours, while the parking time on weekends is longer, 4-5 hours, predictions will be made based on the time period to optimize parking space recommendations. The parking time is predicted in advance so that the system can better match parking spaces. For example, for short-term parking owners, parking spaces close to the exit will be recommended first, while for long-term parking owners, parking spaces farther away but with more vacant spaces will be recommended to balance parking resources;

[0128] S3.2: Collect the occupancy status of parking spaces in the parking lot in real time, generate occupancy status data of parking spaces (free, occupied, faulty), combine the parking behavior database (such as proximity to exits, parking duration, etc.) with the real-time occupancy status data, and select parking spaces that meet the needs of car owners;

[0129] The parking space status of the parking lot is updated in real time through IoT devices to ensure that the recommended parking spaces are free and fault-free. Based on the parking behavior preferences of the car owner (for example, the car owner prefers parking spaces near the exit), the matching algorithm is used to screen the qualified parking spaces. If some areas of the parking lot are under maintenance, these parking spaces are detected as "faulty" and the parking spaces near the exit are detected as occupied. Based on the car owner's preferences (for example, he likes to be near the elevator), faulty and occupied parking spaces are excluded and the free parking spaces closest to the elevator are recommended.

[0130] S3.3: A weighted scoring model is used to assign different weights to various attributes of each parking space to be matched (such as parking space distance, parking space width, and parking space surrounding environment) to calculate the priority score of each parking space;

[0131] The weighted scoring model improves the accuracy of recommendations. For example, some car owners may be more concerned about the spaciousness of the parking space, while others may be more concerned about the safety of the parking space. By flexibly adjusting the weights, the most suitable parking space can be provided to each car owner;

[0132] S3.4: Based on the priority score of each parking space, the parking space with the highest score is selected and recommended to the car owner. The recommendation results are displayed in an intuitive way, such as displaying the specific location of the parking space, the distance between the parking space and the destination, and generating a navigation path from the parking lot entrance to the recommended parking space in the car owner's mobile app. The car owner can view the navigation information in real time through the mobile device or the in-vehicle system.

[0133] See also Figure 5 , a structural diagram of a weighted scoring model according to an embodiment of the present invention, wherein the weighted scoring model comprises:

[0134] S3.3.1: Detect the occupancy status of both sides of the parking space to be matched and calculate the parking difficulty;

[0135] First, the occupancy status on both sides of the parking space to be matched is detected. If there are vehicles or obstacles on both sides of the parking space, the difficulty of parking will increase significantly. By combining sensors (such as ultrasonic sensors) with image processing technology, the occupancy status of adjacent parking spaces is detected in real time, and the parking difficulty is calculated. The difficulty of parking depends not only on the presence of vehicles on both sides, but also on the distance between the vehicle and the parking space. The adjacent occupancy status of the parking space is collected in real time through sensors, and the distance sensor is used to determine the distance between the vehicles on both sides of the parking space. Based on the principle that the closer the distance between the vehicles on both sides, the greater the difficulty of parking, the parking difficulty coefficient is calculated;

[0136] Through accurate calculation of parking difficulty, the usability of parking spaces can be effectively evaluated, and more appropriate recommendations can be provided to car owners. For car owners who are not proficient in parking skills, parking spaces with low parking difficulty can be recommended first;

[0137] S3.3.2: Generate the shortest path from the entrance to the parking space to be matched according to the path planning algorithm, and calculate the accessibility of the parking space to be matched according to the shortest path. The parking space with higher path complexity has poorer accessibility, and its score will be reduced in the priority score;

[0138] S3.3.3: Calculate parking safety according to the parking environment of the parking space to be matched, wherein the parking environment includes monitoring coverage and parking lighting conditions;

[0139] The safety of a parking space is calculated using camera monitoring coverage and the surrounding environment (such as whether there are obstacles and whether the lighting is good). If the parking space is in the blind spot of the surveillance camera or the lighting is poor, the safety score will be low;

[0140] S3.3.4: Calculate the parking space priority using a priority calculation formula, the priority calculation formula is:

[0141] ,

[0142] in, Indicates the parking space priority of the parking space to be matched, K indicates the parking suggestion index, if there are no vehicles or obstacles on both sides, K=3, if there are vehicles or obstacles on one side, K=2, if there are vehicles or obstacles on both sides, K=1, Indicates the parking difficulty coefficient, which is used to measure the parking difficulty. The wider the parking space, the easier it is to park. The larger the value, Indicates the width of the area to be matched. Indicates the distance from the left line of the parking space to be matched to the nearest obstacle. Indicates the distance from the right line of the parking space to be matched to the nearest obstacle. Represents the angle correction coefficient of the adjacent vehicle, which is used to measure the angle adaptability between the adjacent vehicle and the parking space. If it is completely parallel to the parking space, then =1, if there is an angle deviation, the larger the deviation angle, the The smaller the value, the more likely it is that there is no adjacent vehicle. =1, Represents the accessibility impact coefficient, which is used to measure the complexity of the path to the parking space. The more complex the path, The lower the value, Indicates the shortest path length from the entrance to the parking space to be matched, represents the number of turns in the shortest path, represents the turning weight coefficient, which can be obtained by those skilled in the art through a large number of experiments. Indicates the impact coefficient of parking safety, which is used to measure the safety of parking spaces. Indicates the coverage of the surveillance camera. Indicates the parking space lighting intensity of the parking space to be matched. Indicates historical risk records, indicating the number of accidents that have occurred in the parking space in the past. It reduces the priority of parking spaces with high risks and high accident rates, and gives priority to recommending parking spaces in safe areas. Represents a constant greater than 0.

[0143] The specific steps of S5 are as follows:

[0144] S5.1: Collect real-time data from various sensors and systems in the parking lot, including parking space occupancy, vehicle flow, parking duration, and parking space regional distribution information;

[0145] Parking space occupancy: Use IoT devices to detect whether each parking space is occupied, free, or faulty;

[0146] Traffic flow: record the number of vehicles entering and leaving the parking lot during each time period;

[0147] Parking time: Automatically record the parking time of each vehicle as the basis for fee adjustment;

[0148] Parking space area distribution: Parking spaces are classified according to the area where they are located (such as near exits, shopping malls, etc.);

[0149] S5.2: Analyze the current supply and demand relationship of the parking lot based on the parking space occupancy rate and vehicle flow in the parking lot, and adjust the parking fee based on the analysis results;

[0150] When the parking space occupancy rate is high, the supply and demand are tight, and the parking fee will be increased; when the parking space is sufficient, the fee will be appropriately reduced to attract more car owners to enter the parking lot. When the parking space occupancy rate exceeds 80%, the "supply and demand are tight" mode will be activated, and the parking fee will increase at a higher rate; if the occupancy rate is less than 50%, the "off-peak period" pricing strategy will be activated to reduce the parking fee;

[0151] Automatically adjust prices based on real-time supply and demand changes, and flexibly adjust parking fees using supply and demand analysis formulas to avoid resource waste caused by fixed price models;

[0152] S5.3: adjusting the parking fee according to the location of the parking space in the parking lot;

[0153] Differentiate the pricing based on the location of the parking space in the parking lot. Parking spaces near the mall entrance, elevator or exit are usually in higher demand, so the price is relatively higher; while parking spaces far away from these key locations are cheaper, so as to balance the utilization of parking spaces in different areas;

[0154] Automatically set differentiated prices for parking spaces in different areas based on the geographical location of the parking lot to make full use of parking space resources in each area. Especially during peak parking hours, the price of parking spaces near elevators or exits will increase further;

[0155] Through differentiated pricing, the utilization rate of parking spaces in different areas of the parking lot can be balanced, preventing car owners from concentrating on parking spaces near the entrance or exit of the shopping mall, thereby improving the overall utilization rate of the parking lot;

[0156] S5.4: Adjust parking fees based on the length of time a vehicle owner parks by using a tiered pricing approach;

[0157] The price of short-term parking (e.g. within 1 hour) is relatively low to encourage short-term parking; while the fee for long-term parking will increase with the length of time, encouraging car owners to choose the parking time reasonably according to their needs. For short-term parking (within 1 hour), only the basic fee may be charged; for more than 2 hours, the tiered pricing will be activated, and the parking fee will increase by a certain percentage for each additional hour;

[0158] Through flexible time control, short-term parking is encouraged to avoid long-term occupancy of parking spaces. This pricing model is particularly suitable for parking lots during peak hours or with high demand. It improves the mobility of parking spaces and effectively prevents parking spaces from being occupied for a long time. For car owners who park for a short time, the price is lower to encourage them to turn over quickly; for long-term parking, the price increases progressively to ensure that the parking lot income is linked to the length of parking time.

[0159] Example 2

[0160] See also Figure 6 , the present invention provides an embodiment: a smart parking management system based on the Internet of Things, the system comprising a parking space detection module, a parking recommendation module, a navigation guidance module, a dynamic pricing module and a monitoring module;

[0161] The parking space detection module is configured with a parking space status detection strategy, which acquires a parking space image through an image sensor, processes the parking space image, and generates occupancy status data;

[0162] The parking recommendation module is configured with a parking space allocation strategy, which analyzes the parking preference and parking time of the car owner and recommends a parking space to the car owner according to the occupancy status data;

[0163] The navigation guidance module is connected to the parking recommendation module, generates a parking navigation path according to the parking space, and guides the car owner to the parking space;

[0164] The dynamic pricing module is configured with an intelligent pricing strategy, which is used to dynamically adjust the parking fee according to the real-time status in the parking lot;

[0165] The monitoring module is used to monitor the situation in the parking lot.

[0166] The parking space detection module comprises:

[0167] An image acquisition unit, used to acquire real-time images of parking spaces in the parking lot through an image sensor;

[0168] A target detection unit, used to process the real-time image of the parking space by image processing technology to detect whether there is a vehicle in the parking space area;

[0169] A parking space status analysis unit, used to analyze the detection result output by the target detection unit to generate occupancy status data of each parking space;

[0170] The parking space status detection strategy includes image processing logic and status judgment logic;

[0171] The image processing logic is used to perform edge detection on the real-time image of the parking space, crop the target area image, and identify the target area image through neural network technology to determine whether there is a vehicle in the area. The image processing logic is configured in the target detection unit;

[0172] The state judgment logic is used to analyze the recognition results of the image processing logic and determine the occupancy status of the parking space according to the detection situation of the vehicle. At the same time, when the detection result is unclear, the abnormal state mark is triggered and the image is requested to be re-captured. The state judgment logic is configured in the parking space state analysis unit.

[0173] The parking recommendation module includes:

[0174] Parking data analysis unit, used to analyze the parking preferences of car owners and predict parking time;

[0175] An occupancy status matching unit, configured to output a parking space to be matched according to the output result of the parking data analysis unit and the occupancy status data;

[0176] A parking space priority calculation unit, used to perform priority scoring on the parking spaces to be matched by using a weighted algorithm;

[0177] A personalized recommendation unit, used to recommend the parking space with the highest priority as the best parking space to the car owner according to the parking space priority calculation result;

[0178] The parking space allocation strategy includes data analysis logic, free parking space matching logic and parking space recommendation logic;

[0179] The data analysis logic is configured in the parking data analysis unit, and is used to analyze the parking habits of the vehicle owner based on the vehicle owner's historical parking data, and predict the vehicle owner's parking time;

[0180] The free parking space matching logic is configured in the occupancy status matching unit, and is used to filter out free parking spaces from the occupancy status data, and further match the free parking spaces to obtain parking spaces to be matched;

[0181] The parking space recommendation logic is configured in the parking space priority calculation unit and is used to perform priority scoring on the parking spaces to be matched.

[0182] The dynamic pricing module includes:

[0183] Real-time data collection unit, used to collect real-time data from various sensors and systems in the parking lot, including parking space occupancy, vehicle flow, parking duration and parking space area distribution information;

[0184] The pricing unit is used to analyze the supply and demand of parking spaces in the parking lot and dynamically adjust the parking fees in different time periods and areas;

[0185] A settlement unit, used to perform real-time settlement based on the output of the pricing unit and provide parking fee information when the car owner leaves the parking space;

[0186] The intelligent pricing strategy includes a dynamic pricing logic, which is configured in the pricing unit and is used to dynamically adjust the parking fee according to the real-time situation of the parking lot.

[0187] The monitoring module specifically includes:

[0188] An anomaly detection unit, which is used to determine whether an abnormal situation occurs based on real-time data collected from various sensors and systems in the parking lot;

[0189] A notification unit for immediately sending an alert to the parking lot manager when an abnormal situation is detected;

[0190] The remote monitoring unit is used to remotely monitor the parking lot situation through smart devices and view real-time monitoring video parking spaces.

[0191] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A smart parking management system based on the Internet of Things, characterized in that: The system includes a parking space detection module, a parking recommendation module, a navigation guidance module and a dynamic pricing module; The parking space detection module is configured with a parking space status detection strategy, which acquires a parking space image through an image sensor, processes the parking space image, and generates occupancy status data; The parking recommendation module is configured with a parking space allocation strategy, which analyzes the parking preference and parking time of the car owner and recommends a parking space to the car owner according to the occupancy status data, wherein the parking space allocation strategy includes data analysis logic, free parking space matching logic and parking space recommendation logic; The navigation guidance module is connected to the parking recommendation module, generates a parking navigation path according to the parking space, and guides the car owner to the parking space; The dynamic pricing module is configured with an intelligent pricing strategy, which is used to dynamically adjust the parking fee according to the real-time status in the parking lot; The parking space recommendation logic includes: Detect the occupancy status of both sides of the matching parking space and calculate the parking difficulty; Generate the shortest path from the entrance to the parking space to be matched according to the path planning algorithm, and calculate the accessibility of the parking space to be matched according to the shortest path; Calculate parking safety based on the monitoring coverage and lighting conditions of the parking spaces to be matched; The parking space priority is calculated by the priority calculation formula, and the priority calculation formula is: ， in, Indicates the parking space priority of the parking space to be matched, K indicates the parking suggestion index, if there are no vehicles or obstacles on both sides, K=3, if there are vehicles or obstacles on one side, K=2, if there are vehicles or obstacles on both sides, K=1, represents the impact coefficient of parking difficulty, Indicates the width of the area to be matched. Indicates the distance from the left line of the parking space to be matched to the nearest obstacle. Indicates the distance from the right line of the parking space to be matched to the nearest obstacle. Indicates the angle correction coefficient of the adjacent vehicle. If there is no adjacent vehicle, =1, represents the influence coefficient of accessibility, Indicates the shortest path length from the entrance to the parking space to be matched, represents the number of turns in the shortest path, represents the turning weight coefficient, Indicates the impact coefficient of parking safety, Indicates the coverage of the surveillance camera. Indicates the parking space lighting intensity of the parking space to be matched. Represents historical risk records, Represents a constant greater than 0.

2. According to claim 1, a smart parking management system based on the Internet of Things is characterized in that: The parking space detection module comprises: An image acquisition unit, used to acquire real-time images of parking spaces in the parking lot through an image sensor; A target detection unit, used to process the real-time image of the parking space by image processing technology to detect whether there is a vehicle in the parking space area; A parking space status analysis unit, used to analyze the detection result output by the target detection unit to generate occupancy status data of each parking space; The parking space status detection strategy includes image processing logic and status judgment logic; The image processing logic is used to perform edge detection on the real-time image of the parking space, crop the target area image, and identify the target area image through neural network technology to determine whether there is a vehicle in the area. The image processing logic is configured in the target detection unit; The state judgment logic is used to analyze the recognition results of the image processing logic and determine the occupancy status of the parking space according to the detection situation of the vehicle. At the same time, when the detection result is unclear, the abnormal state mark is triggered and the image is requested to be re-captured. The state judgment logic is configured in the parking space state analysis unit.

3. According to claim 2, a smart parking management system based on the Internet of Things is characterized in that: The image processing logic comprises: Performing weighted sampling on the real-time image of the parking space through an upsampling function, identifying the real-time image of the parking space after weighted sampling through an edge detection operator, and extracting a regional contour map; The regional contour map is divided into a foreground area and a background area according to a mixed Gaussian model, and the foreground area is expanded and eroded to calculate the contour feature, wherein the calculation formula of the contour feature is: ， in, Represents the contour features of the foreground area, i represents a single contour point in the foreground area, and n represents the total number of contour points in the foreground area. Represents the weighting coefficient, which is related to the distance between the i-th contour point and the contour center. represents the edge strength of the i-th contour point, X represents the foreground area of ​​the regional contour map, represents a single curvature point in the foreground area, m represents the total number of curvature points in the foreground area, represents the curvature value of the jth curvature point, represents the normalized value of the local edge strength of the jth curvature point, represents the curvature adjustment parameter of the j-th curvature point, represents the corrosion operation, represents the expansion operation; A vehicle recognition network is constructed, and the contour features are used as input parameters of the vehicle recognition network. The vehicle recognition network is trained to determine whether there is a vehicle in the parking area, and a recognition result is output.

4. According to claim 3, a smart parking management system based on the Internet of Things is characterized in that: The vehicle identification network comprises: The sampling layer is used to expand the plane dimension of the input parameters according to the bilinear interpolation, and transform the contour features of different levels and scales into a unified scale; The convolution layer is used to expand the channel of the output of the sampling layer and fuse the output of the sampling layer using a 1×1 convolution fusion channel; The detection layer is used to activate the output of the convolution layer, perform error back propagation on the output of the convolution layer after the activation processing according to the loss function, and output the recognition result after reaching the convergence condition, wherein the activation processing includes plane dimension activation mapping and stereo dimension activation mapping.

5. According to claim 1, the smart parking management system based on the Internet of Things is characterized in that: The parking recommendation module includes: Parking data analysis unit, used to analyze the parking preferences of car owners and predict parking time; An occupancy status matching unit, configured to output a parking space to be matched according to the output result of the parking data analysis unit and the occupancy status data; A parking space priority calculation unit, used to perform priority scoring on the parking spaces to be matched by using a weighted algorithm; A personalized recommendation unit, used to recommend the parking space with the highest priority as the best parking space to the car owner according to the parking space priority calculation result; The data analysis logic is configured in the parking data analysis unit, and is used to analyze the parking habits of the vehicle owner based on the vehicle owner's historical parking data, and predict the vehicle owner's parking time; The free parking space matching logic is configured in the occupancy status matching unit, and is used to filter out free parking spaces from the occupancy status data, and further match the free parking spaces to obtain parking spaces to be matched; The parking space recommendation logic is configured in the parking space priority calculation unit and is used to perform priority scoring on the parking spaces to be matched.

6. According to claim 1, a smart parking management system based on the Internet of Things is characterized in that: The dynamic pricing module includes: Real-time data collection unit, used to collect real-time data from various sensors and systems in the parking lot, including parking space occupancy, vehicle flow, parking duration and parking space area distribution information; The pricing unit is used to analyze the supply and demand of parking spaces in the parking lot and dynamically adjust the parking fees in different time periods and areas; A settlement unit, used to perform real-time settlement based on the output of the pricing unit and provide parking fee information when the car owner leaves the parking space; The intelligent pricing strategy includes a dynamic pricing logic, which is configured in the pricing unit and is used to dynamically adjust the parking fee according to the real-time situation of the parking lot.

7. According to claim 6, a smart parking management system based on the Internet of Things is characterized in that: The dynamic pricing logic includes: By collecting the parking space occupancy rate and vehicle flow of the parking lot, the current supply and demand relationship of the parking lot is analyzed, and the parking fee is adjusted according to the analysis results; adjusting parking fees according to the location of the parking space in the parking lot; Through the tiered pricing method, the parking fee is adjusted according to the parking time of the car owner.

8. According to claim 7, a smart parking management system based on the Internet of Things is characterized in that: The system also includes a monitoring module, which is used to monitor the situation in the parking lot and provide a remote monitoring function to the parking lot manager, specifically including: An anomaly detection unit, which is used to determine whether an abnormal situation occurs based on the real-time data collected from various sensors and systems in the parking lot; A notification unit for immediately sending an alert to the parking lot manager when an abnormal situation is detected; The remote monitoring unit is used to provide the functions of remotely monitoring parking lot conditions and viewing real-time monitoring video parking spaces through smart devices.

9. A smart parking management method based on the Internet of Things, the method is implemented based on the smart parking management system based on the Internet of Things as claimed in any one of claims 1 to 8, characterized in that: The specific steps are as follows: S1: Collect parking space images in the parking lot through image sensors to obtain real-time image data of each parking space; S2: Process the real-time image data, identify the target area image through neural network technology, and generate the occupancy status data of each parking space; S3: Based on the occupancy status data, combined with the owner's parking preferences and historical parking time, the best parking space is recommended to the owner through the parking space allocation strategy; S4: generating a navigation path for the vehicle owner to the recommended parking space and providing real-time guidance information; S5: adjusting parking fees based on real-time parking space occupancy, parking duration, and parking space location, and providing updated charging information to the car owner; S6: The parking lot is monitored in real time through the remote monitoring system. If any abnormal situation is detected, an alarm will be issued and the management staff will be notified.

Images