Intelligent parking space status recognition system and method based on video detection

Abstract

The present invention discloses an intelligent parking space status recognition system and method based on video detection, belonging to the technical field of parking space status recognition. The method specifically comprises: collecting video data of a parking area, and preprocessing the collected video data of the parking area, modeling the parking lot background based on a time series, and updating it in real time, detecting the foreground based on the background modeling result, extracting the foreground area, performing contour analysis on the detected foreground area, judging the occupancy status of each parking space, generating a real-time parking space distribution map of the parking lot according to the occupancy status of all parking spaces, and displaying the occupancy or idle status of each parking space. The intelligent parking space status recognition system based on video detection can maintain high accuracy in actual scenarios such as complex lighting conditions, complex environments, different types of vehicles and angles, and help drivers quickly find vacant parking spaces, thereby reducing traffic congestion caused by searching for parking spaces.

Description

Technical Field

[0001] The present invention belongs to the technical field of parking space status recognition, and in particular to an intelligent parking space status recognition system and method based on video detection. Background Art

[0002] With the acceleration of urbanization, the number of private cars has increased rapidly, and the problem of parking difficulties has become increasingly prominent, especially in commercial areas, residential areas and public places in cities. Traditional parking management methods rely on manual inspections or simple parking sensors to monitor the occupancy status of parking spaces. This method is not only inefficient, but may also result in errors or delays, resulting in a waste of parking resources and increased time costs for users. Intelligent parking space management systems have emerged as the times require. They can achieve intelligent management of parking spaces through technical means and improve the utilization rate of parking lots.

[0003] Video detection technology combines the advantages of modern computer vision and artificial intelligence. By analyzing the real-time monitoring video of the parking lot, it can automatically identify whether the parking space is occupied. Compared with traditional sensor solutions, video detection technology does not need to install independent sensors on each parking space, which reduces hardware costs, and can cover a wider range and identify the status of multiple parking spaces. However, it cannot maintain high accuracy in actual scenarios such as complex lighting conditions, complex environments, different types of vehicles and angles, and there is a certain error.

[0004] For example, a Chinese patent with the authorization announcement number CN109410635B discloses a method for identifying the status of roadside parking spaces, including: inputting or collecting roadside video images; obtaining occlusion parameters, parking space areas, areas of interest, and non-parking space areas, performing license plate recognition, and using the recognized license plate as a to-be-matched license plate; matching the to-be-matched license plate with the current parking space license plate, assigning the matched license plate to the corresponding parking space, matching the current non-parking space license plate with the parking space license plate, and the remaining unassigned license plate with the non-parking space license plate, respectively, and assigning the license plate according to the matching results; when the occlusion parameter is 0, performing the first or second counting process on the current parking space; when the occlusion parameter is 1, performing the first or second counting process on the current parking space; updating the current parking space status; and outputting the parking space status and license plate information of all parking spaces. Compared with the prior art, the invention can effectively solve the license plate occlusion problem in roadside parking space status recognition and identify the parking space status.

[0005] For example, a Chinese patent with the authorization announcement number CN111192478B discloses a method for identifying the status of road parking spaces based on millimeter wave radar. The method includes using a radar module to transmit and receive the returned probe wave, processing the returned probe wave to obtain a point trace with radial velocity vector Vi, distance ρi and angle θi information; calculating the target longitudinal distance YL according to the target distance ρmean and angle θmean, screening suspected parked vehicles according to the target longitudinal distance YL and the target radial velocity vector Vmean; calculating the point trace distance variance D of the suspected parked vehicle, judging whether the suspected parked vehicle is a motor vehicle according to the point trace distance variance D of the suspected parked vehicle; judging the parking space status according to the target radial velocity vector Vmean and the longitudinal distance YL. The invention can avoid interference from vehicles passing by the motor vehicle lane outside the parking frame, effectively distinguish between motor vehicles and non-motor vehicles, effectively judge the parking entry / stop / exit actions of the vehicle, and avoid passing vehicles that do not occupy the parking space.

[0006] The defects of the above patent are: it cannot maintain high accuracy in actual scenarios such as complex lighting conditions, complex environments, different types of vehicles and angles. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention proposes an intelligent parking space status recognition system and method based on video detection. The method can accurately and real-time identify the parking space status in a complex environment by introducing a multi-stage processing flow of dynamic background modeling, regional feature extraction and parking space occupancy status judgment, thereby avoiding the detection errors and instability in the prior art.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method for intelligent parking space status recognition based on video detection, comprising:

[0010] Collecting video data of the parking area, and preprocessing the collected video data of the parking area;

[0011] Model the parking lot background based on time series and update it in real time;

[0012] Based on the background modeling results, the foreground is detected and the foreground area is extracted;

[0013] Perform contour analysis on the detected foreground area to determine the occupancy status of each parking space;

[0014] Generate a real-time parking lot distribution map based on the occupancy status of all parking spaces, showing the occupied or idle status of each parking space.

[0015] Specifically, the preprocessing of the collected parking area video data includes: frame segmentation, grayscale conversion and denoising;

[0016] The frame division is used to divide the video data of the parking area into images of the parking area;

[0017] The grayscale is used to grayscale the framed parking area image data;

[0018] The denoising is used to denoise the grayscale parking area image.

[0019] Specifically, the modeling of the parking lot background based on the time series includes:

[0020] Model the background of each pixel (x, y). The specific formula is:

[0021] ,

[0022] Among them, P(x,y,t) represents the background model of the pixel point (x,y), that is, the pixel intensity distribution at the image coordinate (x,y) at time t. represents the weight of the i-th Gaussian distribution and satisfies , K represents the number of Gaussian distributions, represents the probability density function of the i-th Gaussian distribution, I(x,y,t) represents the pixel value at (x,y) in the parking area image after preprocessing at time t, represents the mean of the i-th Gaussian distribution at time t, represents the variance of the i-th Gaussian distribution at time t.

[0023] Specifically, the real-time updating includes:

[0024] At time t1, the newly arrived pixel value I(x, y, t1) is judged whether it can match the existing background model. The judgment formula is: , if the judgment formula is established, the pixel value I(x,y,t1) matches the existing background model, and the existing background model is updated;

[0025] Update the existing background model. The specific formula is:

[0026] ,

[0027] Among them, B(x,y,t1) represents the background model of the updated pixel point (x,y), that is, the pixel intensity distribution at the image coordinate (x,y) at time t1, and B represents the number of updated Gaussian distributions. represents the weight of the i-th Gaussian distribution after update and satisfies , represents the probability density function of the i-th Gaussian distribution after update, represents the mean of the i-th Gaussian distribution at time t1, Represents the variance of the i-th Gaussian distribution at time t1.

[0028] Specifically, the foreground detection and foreground area extraction include:

[0029] Based on the background modeling results, foreground detection is performed to determine whether each pixel belongs to the foreground. The judgment formula for foreground detection is:

[0030] ,

[0031] Among them, F(x,y,t) represents the judgment formula of foreground detection. represents the mean value at (x,y) in the background model, represents the standard deviation at (x,y) in the background model, Represents the parameters for adjusting the foreground and background discrimination thresholds, It represents the absolute value function. When F(x,y,t)=1, it means that the pixel point (x,y) at time t belongs to the foreground image. When F(x,y,t)=0, it means that the pixel point (x,y) at time t belongs to the background image.

[0032] The foreground image is expanded to connect the discontinuous areas caused by noise or occlusion. The specific formula of the expansion operation is:

[0033] ,

[0034] in, represents the foreground image after the dilation operation, S represents the dilation kernel, and p and q represent the dilation points;

[0035] Perform connectivity analysis on the foreground area, divide the foreground image after the dilation operation into several connected sub-areas, and set R s represents the sth connected region, and the area of ​​the connected region is A(R s );

[0036] Set the area threshold A max and A min , A max Indicates the maximum value of the area threshold, A min It represents the minimum value of the area threshold, which filters out small areas that do not meet the vehicle size and excessive noise areas, and obtains connected areas that meet the parking space size.

[0037] Specifically, the contour analysis of the detected foreground area to determine the occupancy status of each parking space includes:

[0038] In the video image, the actual parking space area Y is marked manually or automatically. aMarked as a polygon, and save the coordinate information in the video image, the actual parking area Y a It is expressed as:

[0039] ,

[0040] Among them, (x1,y1),(x2,y2),...,(x n ,y n ) represents the actual parking area Y a vertices, n represents the number of vertices;

[0041] Determine the occupancy status of the parking space and calculate the actual parking area Y a Foreground image after dilation The overlap rate is calculated as:

[0042] ,

[0043] Among them, O a Indicates the actual parking area Y a Foreground image after dilation The overlap rate, Area() represents the area function, Represents the intersection logical operator, Area(Y a ) represents the actual parking area Y a The area of represents the foreground area of ​​the a-th parking space;

[0044] Set the time window to T frames and calculate the cumulative overlap rate within the time window. The specific formula is:

[0045] ,

[0046] in, represents the cumulative overlap rate within the time window, w0 represents the starting time of the time window, represents the overlap rate of parking space a at the wth frame;

[0047] Set overlap threshold ,when ≥ When , it is determined that parking space a is occupied by a vehicle. < When , it is determined that parking space a is not occupied by a vehicle.

[0048] An intelligent parking space status recognition system based on video detection is used to implement an intelligent parking space status recognition method based on video detection, including: a data acquisition module, a background model building module, a foreground detection module, an occupancy judgment module and a real-time display module;

[0049] The data acquisition module is used to collect video data of the parking area and pre-process the collected video data of the parking area;

[0050] The background model building module is used to model the parking lot background based on the time series and update it in real time;

[0051] The foreground detection module is used to detect the foreground and extract the foreground area based on the background modeling result;

[0052] The occupancy determination module is used to perform contour analysis on the detected foreground area to determine the occupancy status of each parking space;

[0053] The real-time display module is used to generate a real-time parking space distribution map according to the occupancy status of all parking spaces, and display the occupancy or idle status of each parking space.

[0054] Specifically, the background model building module includes: a background modeling unit and a real-time updating unit;

[0055] The background modeling unit is used to model each pixel point of the parking lot background based on the time series;

[0056] The real-time updating unit is used to update the parking lot background model in real time when there are new pixel values.

[0057] Specifically, the occupancy judgment module includes: a parking area marking unit, an overlap rate calculation unit, and a parking state judgment unit;

[0058] The parking space area marking unit is used to mark the actual parking space area in the parking lot video;

[0059] The overlap rate calculation unit is used to calculate the overlap rate between the actual parking space area and the foreground image after the expansion operation, and calculate the cumulative overlap rate within the time window;

[0060] The parking space state determination unit is used to determine whether the parking space is in an occupied state or an unoccupied state.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The present invention proposes an intelligent parking space status recognition method based on video detection. The video-based recognition system can adapt to different types of parking lots, including indoor and outdoor environments, and can recognize many different types of vehicles, such as sedans, SUVs, motorcycles, etc. In addition, it can maintain a high accuracy in actual scenarios such as complex lighting conditions, complex environments, different types of vehicles and angles.

[0063] 2. The present invention proposes an intelligent parking space status recognition method based on video detection, which can help drivers quickly find vacant parking spaces and reduce traffic congestion caused by searching for parking spaces.

[0064] 3. The present invention proposes an intelligent parking space status recognition method based on video detection. Through video detection, the parking management system can monitor the parking space status in real time and record parking data. With the help of big data analysis tools, managers can understand the parking lot utilization rate, parking space turnover rate, popular parking periods and other information, providing data support for optimizing parking lot planning and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flow chart of a method for intelligent parking space status recognition based on video detection provided by the present invention;

[0066] Figure 2 This is an architecture diagram of an intelligent parking space status recognition system based on video detection provided by the present invention. DETAILED DESCRIPTION

[0067] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, in the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like indicate the orientation or position relationship based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply 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 a limitation on the present invention. In addition, the terms "No. 1", "No. 2" and "No. 3" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The present invention is further explained below in conjunction with specific embodiments.

[0068] Example 1

[0069] See also Figure 1 The present invention provides an embodiment: a method for intelligent parking space status recognition based on video detection, comprising the following specific steps:

[0070] Step S1: collecting video data of the parking area through a camera installed above the parking area, and preprocessing the collected video data of the parking area;

[0071] The preprocessing in step S1 includes: framing, graying and denoising;

[0072] The frame division is used to divide the video data of the parking area into images of the parking area;

[0073] The grayscale conversion is used to grayscale the image data after frame division. The specific formula is:

[0074] I(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y), where I(x,y) represents the grayscale image of the parking area, that is, the grayscale value of the parking area grayscale image at the pixel point (x,y), R(x,y) represents the red channel value of the parking area image at the pixel point (x,y), G(x,y) represents the green channel value of the parking area image at the pixel point (x,y), and B(x,y) represents the blue channel value of the parking area image at the pixel point (x,y);

[0075] The denoising is used to denoise the grayscale parking area image, and Gaussian blur filtering is used to perform denoising to reduce noise interference.

[0076] Step S2: Modeling the parking lot background based on the time series and updating it in real time;

[0077] The specific steps of step S2 are:

[0078] Step S201: Model the background of each pixel (x, y). The specific formula is:

[0079] ,

[0080] Among them, P(x,y,t) represents the background model of the pixel point (x,y), that is, the pixel intensity distribution at the image coordinate (x,y) at time t. represents the weight of the i-th Gaussian distribution, that is, the contribution of the distribution to the pixel value, and satisfies , K represents the number of Gaussian distributions, Represents the probability density function of the i-th Gaussian distribution, describing the intensity I(x,y,t) of the pixel (x,y) at time t, which obeys the mean , the variance is Normal distribution, I(x,y,t) represents the pixel value at (x,y) in the parking area image after preprocessing at time t, represents the mean of the i-th Gaussian distribution at time t, represents the variance of the i-th Gaussian distribution at time t;

[0081] The pixel value here is the grayscale value of the image;

[0082] In this embodiment, background modeling is usually used to distinguish the foreground (such as a moving vehicle or pedestrian) from the background (a stationary scene) in a video frame. The mixed Gaussian model assumes that the intensity of each pixel is a mixture of multiple factors, and these factors (such as lighting, shadows, reflections, etc.) are modeled using multiple Gaussian distributions. Each Gaussian distribution describes a possible variation of the pixel intensity, and the weights It indicates the influence of the distribution at the current moment;

[0083] Step S202: The pixel value I(x, y, t1) newly arrived at time t1 is judged whether it can match the existing background model. The judgment formula is: , if the judgment formula is established, the pixel value I(x,y,t1) matches the existing background model, and the existing background model is updated;

[0084] In this embodiment, in each frame of the video, the difference between the current intensity I(x,y,t) of the pixel point (x,y) and each Gaussian distribution in the background model is calculated to determine whether the pixel belongs to the foreground. The difference is not big (such as in 2.5 If the pixel is within the range of , the pixel is considered to belong to the background; otherwise, it is considered to belong to the foreground;

[0085] Step S203: Update the existing background model. The specific formula is:

[0086] ,

[0087] Among them, B(x,y,t1) represents the background model of the updated pixel point (x,y), that is, the pixel intensity distribution at the image coordinate (x,y) at time t1, and B represents the number of updated Gaussian distributions. represents the weight of the i-th Gaussian distribution after update, that is, the contribution of the distribution to the pixel value, and satisfies , usually T=0.7, represents the probability density function of the i-th Gaussian distribution after update, represents the mean of the i-th Gaussian distribution at time t1, Represents the variance of the i-th Gaussian distribution at time t1.

[0088] In this embodiment, the mean of the i-th Gaussian distribution at time t1 is That is the updated mean, the specific formula is: , Represents the first learning rate, which determines the influence of the new pixel value on the distribution mean. Usually it is 0.01~0.05, the variance of the i-th Gaussian distribution at time t1 That is, the updated variance, the specific formula is: , the weight of the i-th Gaussian distribution after update The specific formula is: , Represents the second learning rate.

[0089] Step S3: Based on the background modeling result, the foreground is detected and the foreground area is extracted;

[0090] The specific steps of step S3 are:

[0091] Step S301: Based on the background modeling result, foreground detection is performed to determine whether each pixel belongs to the foreground. The determination formula for foreground detection is:

[0092] ,

[0093] Among them, F(x,y,t) represents the judgment formula of foreground detection, represents the mean value at (x,y) in the background model, represents the standard deviation at (x,y) in the background model, Represents the parameters for adjusting the foreground and background discrimination thresholds, It represents the absolute value function. When F(x,y,t)=1, it means that the pixel point (x,y) at time t belongs to the foreground image. When F(x,y,t)=0, it means that the pixel point (x,y) at time t belongs to the background image.

[0094] In this embodiment, Indicates the degree of deviation between the intensity value of the current pixel and the background mean. If the deviation exceeds , it means that the brightness of the current pixel changes greatly, which may be caused by moving objects (such as vehicles and pedestrians). In this case, the pixel is marked as foreground. If the deviation is less than or equal to , then the pixel is considered to still belong to the background, and its brightness change may be caused by noise or slight background changes (such as illumination changes). The size of determines the sensitivity of foreground judgment. A value of 0 will cause smaller brightness changes to be recognized as foreground, but may increase noise; larger values ​​of A large value increases the threshold for foreground recognition and reduces the impact of noise, but may miss smaller foreground targets (such as shadows and reflective objects);

[0095] The core idea of ​​this formula is to use the mean and standard deviation of the brightness of each pixel in the background model to distinguish the foreground from the background. When the brightness of a pixel changes beyond a certain range, the pixel is considered to be part of the foreground object. The background model continuously updates the mean and standard deviation to adapt to the ambient lighting, subtle changes in the background, etc., to improve the robustness of the overall system.

[0096] Step S302: Perform a dilation operation on the foreground image to connect discontinuous areas caused by noise or occlusion. The specific formula of the dilation operation is:

[0097] ,

[0098] in, represents the foreground image after the dilation operation, S represents the dilation kernel, such as a 3×3 matrix, p and q represent the dilation points, that is, the corresponding values ​​in the dilation kernel matrix. The dilation operation can connect the foreground area split by noise;

[0099] Step S303: Perform connectivity analysis on the foreground area, divide the foreground image after the dilation operation into several connected sub-areas, and set R s represents the sth connected region, and the area of ​​the connected region is A(R s );

[0100] Step S304: Setting the area threshold A max and A min , A max Indicates the maximum value of the area threshold, A min It represents the minimum value of the area threshold, which filters out small areas that do not meet the vehicle size and excessive noise areas, and obtains connected areas that meet the parking space size.

[0101] Step S4: performing contour analysis on the detected foreground area to determine the occupancy status of each parking space;

[0102] The specific steps of step S4 are:

[0103] Step S401: In the video image, the actual parking space area Y is marked manually or automatically. a Marked as a polygon, and save the coordinate information in the video image, the actual parking area Y a It is expressed as:

[0104] ,

[0105] Among them, (x1,y1),(x2,y2),...,(x n ,y n ) represents the actual parking area Y a vertices, n represents the number of vertices;

[0106] Step S402: Determine the occupancy status of the parking space and calculate the actual parking space area Y a Foreground image after dilation The overlap rate is calculated as:

[0107] ,

[0108] Among them, O a Indicates the actual parking area Y a Foreground image after dilation The overlap rate, Area() represents the area function, Represents the intersection logical operator, Area(Y a ) represents the actual parking area Y a The area of represents the foreground area of ​​the a-th parking space;

[0109] In this embodiment, an intersection operation can be performed on the foreground area in the current frame and the known parking space area through an image processing method (such as background subtraction or foreground detection). This intersection represents the portion of the parking space occupied by the foreground object (such as a car). The occupancy of the parking space is quantified by calculating the ratio of the area of ​​the intersection to the total area. The overlap rate value is between 0 and 1, where 0 indicates that the parking space is completely vacant and 1 indicates that the parking space is completely occupied. It can be intuitively concluded whether the parking space is occupied. This method is simple and effective, easy to implement, and has good robustness to environmental noise (such as changes in illumination, movement of small objects, etc.). By setting an overlap rate threshold, the system can distinguish different states of the parking space being completely occupied, partially occupied, or completely vacant. For example, when the overlap rate is less than a certain threshold, the system can determine that the parking space is empty.

[0110] Whether in the natural lighting conditions of outdoor parking lots or in indoor scenes with large changes in lighting, it can flexibly adapt through background modeling, foreground segmentation and other technologies. Even if the vehicle moves slowly or stops temporarily, the dynamic changes in the parking space status can still be judged by the overlap rate of consecutive frames.

[0111] Step S403: Set the time window to T frames and calculate the cumulative overlap rate within the time window. The specific formula is:

[0112] ,

[0113] in, represents the cumulative overlap rate within the time window, w0 represents the starting time of the time window, represents the overlap rate of parking space a at the wth frame;

[0114] In this embodiment, taking into account the short-term interference in the parking lot (such as transient light changes, pedestrians passing by, etc.), a state accumulation method based on a time window is used. By cumulatively analyzing the parking space occupancy within a certain period of time, the cumulative overlap rate formula is the sum of the overlap rates of all frames between w0 and w0+T divided by the total number of frames T. This average value can more robustly reflect the state of the parking space over a period of time, rather than relying on a certain instantaneous occupancy situation;

[0115] The parking space occupancy of a single frame may be affected by illumination, shadow, noise, short-term vehicle movement, etc. By calculating the cumulative overlap rate within the time window, such short-term changes can be effectively smoothed out, thereby improving the reliability of recognition. The status of a parking space changes dynamically and may change frequently in a short period of time (such as a vehicle temporarily staying or leaving). This formula calculates the average value through data from multiple frames to better capture the overall trend of these changes without causing misjudgment due to a single instantaneous change. If the time window is set to a longer time, analysis can be performed based on longer statistical data, which is suitable for scenarios where the status of a parking space needs to be judged over a longer period of time. For example, if a parking space maintains a high overlap rate for a long time, it can be inferred that it is a permanent vehicle or unused.

[0116] Step S404: Setting the overlap rate threshold ,when ≥ When , it is determined that parking space a is occupied by a vehicle. < When , it is determined that parking space a is not occupied by a vehicle.

[0117] In this embodiment, in order to cope with dynamic changes in the parking environment (such as gradual changes in light, camera jitter, etc.), an adaptive threshold adjustment mechanism can be introduced to dynamically update the judgment standard of the overlap rate. By introducing the fusion of historical data and current frame information, the adaptive mechanism can automatically adjust the threshold according to environmental changes. , thereby improving the accuracy of parking space occupancy judgment.

[0118] Step S5: Generate a real-time parking space distribution map according to the occupancy status of all parking spaces, display the occupied or idle status of each parking space, and transmit the result to the parking management system or the user terminal.

[0119] Example 2

[0120] See also Figure 2 ,Another embodiment provided by the present invention: A smart parking space status recognition system based on video detection, comprising: a data acquisition module, a background model building module, a foreground detection module, an occupancy judgment module and a real-time display module;

[0121] The data acquisition module is used to collect video data of the parking area and pre-process the collected video data of the parking area;

[0122] The background model building module is used to model the parking lot background based on the time series and update it in real time;

[0123] The foreground detection module is used to detect the foreground and extract the foreground area based on the background modeling result;

[0124] The occupancy determination module is used to perform contour analysis on the detected foreground area to determine the occupancy status of each parking space;

[0125] The real-time display module is used to generate a real-time parking space distribution map according to the occupancy status of all parking spaces, and display the occupancy or idle status of each parking space.

[0126] The background model building module includes: a background modeling unit and a real-time updating unit;

[0127] The background modeling unit is used to model each pixel point of the parking lot background based on the time series;

[0128] The real-time updating unit is used to update the parking lot background model in real time when there are new pixel values.

[0129] The occupancy judgment module includes: a parking area marking unit, an overlap rate calculation unit, and a parking state judgment unit;

[0130] The parking space area marking unit is used to mark the actual parking space area in the parking lot video;

[0131] The overlap rate calculation unit is used to calculate the overlap rate between the actual parking space area and the foreground image after the expansion operation, and calculate the cumulative overlap rate within the time window;

[0132] The parking space state determination unit is used to determine whether the parking space is in an occupied state or an unoccupied state.

[0133] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0134] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for intelligent parking space status recognition based on video detection, characterized in that: include: Collecting video data of the parking area, and preprocessing the collected video data of the parking area; Model the parking lot background based on time series and update it in real time; Based on the background modeling results, the foreground is detected and the foreground area is extracted; Perform contour analysis on the detected foreground area to determine the occupancy status of each parking space; Generate a real-time parking space distribution map based on the occupancy status of all parking spaces, showing the occupied or idle status of each parking space; The step of performing contour analysis on the detected foreground area to determine the occupancy status of each parking space includes: In the video image, the actual parking space area Y is marked manually or automatically. a Marked as a polygon, and save the coordinate information in the video image, the actual parking area Y a It is expressed as: ， Among them, (x1,y1),(x2,y2),...,(x n ,y n ) represents the actual parking area Y a vertices, n represents the number of vertices; Determine the occupancy status of the parking space and calculate the actual parking area Y a Foreground image after dilation The overlap rate is calculated as: ， Among them, O a Indicates the actual parking area Y a Foreground image after dilation The overlap rate, Area() represents the area function, Represents the intersection logical operator, Area(Y a ) represents the actual parking area Y a The area of represents the foreground area of ​​the a-th parking space; Set the time window to T frames and calculate the cumulative overlap rate within the time window. The specific formula is: ， in, represents the cumulative overlap rate within the time window, w0 represents the starting time of the time window, represents the overlap rate of parking space a at the wth frame; Set overlap threshold ,when ≥ When , it is determined that parking space a is occupied by a vehicle. < When , it is determined that parking space a is not occupied by a vehicle.

2. The method for intelligent parking space status recognition based on video detection according to claim 1, characterized in that: The preprocessing of the collected parking area video data includes: frame segmentation, grayscale conversion and denoising; The frame division is used to divide the video data of the parking area into images of the parking area; The grayscale is used to grayscale the framed parking area image data; The denoising is used to denoise the grayscale parking area image.

3. The method for intelligent parking space status recognition based on video detection as claimed in claim 2, characterized in that: The method of modeling the parking lot background based on the time series includes: Model the background of each pixel (x, y). The specific formula is: ， Among them, P(x,y,t) represents the background model of the pixel point (x,y), that is, the pixel intensity distribution at the image coordinate (x,y) at time t. represents the weight of the i-th Gaussian distribution and satisfies , K represents the number of Gaussian distributions, represents the probability density function of the i-th Gaussian distribution, I(x,y,t) represents the pixel value at (x,y) in the parking area image after preprocessing at time t, represents the mean of the i-th Gaussian distribution at time t, represents the variance of the i-th Gaussian distribution at time t.

4. The method for intelligent parking space status recognition based on video detection as claimed in claim 3, characterized in that: The real-time update includes: At time t1, the newly arrived pixel value I(x, y, t1) is judged whether it can match the existing background model. The judgment formula is: , if the judgment formula is established, the pixel value I(x,y,t1) matches the existing background model, and the existing background model is updated; Update the existing background model. The specific formula is: ， Among them, B(x,y,t1) represents the background model of the updated pixel point (x,y), that is, the pixel intensity distribution at the image coordinate (x,y) at time t1, and B represents the number of updated Gaussian distributions. represents the weight of the i-th Gaussian distribution after update and satisfies , represents the probability density function of the i-th Gaussian distribution after update, represents the mean of the i-th Gaussian distribution at time t1, Represents the variance of the i-th Gaussian distribution at time t1.

5. The method for intelligent parking space status recognition based on video detection as claimed in claim 4, characterized in that: The foreground detection and foreground area extraction include: Based on the background modeling results, foreground detection is performed to determine whether each pixel belongs to the foreground. The judgment formula for foreground detection is: ， Among them, F(x,y,t) represents the judgment formula of foreground detection, represents the mean value at (x,y) in the background model, represents the standard deviation at (x,y) in the background model, Represents the parameters for adjusting the foreground and background discrimination thresholds, It represents the absolute value function. When F(x,y,t)=1, it means that the pixel point (x,y) at time t belongs to the foreground image. When F(x,y,t)=0, it means that the pixel point (x,y) at time t belongs to the background image. The foreground image is expanded to connect the discontinuous areas caused by noise or occlusion. The specific formula of the expansion operation is: ， in, represents the foreground image after the dilation operation, S represents the dilation kernel, and p and q represent the dilation points; Perform connectivity analysis on the foreground area, divide the foreground image after the dilation operation into several connected sub-areas, and set R s represents the sth connected region, and the area of ​​the connected region is A(R s ); Set the area threshold A max and A min , A max Indicates the maximum value of the area threshold, A min It represents the minimum value of the area threshold, which filters out small areas that do not meet the vehicle size and excessive noise areas, and obtains connected areas that meet the parking space size.

6. An intelligent parking space status recognition system based on video detection, used to implement an intelligent parking space status recognition method based on video detection as claimed in any one of claims 1 to 5, characterized in that: include: Data acquisition module, background model building module, foreground detection module, occupancy judgment module and real-time display module; The data acquisition module is used to collect video data of the parking area and pre-process the collected video data of the parking area; The background model building module is used to model the parking lot background based on the time series and update it in real time; The foreground detection module is used to detect the foreground and extract the foreground area based on the background modeling result; The occupancy determination module is used to perform contour analysis on the detected foreground area to determine the occupancy status of each parking space; The real-time display module is used to generate a real-time parking space distribution map according to the occupancy status of all parking spaces, and display the occupancy or idle status of each parking space.

7. The intelligent parking space status recognition system based on video detection as claimed in claim 6, characterized in that: The background model building module includes: a background modeling unit and a real-time updating unit; The background modeling unit is used to model each pixel point of the parking lot background based on the time series; The real-time updating unit is used to update the parking lot background model in real time when there are new pixel values.

8. The intelligent parking space status recognition system based on video detection as claimed in claim 7, characterized in that: The occupancy judgment module includes: a parking area marking unit, an overlap rate calculation unit, and a parking state judgment unit; The parking space area marking unit is used to mark the actual parking space area in the parking lot video; The overlap rate calculation unit is used to calculate the overlap rate between the actual parking space area and the foreground image after the expansion operation, and calculate the cumulative overlap rate within the time window; The parking space state determination unit is used to determine whether the parking space is in an occupied state or an unoccupied state.

Images