An intelligent vehicle detection method for underground garage property management

Abstract

The invention discloses an intelligent detection method for underground garage property management vehicles, belonging to the technical field of image processing. The invention firstly analyzes images in underground garage videos, calculates change feature values, and obtains vehicle driving images, change positions, and static images; then, removes the change positions and performs brightness adjustment; then, by comparing the difference between the static image and the vehicle driving image, locates the vehicle and extracts the image of the vehicle to be detected; subsequently, clustering is performed on the image of the vehicle to be detected, the main color area of ​​the vehicle body is identified, and the color similarity with the stored vehicle image is calculated; then, the class in contact with the main color area of ​​the vehicle body is extracted, a color comparison vector is constructed, and the similarity is further analyzed; finally, by comparing the morphological values ​​of the candidate vehicle image and the image of the vehicle to be detected, accurate identification of the vehicle to be detected is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an intelligent detection method for vehicles in underground garage property management. Background Art

[0002] With the acceleration of urbanization and the continuous growth of the number of motor vehicles, the management and security monitoring of underground garages have become an important issue in the management of modern urban infrastructure. The underground garage passages are intricate and the cameras are usually installed at the top. This layout poses severe challenges to traditional vehicle recognition methods. Due to the limited shooting angle, it is difficult to directly obtain a clear image of the front of the vehicle or the license plate, making the precise positioning of the vehicle position a difficult technical problem. Traditional vehicle management methods mainly rely on manual inspections and static monitoring, which not only have problems such as high labor costs, low monitoring efficiency, and high misjudgment rate, but also cannot achieve accurate vehicle tracking and identification due to spatial structure and viewing angle limitations. Most of the existing vehicle recognition technologies are limited to license plate recognition or simple vehicle counting under ideal lighting conditions, which is difficult to cope with the complex environmental characteristics of underground garages, such as uneven light, shadows, dense vehicles and other actual scenes. Especially in underground spaces with dim light and interlaced vehicles, it is impossible to effectively achieve accurate vehicle identification. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides an intelligent detection method for underground garage property management vehicles, which solves the problem that the prior art cannot accurately identify vehicles in underground garages.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: an intelligent detection method for underground garage property management vehicles, comprising the following steps:

[0005] S1. Calculate the change feature value according to the image in the video of the underground garage, and obtain the vehicle driving image, change position and static image;

[0006] S2, removing the changed position from the vehicle driving image, calculating the ambient brightness value, and performing brightness adjustment in the HSI space to obtain a brightness adjusted image;

[0007] S3, according to the difference between the static image and the vehicle driving image, find out the location of the vehicle, cut out the vehicle area from the brightness adjusted image, and obtain the image of the vehicle to be detected;

[0008] S4, clustering the image of the vehicle to be detected, finding the main color area of ​​the vehicle body, calculating the color similarity between the image of the vehicle to be detected and the stored vehicle image in the main color area of ​​the vehicle body, and screening the image of the vehicle to be detected;

[0009] S5, extracting other classes that are in contact with the main color area of ​​the vehicle body as contact classes, calculating the color contrast between the contact class and the main color area of ​​the vehicle body, constructing a color contrast vector, calculating the similarity between the image of the pending vehicle and the image of the vehicle to be detected on the color contrast vector, and finding candidate vehicle images and suspected similar area pairs;

[0010] S6. Compare the class morphology values ​​of the candidate vehicle image and the vehicle image to be detected on the suspected similar region pair to identify the vehicle image to be detected.

[0011] Furthermore, S1 comprises the following sub-steps:

[0012] S11. In the video of the underground garage, subtract the pixel values ​​of the image at time t+1 from the image at time t at the same pixel position, and take the absolute value of the subtraction result to obtain the picture change image at time t+1;

[0013] S12, adding all pixel values ​​in the picture change image to obtain the change pixel;

[0014] S13, counting the number of pixel values ​​greater than 0 in the picture change image, and calculating the change feature value based on the change pixels;

[0015] S14, when the change characteristic value is greater than the upper limit change characteristic threshold, there is a vehicle in the video, and the image corresponding to time t+1 is a vehicle driving image;

[0016] S15, when the change characteristic value is less than the lower limit change characteristic threshold, the picture in the video is still, and the image corresponding to time t+1 is a static image;

[0017] S16. When the image at time t+1 is a vehicle driving image, mark the pixel points with pixel values ​​greater than 0 in the picture change image at time t+1 to obtain the change position.

[0018] Furthermore, the formula for calculating the variation characteristic value in S13 is: , where ξ is the change feature value, N is the number of pixel values ​​greater than 0 in the picture change image, R is the change pixel, and Z is the number of pixel points in the picture change image.

[0019] Further, S2 includes the following sub-steps:

[0020] S21, removing the changed position from the vehicle driving image to obtain the unchanged area;

[0021] S22, converting the unchanged area to the HSI space, extracting the brightness I component of each pixel, taking the average of the brightness I components of each pixel, and obtaining the ambient brightness value;

[0022] S23. Calculate the brightness adjustment ratio according to the ambient brightness value: γ=Lcontrast / L en , where γ is the brightness adjustment ratio, L contrast is the contrast brightness value, L en is the ambient brightness value;

[0023] S24, converting the vehicle driving image into the HSI space, and adjusting the brightness of the brightness I component of each pixel according to the brightness adjustment ratio: ,in, is the brightness I component of the i-th pixel after adjustment, L i is the brightness I component of the i-th pixel before adjustment, where i is a positive integer;

[0024] S25, restoring the brightness-adjusted image to the RGB space to obtain a brightness-adjusted image.

[0025] Furthermore, S3 includes the following sub-steps:

[0026] S31, subtracting pixel values ​​of the vehicle driving image and the static image at the same pixel position to obtain a pixel difference value of each pixel point, and taking an absolute value of the pixel difference value to obtain a pixel difference;

[0027] S32, when the pixel difference is greater than 0, mark the pixel as a suspicious point;

[0028] S33, when there are no other suspicious points within the neighborhood of a suspicious point, discard the suspicious point;

[0029] S34, dividing the remaining suspicious points into a plurality of simply connected regions, wherein the simply connected regions are not in contact with each other;

[0030] S35. Calculate the area difference between each simply connected region and the vehicle imaging area: , where d E is the area difference, E s is the area of ​​the simply connected region, E v is the vehicle imaging area;

[0031] S36: when the area difference is less than the first difference threshold, the simply connected area is a vehicle area;

[0032] S37, extracting the vehicle region from the brightness-adjusted image to obtain a vehicle image to be detected.

[0033] Further, S4 includes the following sub-steps:

[0034] S41, clustering the pixels according to the pixel values ​​on the image of the vehicle to be detected to obtain a classified image;

[0035] S42, finding the largest class in the classified image, and using this class as the main color area of ​​the vehicle body;

[0036] S43, extracting the mean of the R, G, and B channel values ​​of the main color area of ​​the vehicle body to form a color vector;

[0037] S44, calculating the similarity between the color vector in S43 and the stored color vector storing the vehicle image, to obtain a first similarity;

[0038] S45. When the first similarity is greater than a first similarity threshold, the stored vehicle image corresponding to the stored color vector is a pending vehicle image.

[0039] Further, S5 includes the following sub-steps:

[0040] S51, on the image of the vehicle to be detected, extracting other classes that are in contact with the main color area of ​​the vehicle body as contact classes, calculating the color contrast between the contact class and the main color area of ​​the vehicle body, and constructing a color contrast vector;

[0041] S52, calculating the similarity between the color contrast vector of each contact class and the color contrast vector of the contact class in the to-be-determined vehicle image to obtain a second similarity;

[0042] S53: When the second similarity is greater than a second similarity threshold, mark the pending vehicle image as a candidate vehicle image, and mark the pair of contact classes as a suspected similar region pair.

[0043] Further, S51 includes the following sub-steps:

[0044] S511, on the image of the vehicle to be detected, other classes that are in contact with the main color area of ​​the vehicle body are regarded as contact classes;

[0045] S512, extracting the mean values ​​of R, G, and B channels in the contact category;

[0046] S513, taking the ratio of the R channel mean value of the main color area of ​​the vehicle body to the R channel mean value of the contact type as the R channel contrast;

[0047] S514, taking the ratio of the G channel mean value of the main color area of ​​the vehicle body to the G channel mean value of the contact type as the G channel contrast;

[0048] S515, taking the ratio of the B channel mean value of the main color area of ​​the vehicle body to the B channel mean value of the contact type as the B channel contrast;

[0049] S516: Use the R channel contrast, the G channel contrast, and the B channel contrast as elements to construct a color contrast vector.

[0050] Further, S6 includes the following sub-steps:

[0051] S61, calculating the class morphology value of each contact class in the suspected similar region pair;

[0052] S62, calculating the morphological gap between the class morphological values ​​of the two contact classes in the suspected similar region pair, wherein the morphological gap is the square of the difference between the class morphological values ​​of the two contact classes;

[0053] S63, when the morphological difference is less than a second difference threshold, marking the contact class belonging to the candidate vehicle image in the suspected similar region pair as a confirmed similar region;

[0054] S64: When it is confirmed that there are multiple similar regions in a candidate vehicle image, the candidate vehicle image and the vehicle image to be detected are the same vehicle, and the recognition of the vehicle image to be detected is completed.

[0055] Furthermore, the formula for calculating the class morphology value in S61 is: , where θ is the class morphology value, P is the perimeter of the contact class, A is the area of ​​the contact class, and r j is the curvature of the jth edge pixel in the contact class, M is the number of edge pixels in the contact class, and j is a positive integer.

[0056] The beneficial effects of the present invention are:

[0057] 1. The present invention calculates the change feature value based on the image at each moment in the video, reflects the image change at different moments, extracts the image containing the moving vehicle, that is, the vehicle driving image, obtains the change position and the static image, and then adjusts the brightness of the vehicle driving image to reduce the impact of the darker light in the underground garage on the image, and further eliminates the background area to obtain the vehicle image to be detected that only contains the vehicle, reducing the interference of background and light.

[0058] 2. The present invention performs clustering processing on the image of the vehicle to be detected, finds out the main color area of ​​the vehicle body, performs color comparison on the main color area of ​​the vehicle body between the image of the vehicle to be detected and the stored vehicle image, screens out the stored vehicle images with similar colors in the main color area of ​​the vehicle body, i.e., the image of the vehicle to be determined, and then constructs a color comparison vector based on the class in contact with the main color area of ​​the vehicle body, finds out candidate vehicle images and suspected similar area pairs from multiple images of the vehicle to be determined, determines the similarity of the two images from the class in contact with the main color area of ​​the vehicle body, realizes comparison of local areas, further screens the images, and finally checks the class morphology values ​​of the two classes in the suspected similar area pair, judges whether the shape features of the classes are similar, and determines the identity of the vehicle traveling in the underground garage.

[0059] 3. The present invention adjusts the brightness of the vehicle driving image, so that when performing color comparison, the accuracy of color comparison can be improved, and then the background area is eliminated to obtain the vehicle image to be detected that only contains the vehicle, reducing the interference of the background, and facilitating the subsequent color comparison of the main color area and local area of ​​the vehicle body. After the color comparison, the vehicle is further confirmed through morphological comparison, thereby realizing accurate identification of the vehicle in the underground garage, which is convenient for the property to manage and track the vehicle driving process. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 The present invention is a flow chart of an intelligent vehicle detection method for underground garage property management. DETAILED DESCRIPTION

[0061] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0062] like Figure 1 As shown, a method for intelligent detection of vehicles for property management in underground garages comprises the following steps:

[0063] S1. Calculate the change feature value according to the image in the video of the underground garage, and obtain the vehicle driving image, change position and static image;

[0064] S2, removing the changed position from the vehicle driving image, calculating the ambient brightness value, and performing brightness adjustment in the HSI space to obtain a brightness adjusted image;

[0065] S3, according to the difference between the static image and the vehicle driving image, find out the location of the vehicle, cut out the vehicle area from the brightness adjusted image, and obtain the image of the vehicle to be detected;

[0066] S4, clustering the image of the vehicle to be detected, finding the main color area of ​​the vehicle body, calculating the color similarity between the image of the vehicle to be detected and the stored vehicle image in the main color area of ​​the vehicle body, and screening the image of the vehicle to be detected;

[0067] S5, extracting other classes that are in contact with the main color area of ​​the vehicle body as contact classes, calculating the color contrast between the contact class and the main color area of ​​the vehicle body, constructing a color contrast vector, calculating the similarity between the image of the pending vehicle and the image of the vehicle to be detected on the color contrast vector, and finding candidate vehicle images and suspected similar area pairs;

[0068] S6. Compare the class morphology values ​​of the candidate vehicle image and the vehicle image to be detected on the suspected similar region pair to identify the vehicle image to be detected.

[0069] In this embodiment, S1 includes the following sub-steps:

[0070] S11. In the video of the underground garage, subtract the pixel values ​​of the image at time t+1 from the image at time t at the same pixel position, and take the absolute value of the subtraction result to obtain the picture change image at time t+1;

[0071] S12, adding all pixel values ​​in the picture change image to obtain the change pixel;

[0072] S13, counting the number of pixel values ​​greater than 0 in the picture change image, and calculating the change feature value based on the change pixels;

[0073] S14, when the change characteristic value is greater than the upper limit change characteristic threshold, there is a vehicle in the video, and the image corresponding to time t+1 is a vehicle driving image;

[0074] S15, when the change characteristic value is less than the lower limit change characteristic threshold, the picture in the video is still, and the image corresponding to time t+1 is a static image;

[0075] S16. When the image at time t+1 is a vehicle driving image, mark the pixel points with pixel values ​​greater than 0 in the screen change image at time t+1 to obtain the change position.

[0076] There are images at multiple moments in the video. Therefore, when the picture in the video is still, the image at that moment is extracted as a static image. When there is a vehicle in the video, the image at the corresponding moment is extracted as a vehicle driving image, and then the change position is extracted based on the picture change image at the same moment.

[0077] In this embodiment, since the image storage format in the video is RGB, the pixel value of each pixel in the sub-step S1 is equal to the average value of the three channels of the pixel.

[0078] In this embodiment, the formula for calculating the variation characteristic value in S13 is: , where ξ is the change feature value, N is the number of pixel values ​​greater than 0 in the picture change image, R is the change pixel, and Z is the number of pixel points in the picture change image.

[0079] In an underground garage, a camera is fixed to monitor a fixed area. The present invention subtracts pixel values ​​of images at adjacent moments at the same pixel point position, takes the absolute value of the subtraction result, and uses the absolute value as the pixel value of the same pixel point position to obtain a picture change image.

[0080] The present invention counts the number of pixel values ​​greater than 0 in the picture change image to obtain the size of the picture change area, combined with the ratio of the change area to the entire monitoring area. , as well as the size of pixel changes, calculate the change feature values ​​and improve the accuracy of screening vehicle driving images.

[0081] The present invention sets two thresholds for the change feature value: an upper change feature threshold and a lower change feature threshold. The upper change feature threshold can be specifically set by the changes in the image when the vehicle drives to the area and images are formed, so as to avoid misclassifying the picture changes caused by pedestrians walking as vehicles. The lower change feature threshold is close to 0.

[0082] In this embodiment, S2 includes the following sub-steps:

[0083] S21, removing the changed position from the vehicle driving image to obtain the unchanged area;

[0084] S22, converting the unchanged area to the HSI space, extracting the brightness I component of each pixel, taking the average of the brightness I components of each pixel, and obtaining the ambient brightness value;

[0085] S23. Calculate the brightness adjustment ratio according to the ambient brightness value: γ=L contrast / L en , where γ is the brightness adjustment ratio, L contrast is the contrast brightness value, L en is the ambient brightness value;

[0086] S24, converting the vehicle driving image into the HSI space, and adjusting the brightness of the brightness I component of each pixel according to the brightness adjustment ratio: ,in, is the brightness I component of the i-th pixel after adjustment, L i is the brightness I component of the i-th pixel before adjustment, where i is a positive integer;

[0087] S25, restoring the brightness-adjusted image to the RGB space to obtain a brightness-adjusted image.

[0088] The present invention removes changed positions from a vehicle driving image, retains unchanged positions, obtains unchanged areas, converts the unchanged areas to an HSI space, extracts a brightness I component, takes an average, obtains an ambient brightness value, and then obtains a brightness adjustment ratio based on a ratio of a contrast brightness value to an ambient brightness value. The brightness adjustment ratio is used to adjust the brightness of the vehicle driving image in the HSI space, thereby improving the accuracy of subsequent color contrast.

[0089] When a vehicle enters the underground garage, there is a gate at the entrance for obtaining the vehicle image and the license plate image. Therefore, in this embodiment, the contrast brightness value L contrastThe ambient brightness value of the unchanged area in the vehicle image when entering the gate in the HSI space can be used to ensure that the background brightness remains consistent during image comparison, making subsequent color comparison more accurate.

[0090] In this embodiment, S3 includes the following sub-steps:

[0091] S31, subtracting pixel values ​​of the vehicle driving image and the static image at the same pixel position to obtain a pixel difference value of each pixel point, and taking an absolute value of the pixel difference value to obtain a pixel difference;

[0092] S32, when the pixel difference is greater than 0, mark the pixel as a suspicious point;

[0093] S33, when there are no other suspicious points within the neighborhood of a suspicious point, discard the suspicious point;

[0094] S34, dividing the remaining suspicious points into a plurality of simply connected regions, wherein the simply connected regions are not in contact with each other;

[0095] S35. Calculate the area difference between each simply connected region and the vehicle imaging area: , where d E is the area difference, E s is the area of ​​the simply connected region, E v is the vehicle imaging area;

[0096] S36: when the area difference is less than the first difference threshold, the simply connected area is a vehicle area;

[0097] S37, extracting the vehicle region from the brightness-adjusted image to obtain a vehicle image to be detected.

[0098] The static image of the present invention is an image without a vehicle traveling, that is, a background image. When the pixel difference is greater than 0, the pixel point is marked as a suspicious point, and the isolated suspicious points are discarded. The suspicious points are classified according to the clustering situation so that the contacting suspicious points are divided into one area to form a simply connected area. The suspicious points clustered together are classified into a simply connected area, and then the difference between the area of ​​the simply connected area and the vehicle imaging area is calculated. When the difference is small, the simply connected area is the vehicle area. When the position of the vehicle area is known, the vehicle area is cut out from the brightness adjustment image to obtain the vehicle image to be detected.

[0099] The vehicle imaging area is the number of pixels occupied by the vehicle in the image.

[0100] In step S37, the vehicle area on the brightness-adjusted image is retained and other areas are discarded.

[0101] The area of ​​the simply connected region and the vehicle imaging area can both be represented by the number of pixels.

[0102] In this embodiment, the first difference threshold is a threshold set for the area difference, and its specific value can be set based on experience or experiments.

[0103] In this embodiment, S4 includes the following sub-steps:

[0104] S41, clustering the pixels according to the pixel values ​​on the image of the vehicle to be detected to obtain a classified image;

[0105] In this embodiment, the pixel value at step S41 can be the R, G, B channel value, or the average value of the three channels of R, G, B of the pixel point, and both can realize the classification of pixel points with similar pixel values ​​into one category;

[0106] S42, finding the largest class in the classified image, and using this class as the main color area of ​​the vehicle body;

[0107] In the present invention, the largest class refers to the class containing the largest number of pixels;

[0108] S43, extracting the mean of the R, G, and B channel values ​​of the main color area of ​​the vehicle body to form a color vector;

[0109] S44, calculating the similarity between the color vector in S43 and the stored color vector storing the vehicle image, to obtain a first similarity;

[0110] S45. When the first similarity is greater than a first similarity threshold, the stored vehicle image corresponding to the stored color vector is a pending vehicle image.

[0111] In the present invention, the elements in the color vector include: an R channel mean, a G channel mean and a B channel mean.

[0112] The present invention classifies pixels with similar values ​​into one class according to the pixel values ​​on the image of the vehicle to be detected, obtains a classified image, takes the largest class as the main color area of ​​the vehicle body, compares the color vector with the stored color vector, and screens out the image of the vehicle to be detected.

[0113] In this embodiment, the stored vehicle image is a vehicle image taken when the vehicle enters the underground garage, and the stored color vector is a color vector of a main color area of ​​the vehicle body of the stored vehicle image.

[0114] In this embodiment, S5 includes the following sub-steps:

[0115] S51, on the image of the vehicle to be detected, extracting other classes that are in contact with the main color area of ​​the vehicle body as contact classes, calculating the color contrast between the contact class and the main color area of ​​the vehicle body, and constructing a color contrast vector;

[0116] S52, calculating the similarity between the color contrast vector of each contact class and the color contrast vector of the contact class in the to-be-determined vehicle image to obtain a second similarity;

[0117] S53: When the second similarity is greater than a second similarity threshold, mark the pending vehicle image as a candidate vehicle image, and mark the pair of contact classes as a suspected similar region pair.

[0118] In this embodiment, the first similarity threshold is a threshold set for the first similarity, and the second similarity threshold is a threshold set for the second similarity, and the specific values ​​can be set according to experience or experiments. The similarity between vectors can be calculated using cosine similarity. When using cosine similarity, the first similarity threshold and the second similarity threshold can be set to 0.5.

[0119] The present invention extracts other classes that are in contact with the main color area of ​​the vehicle body as contact classes. In order to further eliminate the influence of the environment on the imaging effect, the color contrast between these contact classes and the main color area of ​​the vehicle body is calculated, and a color contrast vector is constructed to obtain the color contrast vector of each contact class. The color contrast vectors are compared on the contact class in the vehicle image to be detected and the vehicle image to be determined, and the candidate vehicle images are further screened out, and the two classes corresponding to the second similarity greater than the second similarity threshold are marked as suspected similar area pairs.

[0120] In this embodiment, S51 includes the following sub-steps:

[0121] S511, on the image of the vehicle to be detected, other classes that are in contact with the main color area of ​​the vehicle body are regarded as contact classes;

[0122] S512, extracting the mean values ​​of R, G, and B channels in the contact category;

[0123] S513, taking the ratio of the R channel mean value of the main color area of ​​the vehicle body to the R channel mean value of the contact type as the R channel contrast;

[0124] S514, taking the ratio of the G channel mean value of the main color area of ​​the vehicle body to the G channel mean value of the contact type as the G channel contrast;

[0125] S515, taking the ratio of the B channel mean value of the main color area of ​​the vehicle body to the B channel mean value of the contact type as the B channel contrast;

[0126] S516: Use the R channel contrast, the G channel contrast, and the B channel contrast as elements to construct a color contrast vector.

[0127] The method for obtaining the color contrast vector of the contact class in the pending vehicle image is consistent with the sub-steps of step S51.

[0128] In this embodiment, S6 includes the following sub-steps:

[0129] S61, calculating the class morphology value of each contact class in the suspected similar region pair;

[0130] S62, calculating the morphological gap between the class morphological values ​​of the two contact classes in the suspected similar region pair, wherein the morphological gap is the square of the difference between the class morphological values ​​of the two contact classes;

[0131] S63, when the morphological difference is less than a second difference threshold, marking the contact class belonging to the candidate vehicle image in the suspected similar region pair as a confirmed similar region;

[0132] S64: When it is confirmed that there are multiple similar regions in a candidate vehicle image, the candidate vehicle image and the vehicle image to be detected are the same vehicle, and the recognition of the vehicle image to be detected is completed.

[0133] The second gap threshold is a threshold set for the morphological gap, and its size can be set based on experience or experiments.

[0134] After screening out suspected similar area pairs, the present invention calculates the class morphological values ​​of the two classes in the suspected similar area pairs, and calculates the morphological gap between the two class morphological values. When the morphological gap is less than a second gap threshold, the two classes are a pair with matching color and morphology. When it is confirmed that there are multiple similar areas in a candidate vehicle image, the candidate vehicle image and the vehicle image to be detected are the same vehicle.

[0135] In this embodiment, the formula for calculating the class morphology value in S61 is: , where θ is the class morphology value, P is the perimeter of the contact class, A is the area of ​​the contact class, and r j is the curvature of the jth edge pixel in the contact class, M is the number of edge pixels in the contact class, and j is a positive integer.

[0136] The present invention evaluates the morphology of this type in terms of perimeter, area and curvature.

[0137] The present invention calculates the change characteristic value according to the image at each moment in the video, reflects the image change at different moments, extracts the image containing the moving vehicle, that is, the vehicle driving image, obtains the change position and the static image, and then adjusts the brightness of the vehicle driving image to reduce the influence of the darker light in the underground garage on the image, and further eliminates the background area to obtain the vehicle image to be detected that only contains the vehicle, reducing the interference of background and light.

[0138] The present invention performs clustering processing on the image of a vehicle to be detected, finds out the main color area of ​​the vehicle body, performs color comparison on the main color area of ​​the vehicle body between the image of the vehicle to be detected and the stored vehicle image, screens out the stored vehicle image with similar color on the main color area of ​​the vehicle body, i.e., the to-be-determined vehicle image, and then constructs a color comparison vector according to the class in contact with the main color area of ​​the vehicle body, finds out candidate vehicle images and suspected similar area pairs from multiple to-be-determined vehicle images, determines the similarity of the two images from the class in contact with the main color area of ​​the vehicle body, realizes comparison of local areas, further screens images, and finally checks the class morphology values ​​of the two classes in the suspected similar area pair, judges whether the shape features of the classes are similar, and determines the identity of the vehicle traveling in the underground garage.

[0139] The present invention adjusts the brightness of the vehicle driving image, so that when performing color comparison, the accuracy of color comparison can be improved, and then the background area is eliminated to obtain the vehicle image to be detected that only contains the vehicle, thereby reducing the interference of the background and facilitating the subsequent color comparison of the main color area and the local area of ​​the vehicle body. After the color comparison, the vehicle is further confirmed through morphological comparison, thereby realizing accurate identification of the vehicle in the underground garage, and facilitating the property to manage and track the vehicle driving process.

[0140] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. 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. An intelligent detection method for underground garage property management vehicles, characterized in that: The following steps are involved: S1. Calculate the change feature value according to the image in the video of the underground garage, and obtain the vehicle driving image, change position and static image; S2, removing the changed position from the vehicle driving image, calculating the ambient brightness value, and performing brightness adjustment in the HSI space to obtain a brightness adjusted image; S3, according to the difference between the static image and the vehicle driving image, find out the location of the vehicle, cut out the vehicle area from the brightness adjusted image, and obtain the image of the vehicle to be detected; S4, clustering the image of the vehicle to be detected, finding the main color area of ​​the vehicle body, calculating the color similarity between the image of the vehicle to be detected and the stored vehicle image in the main color area of ​​the vehicle body, and screening the image of the vehicle to be detected; S5, extracting other classes that are in contact with the main color area of ​​the vehicle body as contact classes, calculating the color contrast between the contact class and the main color area of ​​the vehicle body, constructing a color contrast vector, calculating the similarity between the image of the pending vehicle and the image of the vehicle to be detected on the color contrast vector, and finding candidate vehicle images and suspected similar area pairs; The S5 comprises the following sub-steps: S51, on the image of the vehicle to be detected, extracting other classes that are in contact with the main color area of ​​the vehicle body as contact classes, calculating the color contrast between the contact class and the main color area of ​​the vehicle body, and constructing a color contrast vector; S52, calculating the similarity between the color contrast vector of each contact class and the color contrast vector of the contact class in the to-be-determined vehicle image to obtain a second similarity; S53, when the second similarity is greater than a second similarity threshold, marking the pending vehicle image as a candidate vehicle image, and marking the contact class satisfying the second similarity greater than the second similarity threshold as a suspected similar region pair; S6, comparing the class morphology values ​​of the candidate vehicle image and the vehicle image to be detected on the suspected similar area pair, identifying the vehicle image to be detected; The S6 comprises the following sub-steps: S61, calculating the class morphology value of each contact class in the suspected similar region pair; S62, calculating the morphological gap between the class morphological values ​​of the two contact classes in the suspected similar region pair, wherein the morphological gap is the square of the difference between the class morphological values ​​of the two contact classes; S63, when the morphological difference is less than a second difference threshold, marking the contact class belonging to the candidate vehicle image in the suspected similar region pair as a confirmed similar region; S64, when it is confirmed that there are multiple similar regions in a candidate vehicle image, the candidate vehicle image and the vehicle image to be detected are the same vehicle, and the recognition of the vehicle image to be detected is completed; The formula for calculating the class morphology value in S61 is: , where θ is the class morphology value, P is the perimeter of the contact class, A is the area of ​​the contact class, and r j is the curvature of the jth edge pixel in the contact class, M is the number of edge pixels in the contact class, and j is a positive integer.

2. The intelligent detection method for underground garage property management vehicles according to claim 1 is characterized in that: The S1 comprises the following sub-steps: S11. In the video of the underground garage, subtract the pixel values ​​of the image at time t+1 from the image at time t at the same pixel position, and take the absolute value of the subtraction result to obtain the picture change image at time t+1; S12, adding all pixel values ​​in the picture change image to obtain the change pixel; S13, counting the number of pixel values ​​greater than 0 in the picture change image, and calculating the change feature value based on the change pixels; S14, when the change characteristic value is greater than the upper limit change characteristic threshold, there is a vehicle in the video, and the image corresponding to time t+1 is a vehicle driving image; S15, when the change characteristic value is less than the lower limit change characteristic threshold, the picture in the video is still, and the image corresponding to time t+1 is a static image; S16. When the image at time t+1 is a vehicle driving image, mark the pixel points with pixel values ​​greater than 0 in the picture change image at time t+1 to obtain the change position.

3. The intelligent detection method for underground garage property management vehicles according to claim 2 is characterized in that: The formula for calculating the variation characteristic value in S13 is: , where ξ is the change feature value, N is the number of pixel values ​​greater than 0 in the picture change image, R is the change pixel, and Z is the number of pixel points in the picture change image.

4. The intelligent detection method for underground garage property management vehicles according to claim 1 is characterized in that: The S2 comprises the following sub-steps: S21, removing the changed position from the vehicle driving image to obtain the unchanged area; S22, converting the unchanged area to the HSI space, extracting the brightness I component of each pixel, taking the average of the brightness I components of each pixel, and obtaining the ambient brightness value; S23. Calculate the brightness adjustment ratio according to the ambient brightness value: γ=L contrast / L en , where γ is the brightness adjustment ratio, L contrast is the contrast brightness value, L en is the ambient brightness value; S24, converting the vehicle driving image into the HSI space, and adjusting the brightness of the brightness I component of each pixel according to the brightness adjustment ratio: ,in, is the brightness I component of the i-th pixel after adjustment, L i is the brightness I component of the i-th pixel before adjustment, where i is a positive integer; S25, restoring the brightness-adjusted image to the RGB space to obtain a brightness-adjusted image.

5. The intelligent detection method for underground garage property management vehicles according to claim 1 is characterized in that: The S3 comprises the following sub-steps: S31, subtracting pixel values ​​of the vehicle driving image and the static image at the same pixel position to obtain a pixel difference value of each pixel point, and taking an absolute value of the pixel difference value to obtain a pixel difference; S32, when the pixel difference is greater than 0, mark the pixel as a suspicious point; S33, when there are no other suspicious points within the neighborhood of a suspicious point, discard the suspicious point; S34, dividing the remaining suspicious points into a plurality of simply connected regions, wherein the simply connected regions are not in contact with each other; S35. Calculate the area difference between each simply connected region and the vehicle imaging area: , where d E is the area difference, E s is the area of ​​the simply connected region, E v is the vehicle imaging area; S36: when the area difference is less than the first difference threshold, the simply connected area is a vehicle area; S37, extracting the vehicle region from the brightness-adjusted image to obtain a vehicle image to be detected.

6. The intelligent detection method for underground garage property management vehicles according to claim 1 is characterized in that: The S4 comprises the following sub-steps: S41, clustering the pixels according to the pixel values ​​on the image of the vehicle to be detected to obtain a classified image; S42, finding the largest class in the classified image, and using this class as the main color area of ​​the vehicle body; S43, extracting the mean of the R, G, and B channel values ​​of the main color area of ​​the vehicle body to form a color vector; S44, calculating the similarity between the color vector in S43 and the stored color vector storing the vehicle image, to obtain a first similarity; S45. When the first similarity is greater than a first similarity threshold, the stored vehicle image corresponding to the stored color vector is a pending vehicle image.

7. The intelligent detection method for underground garage property management vehicles according to claim 1 is characterized in that: The S51 comprises the following sub-steps: S511, on the image of the vehicle to be detected, other classes that are in contact with the main color area of ​​the vehicle body are regarded as contact classes; S512, extracting the mean values ​​of R, G, and B channels in the contact category; S513, taking the ratio of the R channel mean value of the main color area of ​​the vehicle body to the R channel mean value of the contact type as the R channel contrast; S514, taking the ratio of the G channel mean value of the main color area of ​​the vehicle body to the G channel mean value of the contact type as the G channel contrast; S515, taking the ratio of the B channel mean value of the main color area of ​​the vehicle body to the B channel mean value of the contact type as the B channel contrast; S516: Use the R channel contrast, the G channel contrast, and the B channel contrast as elements to construct a color contrast vector.

Images