Vehicle logo detection and identification method based on vehicle logo repositioning

Abstract

The invention discloses a vehicle logo detection and identification method based on vehicle logo repositioning, relates to the technical field of vehicle logo detection and identification, and is used for solving the problems of insufficient detection and positioning stability and precision caused by multi-angle transformation and easy interference of dynamic factors such as vehicle vibration and illumination variation. After preprocessing, an edge detection algorithm is used to extract vehicle contour information, and a target vehicle image range is screened out. And hub information is identified in the range, and vehicle hub features are generated and used for judging whether vehicle logo detection and positioning are carried out or not. The positioned vehicle logo is recorded, the vehicle jitter frequency and the braking condition are monitored in real time, and the vehicle logo influence characteristics are calculated according to the braking condition and used for setting the vehicle logo influence weight. And in combination with the vehicle jitter frequency, the marked vehicles are screened and detected and sorted, and secondary vehicle logo detection and positioning are performed on the marked vehicles according to a sorting result, so that accurate identification of the vehicle logo is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle logo detection and recognition. More specifically, the present invention relates to a vehicle logo detection and recognition method based on vehicle logo repositioning. Background Art

[0002] With the continuous development of intelligent transportation systems, vehicle recognition and dynamic detection technologies based on video image processing technology have been widely applied in scenarios such as road monitoring, vehicle management, and traffic law enforcement. Among them, the vehicle logo, as an important symbol for vehicle manufacturers to identify, its accurate recognition is crucial for tasks such as vehicle attribute extraction, brand classification, and traffic flow feature analysis.

[0003] The existing technologies have the following deficiencies:

[0004] Currently, vehicle recognition mainly relies on traditional license plate recognition and vehicle image feature extraction methods. However, in complex environments, such as when the license plate is soiled, blocked, or the image is blurred, the accuracy and robustness of license plate recognition significantly decrease. To improve the comprehensiveness and redundancy of recognition, some studies have proposed using vehicle logo recognition as an auxiliary information source. However, the vehicle logo itself is small in area, has multiple-angle transformations, and is easily interfered by dynamic factors such as vehicle vibration and light changes, resulting in insufficient stability and accuracy in its detection and positioning. Therefore, a vehicle logo detection and recognition method based on vehicle logo repositioning is proposed.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a vehicle logo detection and recognition method based on vehicle logo repositioning, and solves the problems raised in the above background art by applying a wheel hub feature assisted judgment mechanism and a vehicle dynamic behavior feature weighting strategy.

[0007] To achieve the above object, the present invention provides the following technical solution. A vehicle logo detection and recognition method based on vehicle logo repositioning includes the following steps:

[0008] Step S1, collect vehicle images through a camera, and after preprocessing, use an edge detection algorithm to extract the vehicle contour information in the vehicle images and screen out the target vehicle image range;

[0009] Step S2, identify the wheel hub information within the target vehicle image range, generate the wheel hub features of the vehicle, and determine whether to detect and position its vehicle logo according to the wheel hub features of the vehicle;

[0010] Step S3: Record the detected and located vehicle logo, monitor the vehicle jitter frequency and the vehicle braking condition in real time, and calculate the vehicle logo influence feature according to the vehicle braking condition;

[0011] Step S4: Set the vehicle logo influence weight according to the vehicle logo influence feature, screen out the marked vehicles in combination with the vehicle jitter frequency and perform detection sorting, and perform secondary detection and location on the marked vehicles according to the detection sorting.

[0012] In a preferred embodiment, in step S1, the camera continuously acquires vehicle images at a preset frame rate, combines the vehicle images into a vehicle image frame sequence in chronological order, and preprocesses the vehicle images through an image processing tool, including grayscale processing and noise suppression processing;

[0013] The grayscale processing is used to convert the vehicle image into a grayscale image;

[0014] The noise suppression processing is used to smooth and reduce the noise of the grayscale image to obtain the filtered vehicle image.

[0015] In a preferred embodiment, in step S1, the edge contour pixels of the vehicle image are extracted by the Canny edge detection algorithm. The Canny edge detection algorithm includes gradient calculation, non-maximum suppression, and double-threshold connection method. The specific steps are as follows:

[0016] After obtaining the gradient intensity value and gradient direction of each pixel through gradient calculation, an edge response map is generated based on non-maximum suppression, and then the edge contour pixels are screened out according to double-threshold connection.

[0017] In a preferred embodiment, in step S1, the edge contour pixels of the target vehicle are screened, the strong edges are detected through an image processing tool, and the area of the closed edge contour in the vehicle image is extracted;

[0018] The minimum circumscribed rectangle is obtained by boundary fitting for the contour area, and its aspect ratio is calculated. A first aspect ratio threshold and a second aspect ratio threshold are preset;

[0019] If the aspect ratio is between the first aspect ratio threshold and the second aspect ratio threshold, it is determined that the contour area is the target vehicle area; otherwise, it is determined that the contour area is the non-target vehicle area.

[0020] In a preferred embodiment, in step S2, based on the target vehicle area, the Hough circle transform method is called through an image processing tool to identify the standard wheel hub area. The identification process satisfies the following rules:

[0021] Rule 1: If the radius of the identified circular area is within the preset radius range, the circular area is retained; otherwise, the circular area is not retained;

[0022] Rule 2: Determine whether the recognized circular area contains edge density and calculate the edge density. If the edge density is greater than the preset edge density threshold, retain the circular area; otherwise, do not retain the circular area.

[0023] If all the above rules are satisfied, then determine that the circular area is the hub area.

[0024] In a preferred embodiment, extract the size feature and texture feature of the hub in the hub area. The hub size feature is the radius of the circular area, and the hub texture feature is the edge density.

[0025] Based on the hub size feature and hub texture feature, construct a logistic regression model to determine whether the target vehicle area is a vehicle area:

[0026] Normalize and sum the hub size feature and hub texture feature as the logistic regression parameter of the hub. Based on the logistic regression parameter of the hub, construct a logistic regression model to obtain the vehicle prediction probability.

[0027] If the vehicle prediction probability is greater than the preset probability threshold, determine that the target vehicle area is a vehicle area and detect and locate the vehicle logo; otherwise, determine that the current area is a non-vehicle area.

[0028] In a preferred embodiment, in step S3, detect and locate the vehicle logo, obtain the center coordinate point of the vehicle logo area, retrieve the center coordinate points of the target areas of each frame in the vehicle image frame sequence, and calculate the change amount of the vehicle logo position between adjacent frames.

[0029] When the change amount of the vehicle logo position is greater than the preset change amount threshold, the corresponding vehicle shakes. Calculate the vehicle shake frequency in the vehicle image frame sequence based on the number of vehicle shakes and the total number of vehicle images in the vehicle image frame sequence.

[0030] Calculate the vehicle speed through the change amount of the vehicle logo position between adjacent frames, and merge the speeds in the vehicle image frame sequence into a vehicle speed sequence.

[0031] The number of braking times is the number of times the speed decreases in the vehicle speed sequence, and only the continuously decreasing part is counted as one time.

[0032] The ratio of the average speed to the maximum speed of the vehicle in the vehicle image frame sequence is used as the braking speed ratio.

[0033] In a preferred embodiment, in step S3, comprehensively utilize the number of braking times and the braking speed ratio to evaluate the vehicle logo influence feature by fuzzy inference:

[0034] Define the number of braking times and the braking speed ratio as input variables and divide them into different fuzzy sets respectively.

[0035] Define the vehicle logo influence feature as the output variable and divide it into a fuzzy set.

[0036] Formulate a set of fuzzy rules to describe the influence of different input variables on the output variable, and evaluate the vehicle logo influence feature according to fuzzy inference.

[0037] In a preferred embodiment, in step S4, obtain the membership degree value in each fuzzy set as the vehicle logo influence weight according to the preset membership function.

[0038] Multiply the vehicle logo influence weight by the vehicle jitter frequency as the comprehensive score value. If the comprehensive score value is greater than the comprehensive score threshold, mark the vehicle; otherwise, do not mark the vehicle.

[0039] Arrange the comprehensive score values of the marked vehicles in descending order, and perform a secondary detection of the vehicle logo images on the marked vehicles in sequence according to the order.

[0040] Technical effects and advantages of the present invention:

[0041] 1. The present invention collects vehicle images through a camera, extracts vehicle contour information in the vehicle images using an edge detection algorithm after preprocessing, and screens out the target vehicle image range. Identify the wheel hub information within the target vehicle image range to generate the wheel hub feature of the vehicle, which is used to determine whether to detect and locate its vehicle logo. Record the detected and located vehicle logo, monitor the vehicle jitter frequency and vehicle braking situation in real time, calculate the vehicle logo influence feature according to the vehicle braking situation, which is used to set the vehicle logo influence weight, and screen out the marked vehicles in combination with the vehicle jitter frequency and perform detection sorting. Perform secondary detection and positioning on the marked vehicles according to the detection sorting, realizing the accurate identification of the vehicle logo. Brief Description of the Drawings

[0042] Figure 1 It is a flowchart for implementing a vehicle logo detection and recognition method based on vehicle logo repositioning according to the present invention.

[0043] Figure 2 It is a step schematic diagram of a vehicle logo detection and recognition method based on vehicle logo repositioning according to the present invention. Detailed Embodiment

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0045] Embodiment 1, a vehicle logo detection and recognition method based on vehicle logo repositioning, as Figures 1 to 2 shown, includes the following steps:

[0046] Step S1: Collect vehicle images through a camera. After preprocessing, use an edge detection algorithm to extract the vehicle contour information in the vehicle images and screen out the target vehicle image range;

[0047] Step S2: Identify the wheel hub information within the target vehicle image range, generate the wheel hub features of the vehicle, and determine whether to detect and locate its vehicle logo based on the wheel hub features of the vehicle;

[0048] Step S3: Record the detected and located vehicle logo, monitor the vehicle jitter frequency and vehicle braking condition in real time, and calculate the vehicle logo influence features according to the vehicle braking condition;

[0049] Step S4: Set the vehicle logo influence weight according to the vehicle logo influence features, screen out the marked vehicles in combination with the vehicle jitter frequency and perform detection sorting, and perform secondary detection and location on the marked vehicles according to the detection sorting.

[0050] The specific implementation is as follows:

[0051] In step S1, the camera continuously collects vehicle images at a preset frame rate, and combines the vehicle images into a vehicle image frame sequence in chronological order. The vehicle images are preprocessed through an image processing tool, including grayscale processing and noise suppression processing;

[0052] The grayscale processing is used to convert the vehicle image into a grayscale image, removing the interference of color information on edge detection. The weighted average method is used to calculate the grayscale value by weighting the three red, green, and blue channels in proportion: , where a, b, and c are preset weight coefficients, , G, and B are the pixel values corresponding to the three red, green, and blue channels respectively, is the pixel coordinate point is the grayscale value of the pixel coordinate point, and the grayscale values of each pixel are arranged in a two-dimensional matrix to form a grayscale image;

[0053] The noise suppression processing is used to smooth and denoise the grayscale image, and a Gaussian function is used to remove the random noise in the grayscale image to obtain the processed vehicle image. The Gaussian function determines the weight factor, and the weighted sum of the pixel values in the neighborhood is used as the central pixel value: , where, is the standard deviation, is the weight factor; Multiply the weight factor by each neighborhood of the grayscale image and sum: , where, is the grayscale value of the neighborhood pixel in the grayscale image, k is the preset kernel radius, is the vehicle image;

[0054] Extract the edge contour pixels of the vehicle image through the Canny edge detection algorithm. The Canny edge detection algorithm includes gradient calculation, non-maximum suppression, and double-threshold connection methods. The specific steps are as follows:

[0055] Gradient calculation method: Use the first-order differential operator to convolve the X-axis and Y-axis of the filtered image respectively, and calculate the gray-scale change rates of each pixel in the vehicle image on the X-axis and Y-axis: , where and are the gray-scale change rates of the vehicle image on the x-axis and y-axis respectively;

[0056] Obtain the gradient intensity of the pixel based on the Euclidean distance: , where G is the pixel gradient intensity value; the pixel gradient intensity value represents the degree of gray-scale change at the corresponding pixel. The larger the pixel gradient intensity value, the more likely it is to be an edge point;

[0057] Determine the corresponding gradient direction through the arctangent function: , where is the gradient direction of the pixel edge.

[0058] Non-maximum suppression method: Retain the pixel with the largest gradient intensity among the adjacent pixels in the gradient direction of the pixel edge, and suppress the gradient response in the non-edge region; compare the gradient intensity value of the current pixel with the gradient intensity values of the two adjacent pixels along the corresponding direction to obtain the edge response map: If the gradient intensity value of the current pixel is not the maximum value in the corresponding positive direction, set the current gradient intensity value to zero, otherwise retain the current gradient intensity value.

[0059] Double-threshold connection method: Preset the first gradient intensity threshold and the second gradient intensity threshold, and the first gradient intensity threshold is greater than the second gradient intensity threshold. Divide the pixel gradient intensity values in the edge response map into different edge types, and screen out the edge contour pixels:

[0060] If the pixel gradient intensity value is greater than the first gradient intensity threshold, it is marked as a strong edge pixel. If the pixel gradient intensity value is less than or equal to the first gradient intensity threshold and greater than the second gradient intensity threshold, it is marked as a weak edge pixel. If the pixel gradient intensity value is less than or equal to the second gradient intensity threshold, it is marked as a non-edge pixel;

[0061] When marked as a strong edge pixel, the corresponding pixel is the edge contour; when marked as a weak edge pixel, if there is a strong edge pixel in the pixel neighborhood, the corresponding pixel is the edge contour, otherwise it is marked as a non-edge pixel.

[0062] Further screen the edge contour pixels of the target vehicle, detect the edge contour through an image processing tool, and extract the area of the edge closed contour in the vehicle image;

[0063] Calculate the aspect ratio of the contour area and determine whether it is a vehicle area: For the contour area, obtain the minimum bounding rectangle by boundary fitting. The height of the minimum bounding rectangle is used as the numerator, the width of the minimum bounding rectangle is used as the denominator, and their ratio is used as the aspect ratio. Preset aspect ratio threshold 1 and aspect ratio threshold 2. If the aspect ratio is between aspect ratio threshold 1 and aspect ratio threshold 2, then determine that the contour area is the target vehicle area; otherwise, determine that the contour area is a non-target vehicle area;

[0064] For example, aspect ratio threshold 1 and aspect ratio threshold 2 are set to 1.2 and 4 respectively. If the aspect ratio is 2, then determine that the contour area is a vehicle contour.

[0065] Through structured image preprocessing and edge analysis, background interference is effectively excluded, the target vehicle area is accurately extracted, ensuring that subsequent steps focus on the effective vehicle image area, and improving the detection efficiency and accuracy.

[0066] It should be noted that the preset frame rate is a parameter configured by the camera, indicating the number of images continuously captured per second, denoted as F. For example, 30 FPS means that 30 frames of vehicle images are collected per second, then the time interval between two adjacent frames is: ; In the Gaussian function, the standard deviation controls the degree of blurring. The larger its value, the flatter the weight distribution, the stronger the filtering effect, and the weaker the retention of image details. It is set by professionals and will not be elaborated here; The preset kernel radius controls the size of the pixel neighborhood participating in the convolution operation and has a linear relationship with the standard deviation, which is set by professionals; The image processing tool is an open-source computer vision library that can perform preprocessing operations and feature extraction on vehicle images, etc.; The Canny edge detection algorithm is a multi-stage edge detection algorithm that suppresses noise through the Gaussian function, combines the first-order differential operator to extract the image gradient information, and accurately retains the real edge structure through non-maximum suppression operation. Finally, the double-threshold connection strategy is used to classify and determine the edge pixels, so as to efficiently extract the vehicle contour boundary; The preset first gradient intensity threshold and second gradient intensity threshold are the classification parameters for edge pixels, which are set by professionals. For example, they are set according to experience or dynamically calculated according to the local or global gradient distribution of the image, etc.;

[0067] In step S2, based on the target vehicle area, the Hough circle transform method is called through the image processing tool to identify the standard wheel hub area. The identification process satisfies the following rules:

[0068] Rule 1: If the radius of the identified circular area is within the preset radius range, then retain the circular area; otherwise, do not retain the circular area;

[0069] Rule 2: Whether the identified circular area contains edge density, excluding the misjudgment of smooth contours or pseudo-circles as wheel hubs, and calculate the edge density index: , where, is the number of edge pixels in the circular region, is the total number of pixels in the circular region, and D is the edge density; if the edge density is greater than the preset edge density threshold then the circular region is retained; otherwise, the circular region is not retained;

[0070] If all the above rules are satisfied, then the circular region is determined to be the wheel hub region;

[0071] Extract the size feature and texture feature of the wheel hub in the wheel hub region. The size feature of the wheel hub is the radius of the circular region identified in Rule 1 above, and the texture feature of the wheel hub is the edge density in Rule 2 above;

[0072] Construct a logistic regression model based on the size feature and texture feature of the wheel hub to determine whether the target vehicle region is an automobile region:

[0073] Normalize and sum the size feature and texture feature of the wheel hub as the logistic regression parameter of the wheel hub, and construct a logistic regression model with the logistic regression parameter of the wheel hub: , where z is the logistic regression parameter of the wheel hub, e is the natural base, L is the calculation result of the logistic regression model corresponding to the wheel hub, and the calculation result of the logistic regression model of the wheel hub is used as the automobile prediction probability;

[0074] If the automobile prediction probability is greater than the preset probability threshold, determine that the target vehicle region is an automobile region and detect and locate the vehicle logo; otherwise, determine that the current region is a non-automobile region.

[0075] Use the stability and unique structure of the vehicle wheel hub as the preliminary screening criterion to realize the preliminary screening of the target vehicle from the image structure level, effectively reduce the number of times of processing invalid images, and reduce the false detection rate of vehicle logo detection.

[0076] It should be noted that the Hough circle transform is an image circular region detection algorithm based on the parameter space voting mechanism. Using the geometric characteristics of the circle, the circular detection problem in the image space is converted into an accumulative voting process in the parameter space, which can be realized through an open-source computer vision library; the preset probability threshold is set by professionals and will not be elaborated here;

[0077] In step S3, for vehicle logo detection and location, obtain the center coordinate point of the vehicle logo region and record it as , retrieve the center coordinate points of the target regions of each frame in the vehicle image frame sequence, and calculate the change amount of the vehicle logo position between adjacent frames: , is the change amount of the vehicle logo position;

[0078] When the change amount of the vehicle logo position is greater than the preset change amount threshold, then the corresponding vehicle shakes, and calculate the vehicle shake frequency in the vehicle image frame sequence: , where, where is the number of vehicle jitters and N is the total number of vehicle images in the vehicle image frame sequence, is the vehicle jitter frequency;

[0079] Calculate the vehicle speed by the change in the position of the vehicle logo between adjacent frames: , where, is the time interval between adjacent frames, is the speed, and the speeds in the vehicle image frame sequence are combined into a vehicle speed sequence;

[0080] The number of times the speed decreases in the vehicle speed sequence is used as the number of braking times, and the continuously decreasing part is only counted once. The formula for the number of braking times is: , where, means that the speed at a certain moment is lower than the previous moment, is the number of braking times;

[0081] The ratio of the average speed to the maximum speed of the vehicle in the vehicle image frame sequence is used as the braking speed ratio: , where, is the maximum speed in the vehicle image frame sequence, is the average speed in the vehicle image frame sequence, and R is the braking speed ratio.

[0082] Comprehensively use the number of braking times and the braking speed ratio to evaluate the vehicle logo influence characteristics by fuzzy inference. The specific steps are as follows:

[0083] Define the number of braking times and the braking speed ratio as input variables and divide them into different fuzzy sets. For example, "high" and "low" for the number of braking times, and "high" and "low" for the braking speed ratio.

[0084] Define the vehicle logo influence characteristic as the output variable and divide it into fuzzy sets. For example, "low", "medium", and "high" for the vehicle logo influence characteristic.

[0085] Formulate a set of fuzzy rules to describe the influence of different input variables on the output variable, and evaluate the vehicle logo influence characteristic according to fuzzy inference. The fuzzy rules can be formulated according to the actual situation. For example:

[0086] When the number of braking times is high and the braking speed ratio is high, the vehicle logo influence degree is high;

[0087] When the number of braking times is medium and the braking speed ratio is medium, the vehicle logo influence degree is medium;

[0088] When the number of braking times is low and the braking speed ratio is low, the vehicle logo influence degree is low; ...

[0090] And so on to form a fuzzy rule set.

[0091] By calculating the vehicle state and the logo position, and combining with the braking behavior analysis, a fuzzy rule set is constructed to obtain the logo influence characteristics, providing key parameter support for subsequent secondary detection.

[0092] It should be noted that the division of the fuzzy set can be adjusted according to the actual situation. For example, if the number of vehicle brakings is less than 3 times, it is calibrated as the "low" fuzzy set; if the number of brakings is between 3 and 7 times, it is calibrated as the "medium" fuzzy set. The preset change amount threshold and the preset probability threshold are set by professionals and will not be elaborated here.

[0093] In step S4, the preset membership function represents the belonging degree of the input variable in the fuzzy rule under different fuzzy sets. According to the preset membership function, the membership degree value in each fuzzy set is obtained as the logo influence weight, denoted as W.

[0094] For example, when the logo influence degree is high, the membership degree is 0.7; when the logo influence degree is medium, the membership degree is 0.2; when the logo influence degree is low, the membership degree is 0.1.

[0095] Calculate the comprehensive score value by integrating the logo influence weight and the vehicle jitter frequency, and mark the vehicles for secondary detection:

[0096] Calculate and sort the comprehensive score values of the vehicles marked for secondary detection: , where S is the comprehensive score value;

[0097] If the comprehensive score value is greater than the comprehensive score threshold, the vehicle is marked; otherwise, the vehicle is not marked.

[0098] Arrange the comprehensive score values of the marked vehicles in descending order, and sequentially perform secondary detection on the logo images of the marked vehicles according to the order.

[0099] Introduce the logo influence weight as a reference for re-identification priority, improve the adaptability of the recognition system to dynamic complex scenarios, avoid missed detection or misdetection of important target logos due to dynamic disturbances, and improve the recognition accuracy and practical application reliability of the system.

[0100] It should be noted that the membership function is used to describe the degree to which the input variable value belongs to a certain fuzzy set, ranging from 0 to 1. The preset membership function is determined by domain professionals according to vehicle operation data and empirical knowledge to determine the starting point, peak point, and ending point of each input variable corresponding to the fuzzy set, thus forming a set of continuous membership function curves; the comprehensive score threshold is set by professionals and will not be elaborated here.

[0101] The above formulas are all dimensionless and only take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulations to get a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0102] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0103] It should be understood that in various embodiments of the present application, the magnitudes of the sequence numbers of the above processes do not mean the order of execution is prior or subsequent. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0104] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0105] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0106] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0107] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0108] In addition, in each embodiment of this application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0109] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of this application, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0110] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claimed rights.

Claims

1. A method for vehicle logo detection and recognition based on vehicle logo repositioning, characterized in that: It includes the following steps: Step S1: Collect vehicle images through a camera. After preprocessing, use an edge detection algorithm to extract the vehicle contour information in the vehicle images and screen out the target vehicle image range; Step S2: Identify the wheel hub information within the target vehicle image range, generate the wheel hub features of the vehicle, and determine whether to detect and locate its vehicle logo based on the wheel hub features of the vehicle; Step S3: Record the detected and located vehicle logo, monitor the vehicle jitter frequency and vehicle braking condition in real time, and calculate the vehicle logo influence features according to the vehicle braking condition; Step S4: Set the vehicle logo influence weight according to the vehicle logo influence features, screen out the marked vehicles in combination with the vehicle jitter frequency and perform detection sorting, and perform secondary detection and location on the marked vehicles according to the detection sorting.

2. A vehicle logo detection and recognition method based on vehicle logo repositioning according to claim 1, characterized in that: In step S1, the camera continuously collects vehicle images at a preset frame rate, and combines the vehicle images into a vehicle image frame sequence in chronological order. The vehicle images are preprocessed through an image processing tool, including grayscale processing and noise suppression processing; The grayscale processing is used to convert the vehicle images into grayscale images; The noise suppression processing is used to smooth and reduce noise on the grayscale images to obtain the filtered vehicle images.

3. A vehicle logo detection and recognition method based on vehicle logo repositioning according to claim 2, characterized in that: In step S1, extract the edge contour pixels of the vehicle images through the Canny edge detection algorithm. The Canny edge detection algorithm includes gradient calculation, non-maximum suppression, and double-threshold connection methods. The specific steps are as follows: After obtaining the gradient intensity values and gradient directions of each pixel through gradient calculation, generate an edge response map based on non-maximum suppression, and then screen out the edge contour pixels according to double-threshold connection.

4. A vehicle logo detection and recognition method based on vehicle logo repositioning according to claim 3, characterized in that: In step S1, screen the edge contour pixels of the target vehicle, detect the strong edges through an image processing tool, and extract the area of the edge closed contour in the vehicle images; Obtain the minimum circumscribed rectangle for the contour area through boundary fitting, calculate its aspect ratio, and preset the first aspect ratio threshold and the second aspect ratio threshold; If the aspect ratio is between the first aspect ratio threshold and the second aspect ratio threshold, it is determined that the contour area is the target vehicle area; otherwise, it is determined that the contour area is a non-target vehicle area.

5. A vehicle logo detection and recognition method based on vehicle logo repositioning according to claim 4, characterized in that: In step S2, based on the target vehicle area, call the Hough circle transform method through an image processing tool to identify the standard wheel hub area. The recognition process satisfies the following rules: Rule 1: If the radius of the identified circular area is within the preset radius range, the circular area is retained; otherwise, the circular area is not retained; Rule 2: Whether there is edge density inside the identified circular area, calculate the edge density; if the edge density is greater than the preset edge density threshold, the circular area is retained; otherwise, the circular area is not retained; If both of the above rules are satisfied, it is determined that the circular area is the wheel hub area.

6. A method for vehicle logo detection and recognition based on vehicle logo repositioning according to claim 5, characterized in that: Extract the size feature and texture feature of the hub in the hub area. The hub size feature is the radius of the circular area, and the hub texture feature is the edge density; Construct a logistic regression model based on the hub size feature and the hub texture feature to determine whether the target vehicle area is a vehicle area: Normalize and sum the hub size feature and the hub texture feature as the logistic regression parameter of the hub. Construct a logistic regression model based on the logistic regression parameter of the hub to obtain the vehicle prediction probability; If the vehicle prediction probability is greater than the preset probability threshold, determine that the target vehicle area is a vehicle area and detect and locate the vehicle logo; otherwise, determine that the current area is a non-vehicle area.

7. A method for vehicle logo detection and recognition based on vehicle logo repositioning according to claim 6, characterized in that: In step S3, detect and locate the vehicle logo, obtain the central coordinate point of the vehicle logo area, retrieve the central coordinate points of the target areas of each frame in the vehicle image frame sequence, and calculate the change amount of the vehicle logo position between adjacent frames; When the change amount of the vehicle logo position is greater than the preset change amount threshold, the corresponding vehicle shakes. Calculate the vehicle shake frequency in the vehicle image frame sequence based on the number of vehicle shakes and the total number of vehicle images in the vehicle image frame sequence; Calculate the vehicle speed based on the change amount of the vehicle logo position between adjacent frames, and merge the speeds in the vehicle image frame sequence into a vehicle speed sequence; The number of braking times is the number of times the speed decreases in the vehicle speed sequence, and only the continuously decreasing part is counted as one time; The ratio of the average speed to the maximum speed of the vehicle in the vehicle image frame sequence is used as the braking speed ratio.

8. A method for vehicle logo detection and recognition based on vehicle logo repositioning according to claim 7, characterized in that: In step S3, comprehensively utilize the number of braking times and the braking speed ratio to evaluate the vehicle logo influence feature by fuzzy inference: Define the number of braking times and the braking speed ratio as input variables and divide them into different fuzzy sets respectively; Define the vehicle logo influence feature as the output variable and divide it into a fuzzy set; Formulate a set of fuzzy rules to describe the influence of different input variables on the output variable, and evaluate the vehicle logo influence feature according to fuzzy inference.

9. A method for vehicle logo detection and recognition based on vehicle logo repositioning according to claim 8, characterized in that: In step S4, obtain the membership degree value in each fuzzy set according to the preset membership function as the vehicle logo influence weight; The product of the vehicle logo influence weight and the vehicle shake frequency is used as the comprehensive score value. If the comprehensive score value is greater than the comprehensive score threshold, mark the vehicle; otherwise, do not mark the vehicle; Arrange the comprehensive score values of the marked vehicles in descending order, and perform secondary detection of the vehicle logo images on the marked vehicles in sequence.

Images