Vehicle-mounted infrared system object identification method and system based on image identification

Abstract

The invention discloses a vehicle-mounted infrared system object recognition method and system based on image recognition, and relates to the technical field of image recognition, and the method comprises the following steps: S100, carrying out the modeling of a historical infrared image, obtaining an infrared image model, S200, constructing the association relation of a target feature, a background feature and a noise feature according to the infrared image model, and S300, carrying out the recognition of the target feature, the background feature and the noise feature according to the association relation. And based on the association relationship, identifying the object, and outputting an identification result. Through modeling and feature association, targets, backgrounds and noise are effectively distinguished, the false drop rate is reduced, modeling is carried out based on historical data, infrared image changes of different time, weather and seasons can be adapted, noise features are clearly separated, the stability of a system in a complex environment is improved, continuous updating can be carried out along with data accumulation, incremental learning of new target types is supported, and the method is suitable for large-scale popularization and application. Infrared imaging does not depend on visible light, and reliable recognition under severe conditions such as night, foggy days and strong light is achieved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and more particularly to an object recognition method and system for vehicle-mounted infrared systems based on image recognition. Background Technology

[0002] In recent years, vehicle-mounted infrared object recognition technology has developed rapidly, achieving intelligent perception systems. It adopts deep learning frameworks, such as the YOLO series algorithms and Faster R-CNN, to achieve real-time detection and classification. It integrates biometric technologies such as gait analysis, body temperature distribution, and contour features. Through continuous learning and algorithm optimization, it reduces the false judgment rate. The fusion of infrared and other sensors solves the problem of perception failure in complex environments. It relies on AI algorithms to improve recognition accuracy and multi-sensor fusion to enhance the reliability of environmental perception.

[0003] Currently, Chinese invention patent CN119821153A discloses a vehicle driving warning and braking control method and system based on visible light and infrared. This method collects dual-modal data through a visible light camera and an infrared thermal imager, preprocesses the dual-modal data, fuses the features of the preprocessed dual-modal data, and inputs it into a YOLOv5 model for training to obtain a YOLO-DMFF-detect model. The YOLO-DMFF detect model is used to detect target objects and measure distances. Based on the detected target objects and distances, the emergency braking module is controlled to issue voice warnings and perform emergency braking. By fusing recognition and ranging technologies, object recognition and vehicle warning and braking functions are integrated, realizing the functional connection from target detection to vehicle control. However, the related technology relies on visible light and auxiliary features as feature references, making the object recognition process more complex and significantly increasing the calibration difficulty. This is not conducive to adaptability to different scenarios. It does not use the relevant features of the vehicle infrared image itself to screen and search for key recognition areas of objects, which is not conducive to the rapid response and simplicity of object recognition, and has certain limitations. Summary of the Invention

[0004] The technical problem solved by this invention is that the related technologies rely on visible light and auxiliary features as feature references, which makes the object recognition process more complicated and the calibration difficulty is greatly increased. This is not conducive to adaptability to different scenarios. The key recognition areas of the object are not screened and searched based on the relevant features of the vehicle infrared image itself, which is not conducive to the rapid response and simplicity of object recognition, and has certain limitations.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution. In the first aspect, an object recognition method for a vehicle-mounted infrared system based on image recognition includes the following steps: Step S100, modeling historical infrared images to obtain an infrared image model;

[0006] Step S200: Based on the infrared image model, construct the correlation relationship between target features, background features, and noise features;

[0007] Step S300: Based on the association relationship, identify the object and output the identification result;

[0008] The expression for the infrared image model is: ;

[0009] in, Let (i, j) be the k-th historical infrared image, where (i, j) is the pixel coordinate position in the historical infrared image. This represents the infrared feedback signal intensity at position (i, j) of the pixel region after the first marking in the k-th historical infrared image. This represents the infrared feedback signal intensity at position (i, j) in the background region of the k-th historical infrared image. It is represented as the intensity of the infrared feedback signal at position (i, j) in the noise part of the k-th historical infrared image.

[0010] As a preferred embodiment of the image recognition-based object recognition method for vehicle infrared systems according to the present invention, step S100 includes the following sub-steps: step S101, obtaining the vehicle model and retrieving the historical infrared image corresponding to the vehicle model.

[0011] Step S102: Preprocess the historical infrared image, including nonlocal mean filtering noise reduction and CLAHE contrast enhancement.

[0012] Step S103: Select any preprocessed historical infrared image, extract the shape feature quantity of the historical infrared image, and record it as the first feature quantity;

[0013] Step S104: Retrieve the infrared object recognition standard image library, select any standard image from the infrared object recognition standard image library, extract the shape feature quantity of the standard image, and record it as the second feature quantity;

[0014] Step S105: Calculate the similarity between the first feature and the second feature using the cosine similarity formula, set the first value as the similarity threshold, and compare the similarity with the first value.

[0015] When the similarity is greater than or equal to the first value, proceed to step S106;

[0016] When the similarity is less than the first value, proceed to step S104, wherein the selected standard image does not include the already selected standard image;

[0017] Step S106: Set the object name corresponding to the standard image to the object name corresponding to the historical infrared image, mark the pixel part of the object name in the historical infrared image with a first mark, and set the object name as the label of the pixel area after the first mark.

[0018] Step S107: Set the unmarked pixels in the historical infrared image as the background part, calculate the average gray value of the background part, set the second value as the gray value error value, and identify the noise part based on the average gray value and the second value.

[0019] Step S108: Construct an infrared image model based on the background portion, the noise portion, and the pixel region after the first mark.

[0020] As a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems according to the present invention, step S107 further includes the following sub-step: step S1071, obtaining the average gray value of the background portion of any historical infrared image.

[0021] Step S1072: Retrieve the grayscale value lookup table, input the average grayscale value into the grayscale lookup table, and obtain the temperature value corresponding to the average grayscale value.

[0022] Step S1073: Calculate the second value based on the temperature value and the basic temperature fluctuation noise formula, and set the second value as the grayscale error value.

[0023] Step S1074: Calculate the first sum of the average gray value and the second value, and calculate the first difference between the average gray value and the second value. Use the first sum as the upper limit of the gray range and the first difference as the lower limit of the gray range to construct the gray range.

[0024] Step S1075: Compare the gray values ​​of each pixel in the background portion of the historical infrared image with the gray range.

[0025] When the gray value of a pixel is distributed within the gray range, jump to the next pixel and repeat the process of comparing the gray value of the pixel with the gray range. Here, the distribution means that the gray value of the pixel is less than or equal to the upper limit of the gray range and greater than or equal to the lower limit of the gray range.

[0026] When the gray value of a pixel is not distributed within the gray range, the pixel is set as noise. Here, "not distributed" means that the gray value of the pixel is greater than the upper limit of the gray range or the gray value of the pixel is less than the lower limit of the gray range.

[0027] As a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems described in this invention, before constructing the infrared image model, the background portion is automatically updated to the remaining background portion after removing the noise portion.

[0028] In a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems according to the present invention, the expression for the infrared image model is as follows: ;

[0029] in, Let (i, j) be the k-th historical infrared image, where (i, j) is the pixel coordinate position in the historical infrared image. This represents the infrared feedback signal intensity at position (i, j) of the pixel region after the first marking in the k-th historical infrared image. This represents the infrared feedback signal intensity at position (i, j) in the background region of the k-th historical infrared image. It is represented as the intensity of the infrared feedback signal at position (i, j) in the noise part of the k-th historical infrared image.

[0030] As a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems according to the present invention, a new historical infrared image is retrieved, and the background part, noise part, and pixel region after the first mark are extracted from the new historical infrared image according to the infrared image model.

[0031] The statistical characteristics of the background, noise and pixel regions after the first mark are extracted from the new historical infrared image, and the statistical characteristics of the background, noise and pixel regions after the first mark are set as background features, noise features and target features, respectively.

[0032] Statistical characteristics include gray-level variance and gray-level standard deviation;

[0033] The fourth and fifth values ​​are set as variance cutoff values, and the sixth and seventh values ​​are set as standard deviation cutoff values. Based on the variance cutoff values, the various statistical characteristics are classified to obtain the first type of variance, the second type of variance, the third type of variance, the first type of standard deviation, the second type of standard deviation, and the third type of standard deviation.

[0034] Based on the permutation and combination methods and the statistical characteristics after classification, construct the modes;

[0035] The probability of the statistical characteristics of the background, the noise, and the pixel region after the first mark appearing together under each modality is denoted as the first probability.

[0036] The eighth value is set as the probability threshold. Based on the probability threshold and the first probability, the correlation between target features, background features and noise features is constructed.

[0037] As a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems described in this invention, the method for classifying various statistical characteristics according to the variance boundary value includes obtaining target features, background features, noise features, variance boundary value, and standard deviation boundary value.

[0038] Compare the gray-level variance in the target features with the variance threshold, and compare the gray-level standard deviation in the target features with the standard deviation threshold;

[0039] Compare the gray-level variance in the background features with the variance threshold, and compare the gray-level standard deviation in the background features with the standard deviation threshold;

[0040] Compare the gray-level variance in the noise features with the variance threshold, and compare the gray-level standard deviation in the noise features with the standard deviation threshold;

[0041] The comparison process for the gray-level variance and gray-level standard deviation corresponding to the target feature, background feature, and noise feature is logically the same, and is represented as a feature comparison process.

[0042] As a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems described in this invention, the feature comparison process includes classifying the grayscale variance as a first type of variance when the grayscale variance is less than or equal to the fourth value.

[0043] When the grayscale variance is greater than the fourth value and less than or equal to the fifth value, the grayscale variance is classified as the second type of variance.

[0044] When the grayscale variance is greater than the fifth value, the grayscale variance is classified as the third type of variance.

[0045] When the standard deviation of grayscale is less than or equal to the sixth value, the grayscale variance is classified as the first type of standard deviation.

[0046] When the standard deviation of grayscale is greater than the sixth value and less than or equal to the seventh value, the grayscale variance is classified as the second type of standard deviation.

[0047] When the standard deviation of grayscale is greater than the seventh value, the grayscale variance is classified as the third type of standard deviation.

[0048] Based on permutation and combination methods and statistical characteristics after classification, the methods for constructing modes include permutations and combinations corresponding to background features, noise features, and target features, respectively represented as A. w B u and C h ;

[0049] Where w, u, and h represent the w-th permutation and combination corresponding to the background features, the u-th permutation and combination corresponding to the noise features, and the h-th permutation and combination corresponding to the target features;

[0050] The permutations and combinations are first sorted in ascending order by variance classification number, and then in ascending order by standard deviation classification number.

[0051] Modal representation is A w B u C h ;

[0052] The method for calculating the first probability includes calculating the number of new historical infrared images, denoted as the first quantity;

[0053] For any given modality, count the number of new historical infrared images that match that modality, and denote this as the second quantity.

[0054] Calculate the ratio of the second quantity to the first quantity, and denote the ratio of the second quantity to the first quantity as the first probability;

[0055] Iterate through each mode to obtain the first probability corresponding to each mode.

[0056] As a preferred embodiment of the image recognition-based object recognition method for vehicle-mounted infrared systems according to the present invention, the method for constructing the association relationship includes: obtaining a first probability and a probability threshold; comparing the first probability with the probability threshold; and when the first probability is less than or equal to the probability threshold, jumping to the next mode.

[0057] When the first probability is greater than the probability threshold, the mode is marked as a high-confidence mode;

[0058] When all the first probabilities are traversed and no high-confidence mode is found, the probability threshold is lowered by 0.05 and the comparison is repeated until at least one high-confidence mode is found, at which point the probability threshold is stopped from being lowered.

[0059] Obtain each high-confidence mode and its corresponding first probability;

[0060] When there are two or more identical permutations in a high-confidence mode, i.e., the corresponding A... w B u and C h When there are two or more identical permutations and combinations in the classification of the statistical characteristics represented, the first probabilities of the high-confidence modes are compared, the high-confidence modes with the highest first probability are retained, and the remaining high-confidence modes with the highest first probability are deleted.

[0061] When there are no two or more identical permutations in the high-confidence mode, i.e., the corresponding A... w B uand C h When there are no two or more identical permutations or combinations of the statistical characteristics represented, the high-confidence mode is retained.

[0062] Obtain the retained high-confidence modes and construct the mapping relationship between the permutations and combinations corresponding to the target features and the permutations and combinations corresponding to the background features and noise features;

[0063] Construct the association between the permutations and combinations corresponding to the target features and the item names;

[0064] Acquire the infrared image to be identified, and extract the background and noise components based on the infrared image model;

[0065] Extract the statistical features of the background and noise parts, and obtain the permutations and combinations corresponding to the background features and noise features based on the statistical features of the background and noise parts. Retrieve the mapping relationship, input the permutations and combinations corresponding to the background features and noise features into the mapping relationship, and obtain the permutations and combinations corresponding to the target features corresponding to the permutations and combinations corresponding to the background features and noise features.

[0066] Retrieve the association relationships, input the permutations and combinations corresponding to the target features into the association relationships, obtain the item names, and output the item names.

[0067] Secondly, an image recognition-based vehicle infrared system for object recognition includes a construction module, an association module, and a recognition module.

[0068] The construction module models historical infrared images to obtain infrared image models;

[0069] The association module constructs the association relationship between target features, background features, and noise features based on the infrared image model;

[0070] The recognition module identifies objects based on association relationships and outputs the recognition results.

[0071] The beneficial effects of this invention are as follows: by modeling and feature association, it can effectively distinguish between targets, background and noise, reduce the false detection rate, model based on historical data, adapt to changes in infrared images at different times, weather and seasons, clearly separate noise features, improve the stability of the system in complex environments, can be continuously updated with data accumulation, support incremental learning of new target types, and infrared imaging does not depend on visible light, enabling reliable identification under adverse conditions such as night, fog and strong light. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the basic process of an image recognition-based object recognition method for a vehicle-mounted infrared system, provided in one embodiment of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0074] It should be understood that the step numbers used herein are for ease of description only and are not intended to limit the order in which the steps are performed. It should also be understood that the terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention.

[0075] As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0076] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0077] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0078] Example, refer to Figure 1 As an embodiment of the present invention, an object recognition method for a vehicle-mounted infrared system based on image recognition is provided, including the following steps: Step S100, modeling historical infrared images to obtain an infrared image model;

[0079] Step S200: Based on the infrared image model, construct the correlation relationship between target features, background features, and noise features;

[0080] Step S300: Based on the association relationship, identify the object and output the identification result.

[0081] More preferably, by modeling and feature association, the system can effectively distinguish between targets, background and noise, reduce false detection rate, adapt to changes in infrared images at different times, weather and seasons based on historical data modeling, clearly separate noise features, improve the stability of the system in complex environments, can be continuously updated with data accumulation, support incremental learning of new target types, and infrared imaging does not rely on visible light, enabling reliable identification under adverse conditions such as night, fog and strong light.

[0082] Step S100 includes the following sub-steps: Step S101, obtain the vehicle model and retrieve the historical infrared image corresponding to the vehicle model.

[0083] Step S102: Preprocess the historical infrared image, including nonlocal mean filtering noise reduction and CLAHE contrast enhancement.

[0084] Step S103: Select any preprocessed historical infrared image, extract the shape feature quantity of the historical infrared image, and record it as the first feature quantity;

[0085] Step S104: Retrieve the infrared object recognition standard image library, select any standard image from the infrared object recognition standard image library, extract the shape feature quantity of the standard image, and record it as the second feature quantity;

[0086] Step S105: Calculate the similarity between the first feature and the second feature using the cosine similarity formula, set the first value as the similarity threshold, and compare the similarity with the first value.

[0087] When the similarity is greater than or equal to the first value, proceed to step S106;

[0088] When the similarity is less than the first value, proceed to step S104, wherein the selected standard image does not include the already selected standard image;

[0089] Step S106: Set the object name corresponding to the standard image to the object name corresponding to the historical infrared image, mark the pixel part of the object name in the historical infrared image with a first mark, and set the object name as the label of the pixel area after the first mark.

[0090] Step S107: Set the unmarked pixels in the historical infrared image as the background part, calculate the average gray value of the background part, set the second value as the gray value error value, and identify the noise part based on the average gray value and the second value.

[0091] Step S108: Construct an infrared image model based on the background portion, the noise portion, and the pixel region after the first mark.

[0092] Further preferably, the nonlocal mean filtering noise reduction method includes: selecting any pixel in the historical infrared image, selecting a search window centered on the pixel, calculating the grayscale difference between the pixel and each neighboring pixel within the search window, calculating the weighted average of the grayscale differences corresponding to the pixel, setting the weighted average of the grayscale differences corresponding to the pixel as the new grayscale value of the pixel, traversing each pixel in the historical infrared image to obtain the new grayscale value corresponding to each pixel in the historical infrared image;

[0093] Traverse each historical infrared image to obtain the new grayscale value corresponding to each pixel in each historical infrared image.

[0094] At this point, the nonlocal mean filtering noise reduction process is complete.

[0095] More preferably, the nonlocal mean filtering noise reduction method effectively suppresses background clutter and random noise while preserving target details, and significantly improves target distinguishability, especially under low signal-to-noise ratio conditions, thereby enhancing the accuracy and robustness of historical infrared images.

[0096] More preferably, the CLAHE contrast enhancement processing method includes dividing the preprocessed infrared image into local blocks of a first size, performing histogram equalization independently on each block, setting the contrast coefficient to 2, setting the third value as the clipping threshold, truncating histograms exceeding the clipping threshold and distributing them evenly to other histograms, and then performing bilinear interpolation fusion on the equalized local blocks. At this point, the CLAHE contrast enhancement processing steps are completed.

[0097] More preferably, the CLAHE contrast enhancement processing method is used to output historical infrared images with global contrast enhancement and local detail preservation, eliminating block artifacts and providing high-quality input for subsequent target feature analysis.

[0098] More preferably, both the first and second features are extracted using contour extraction algorithms, such as the Canny edge detection algorithm, the Sobel operator algorithm, and the Laplacian operator. Extracting shape features using contour extraction algorithms is an existing technology and will not be elaborated here.

[0099] More preferably, the first mark is represented as a color block in which the pixel portion of the object name in the historical infrared image is filled with a first color and a first brightness;

[0100] The label of the pixel region after the first label is the natural language expression of the name of the object corresponding to that pixel region; that is, the label of the pixel region after the first label is a semantic label.

[0101] Different object names are configured with color blocks of different colors and brightness to ensure that semantic tags are distinguishable in grayscale space.

[0102] More preferably, the average gray value of the background portion is expressed as the arithmetic mean of the gray values ​​of all pixels in the background region in any historical thermal infrared image after the first mark.

[0103] Step S107 also includes the following sub-steps: Step S1071, obtain the average gray value of the background part of any historical infrared image;

[0104] Step S1072: Retrieve the grayscale value lookup table, input the average grayscale value into the grayscale lookup table, and obtain the temperature value corresponding to the average grayscale value.

[0105] Step S1073: Calculate the second value based on the temperature value and the basic temperature fluctuation noise formula, and set the second value as the grayscale error value.

[0106] Step S1074: Calculate the first sum of the average gray value and the second value, and calculate the first difference between the average gray value and the second value. Use the first sum as the upper limit of the gray range and the first difference as the lower limit of the gray range to construct the gray range.

[0107] Step S1075: Compare the gray values ​​of each pixel in the background portion of the historical infrared image with the gray range.

[0108] When the gray value of a pixel is distributed within the gray range, jump to the next pixel and repeat the process of comparing the gray value of the pixel with the gray range. Here, the distribution means that the gray value of the pixel is less than or equal to the upper limit of the gray range and greater than or equal to the lower limit of the gray range.

[0109] When the gray value of a pixel is not distributed within the gray range, the pixel is set as noise. Here, "not distributed" means that the gray value of the pixel is greater than the upper limit of the gray range or the gray value of the pixel is less than the lower limit of the gray range.

[0110] More preferably, the grayscale value lookup table is represented as follows: In the temperature measurement method of a narrow-band infrared thermal imager of 3.2–3.4 micrometers, using an infrared thermal imager and a standard radiation source blackbody, infrared images of the blackbody at different temperatures are acquired at a first ambient temperature. Grayscale values ​​of the blackbody images are calculated using mathematical modeling software, thereby obtaining the correlation between image grayscale values ​​and temperature. Based on the least squares method and interpolation fitting, a blackbody calibration curve is constructed. The ambient temperature is deduced based on the obtained calibration curve and existing grayscale values, thus constructing a table of the correspondence between grayscale values ​​and temperature.

[0111] More preferably, the formula for the base temperature fluctuation noise is: ;

[0112] in, Where is Boltzmann's constant, and T is the temperature value corresponding to the average gray value. For detector bandwidth, This represents the thermal conductivity of a pixel unit. This is represented as the second difference, i.e. The second difference is the grayscale error value corresponding to the temperature value; different temperature values ​​correspond to different grayscale error values.

[0113] Before constructing the infrared image model, the background is automatically updated to the background remaining after noise has been removed.

[0114] The expression for the infrared image model is: ;

[0115] in, Let (i, j) be the k-th historical infrared image, where (i, j) is the pixel coordinate position in the historical infrared image. This represents the infrared feedback signal intensity at position (i, j) of the pixel region after the first marking in the k-th historical infrared image. This represents the infrared feedback signal intensity at position (i, j) in the background region of the k-th historical infrared image. It is represented as the intensity of the infrared feedback signal at position (i, j) in the noise part of the k-th historical infrared image.

[0116] Furthermore, the model can effectively separate the target, background and noise components, providing theoretical support for subsequent adaptive detection. Combined with the temperature fluctuation noise formula and gray value lookup table, it can dynamically correct the signal deviation caused by environmental temperature changes and improve the signal-to-noise ratio of dark targets in complex backgrounds.

[0117] Retrieve new historical infrared images, and extract the background, noise, and pixel regions after the first mark from the new historical infrared images based on the infrared image model;

[0118] The statistical characteristics of the background, noise and pixel regions after the first mark are extracted from the new historical infrared image, and the statistical characteristics of the background, noise and pixel regions after the first mark are set as background features, noise features and target features, respectively.

[0119] Statistical characteristics include gray-level variance and gray-level standard deviation;

[0120] The fourth and fifth values ​​are set as variance cutoff values, and the sixth and seventh values ​​are set as standard deviation cutoff values. Based on the variance cutoff values, the various statistical characteristics are classified to obtain the first type of variance, the second type of variance, the third type of variance, the first type of standard deviation, the second type of standard deviation, and the third type of standard deviation.

[0121] Based on the permutation and combination methods and the statistical characteristics after classification, construct the modes;

[0122] The probability of the statistical characteristics of the background, the noise, and the pixel region after the first mark appearing together under each modality is denoted as the first probability.

[0123] The eighth value is set as the probability threshold. Based on the probability threshold and the first probability, the correlation between target features, background features and noise features is constructed.

[0124] More preferably, this correlation allows for a rapid assessment of the target's confidence level in a dynamic scene, thereby improving the real-time performance and accuracy of target detection.

[0125] Based on variance cutoff values, methods for classifying various statistical characteristics include obtaining target features, background features, noise features, variance cutoff values, and standard deviation cutoff values.

[0126] Compare the gray-level variance in the target features with the variance threshold, and compare the gray-level standard deviation in the target features with the standard deviation threshold;

[0127] Compare the gray-level variance in the background features with the variance threshold, and compare the gray-level standard deviation in the background features with the standard deviation threshold;

[0128] Compare the gray-level variance in the noise features with the variance threshold, and compare the gray-level standard deviation in the noise features with the standard deviation threshold;

[0129] The comparison process for the gray-level variance and gray-level standard deviation corresponding to the target feature, background feature, and noise feature is logically the same, and is represented as a feature comparison process.

[0130] The feature comparison process includes classifying the gray-level variance as a type I variance when the gray-level variance is less than or equal to the fourth value.

[0131] When the grayscale variance is greater than the fourth value and less than or equal to the fifth value, the grayscale variance is classified as the second type of variance.

[0132] When the grayscale variance is greater than the fifth value, the grayscale variance is classified as the third type of variance.

[0133] When the standard deviation of grayscale is less than or equal to the sixth value, the grayscale variance is classified as the first type of standard deviation.

[0134] When the standard deviation of grayscale is greater than the sixth value and less than or equal to the seventh value, the grayscale variance is classified as the second type of standard deviation.

[0135] When the standard deviation of grayscale is greater than the seventh value, the grayscale variance is classified as the third type of standard deviation.

[0136] More preferably, the three types of variance and three types of standard deviation thus constructed form nine modal combinations, that is, there are nine combinations for each of the background features, noise features, and target features. When the three are combined, there are 729 combinations. Each combination corresponds to specific scene priors and noise distribution characteristics. Through training with historical data, high-confidence modal paths can be screened, thereby enabling the selection of reliable modal combinations and the construction of the correlation between target features, background features, and noise features. Target features can be quickly mapped from background features and noise features, simplifying the target detection process, significantly reducing computational complexity, effectively reducing noise interference with target recognition, and improving detection accuracy.

[0137] Based on permutation and combination methods and statistical characteristics after classification, the methods for constructing modes include permutations and combinations corresponding to background features, noise features, and target features, respectively represented as A. w B u and C h ;

[0138] Where w, u, and h represent the w-th permutation and combination corresponding to the background features, the u-th permutation and combination corresponding to the noise features, and the h-th permutation and combination corresponding to the target features;

[0139] The permutations and combinations are first sorted in ascending order by variance classification number, and then in ascending order by standard deviation classification number.

[0140] Modal representation is A w B u C h ;

[0141] The method for calculating the first probability includes calculating the number of new historical infrared images, denoted as the first quantity;

[0142] For any given modality, count the number of new historical infrared images that match that modality, and denote this as the second quantity.

[0143] Calculate the ratio of the second quantity to the first quantity, and denote the ratio of the second quantity to the first quantity as the first probability;

[0144] Iterate through each mode to obtain the first probability corresponding to each mode.

[0145] The method for constructing the association includes obtaining a first probability and a probability threshold, comparing the first probability with the probability threshold, and jumping to the next modality when the first probability is less than or equal to the probability threshold.

[0146] When the first probability is greater than the probability threshold, the mode is marked as a high-confidence mode;

[0147] When all the first probabilities are traversed and no high-confidence mode is found, the probability threshold is lowered by 0.05 and the comparison is repeated until at least one high-confidence mode is found, at which point the probability threshold is stopped from being lowered.

[0148] Obtain each high-confidence mode and its corresponding first probability;

[0149] When there are two or more identical permutations in a high-confidence mode, i.e., the corresponding A... w B u and C h When there are two or more identical permutations and combinations in the classification of the statistical characteristics represented, the first probabilities of the high-confidence modes are compared, the high-confidence modes with the highest first probability are retained, and the remaining high-confidence modes with the highest first probability are deleted.

[0150] When there are no two or more identical permutations in the high-confidence mode, i.e., the corresponding A... w B u and C h When there are no two or more identical permutations or combinations of the statistical characteristics represented, the high-confidence mode is retained.

[0151] Obtain the retained high-confidence modes and construct the mapping relationship between the permutations and combinations corresponding to the target features and the permutations and combinations corresponding to the background features and noise features;

[0152] Construct the association between the permutations and combinations of target features and the item names.

[0153] More preferably, by introducing spatiotemporal consistency constraints of infrared images, the retained high-confidence modes are dynamically weighted and fused to improve the robustness of target features in complex backgrounds.

[0154] Acquire the infrared image to be identified, and extract the background and noise components based on the infrared image model;

[0155] Extract the statistical features of the background and noise parts, and obtain the permutations and combinations corresponding to the background features and noise features based on the statistical features of the background and noise parts. Retrieve the mapping relationship, input the permutations and combinations corresponding to the background features and noise features into the mapping relationship, and obtain the permutations and combinations corresponding to the target features corresponding to the permutations and combinations corresponding to the background features and noise features.

[0156] Retrieve the association relationships, input the permutations and combinations corresponding to the target features into the association relationships, obtain the item names, and output the item names.

[0157] More preferably, by modeling and feature association, the system can effectively distinguish between targets, background and noise, reduce false detection rate, adapt to changes in infrared images at different times, weather and seasons based on historical data modeling, clearly separate noise features, improve the stability of the system in complex environments, can be continuously updated with data accumulation, support incremental learning of new target types, and infrared imaging does not rely on visible light, enabling reliable identification under adverse conditions such as night, fog and strong light.

[0158] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0159] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for object recognition in a vehicle-mounted infrared system based on image recognition, characterized in that, The process includes the following steps: Step S100, modeling historical infrared images to obtain an infrared image model; Step S200: Based on the infrared image model, construct the correlation relationship between target features, background features, and noise features; Step S300: Based on the association relationship, identify the object and output the identification result; The expression for the infrared image model is: ; in, Let (i, j) be the k-th historical infrared image, where (i, j) is the pixel coordinate position in the historical infrared image. This represents the infrared feedback signal intensity at position (i, j) of the pixel region after the first marking in the k-th historical infrared image. This represents the infrared feedback signal intensity at position (i, j) in the background region of the k-th historical infrared image. It is represented as the intensity of the infrared feedback signal at position (i, j) in the noise part of the k-th historical infrared image.

2. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 1, characterized in that, Step S100 includes the following sub-steps: Step S101, obtain the vehicle model and retrieve the historical infrared image corresponding to the vehicle model. Step S102: Preprocess the historical infrared image, including nonlocal mean filtering noise reduction and CLAHE contrast enhancement. Step S103: Select any preprocessed historical infrared image, extract the shape feature quantity of the historical infrared image, and record it as the first feature quantity; Step S104: Retrieve the infrared object recognition standard image library, select any standard image from the infrared object recognition standard image library, extract the shape feature quantity of the standard image, and record it as the second feature quantity; Step S105: Calculate the similarity between the first feature and the second feature using the cosine similarity formula, set the first value as the similarity threshold, and compare the similarity with the first value. When the similarity is greater than or equal to the first value, proceed to step S106; When the similarity is less than the first value, proceed to step S104, wherein the selected standard image does not include the already selected standard image; Step S106: Set the object name corresponding to the standard image to the object name corresponding to the historical infrared image, mark the pixel part of the object name in the historical infrared image with a first mark, and set the object name as the label of the pixel area after the first mark. Step S107: Set the unmarked pixels in the historical infrared image as the background part, calculate the average gray value of the background part, set the second value as the gray value error value, and identify the noise part based on the average gray value and the second value. Step S108: Construct an infrared image model based on the background portion, the noise portion, and the pixel region after the first mark.

3. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 2, characterized in that, Step S107 also includes the following sub-steps: Step S1071, obtain the average gray value of the background part of any historical infrared image; Step S1072: Retrieve the grayscale value lookup table, input the average grayscale value into the grayscale lookup table, and obtain the temperature value corresponding to the average grayscale value. Step S1073: Calculate the second value based on the temperature value and the basic temperature fluctuation noise formula, and set the second value as the grayscale error value. Step S1074: Calculate the first sum of the average gray value and the second value, and calculate the first difference between the average gray value and the second value. Use the first sum as the upper limit of the gray range and the first difference as the lower limit of the gray range to construct the gray range. Step S1075: Compare the gray values ​​of each pixel in the background portion of the historical infrared image with the gray range. When the gray value of a pixel is distributed within the gray range, jump to the next pixel and repeat the process of comparing the gray value of the pixel with the gray range. Here, the distribution means that the gray value of the pixel is less than or equal to the upper limit of the gray range and greater than or equal to the lower limit of the gray range. When the gray value of a pixel is not distributed within the gray range, the pixel is set as noise. Here, "not distributed" means that the gray value of the pixel is greater than the upper limit of the gray range or the gray value of the pixel is less than the lower limit of the gray range.

4. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 2, characterized in that, Before constructing the infrared image model, the background is automatically updated to the background remaining after noise has been removed.

5. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 1, characterized in that, Retrieve new historical infrared images, and extract the background, noise, and pixel regions after the first mark from the new historical infrared images based on the infrared image model; The statistical characteristics of the background, noise and pixel regions after the first mark are extracted from the new historical infrared image, and the statistical characteristics of the background, noise and pixel regions after the first mark are set as background features, noise features and target features, respectively. Statistical characteristics include gray-level variance and gray-level standard deviation; The fourth and fifth values ​​are set as variance cutoff values, and the sixth and seventh values ​​are set as standard deviation cutoff values. Based on the variance cutoff values, the various statistical characteristics are classified to obtain the first type of variance, the second type of variance, the third type of variance, the first type of standard deviation, the second type of standard deviation, and the third type of standard deviation. Based on the permutation and combination methods and the statistical characteristics after classification, construct the modes; The probability of the statistical characteristics of the background, the noise, and the pixel region after the first mark appearing together under each modality is denoted as the first probability. The eighth value is set as the probability threshold. Based on the probability threshold and the first probability, the correlation between target features, background features and noise features is constructed.

6. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 5, characterized in that, Based on variance cutoff values, methods for classifying various statistical characteristics include obtaining target features, background features, noise features, variance cutoff values, and standard deviation cutoff values. Compare the gray-level variance in the target features with the variance threshold, and compare the gray-level standard deviation in the target features with the standard deviation threshold; Compare the gray-level variance in the background features with the variance threshold, and compare the gray-level standard deviation in the background features with the standard deviation threshold; Compare the gray-level variance in the noise features with the variance threshold, and compare the gray-level standard deviation in the noise features with the standard deviation threshold; The comparison process for the gray-level variance and gray-level standard deviation corresponding to the target feature, background feature, and noise feature is logically the same, and is represented as a feature comparison process.

7. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 6, characterized in that, The feature comparison process includes classifying the gray-level variance as a type I variance when the gray-level variance is less than or equal to the fourth value. When the grayscale variance is greater than the fourth value and less than or equal to the fifth value, the grayscale variance is classified as the second type of variance. When the grayscale variance is greater than the fifth value, the grayscale variance is classified as the third type of variance. When the standard deviation of grayscale is less than or equal to the sixth value, the grayscale variance is classified as the first type of standard deviation. When the standard deviation of grayscale is greater than the sixth value and less than or equal to the seventh value, the grayscale variance is classified as the second type of standard deviation. When the standard deviation of grayscale is greater than the seventh value, the grayscale variance is classified as the third type of standard deviation. Based on permutation and combination methods and statistical characteristics after classification, the methods for constructing modes include permutations and combinations corresponding to background features, noise features, and target features, respectively represented as A. w B u and C h ; Where w, u, and h represent the w-th permutation and combination corresponding to the background features, the u-th permutation and combination corresponding to the noise features, and the h-th permutation and combination corresponding to the target features; The permutations and combinations are first sorted in ascending order by variance classification number, and then in ascending order by standard deviation classification number. Modal representation is A w B u C h ; The method for calculating the first probability includes calculating the number of new historical infrared images, denoted as the first quantity; For any given modality, count the number of new historical infrared images that match that modality, and denote this as the second quantity. Calculate the ratio of the second quantity to the first quantity, and denote the ratio of the second quantity to the first quantity as the first probability; Iterate through each mode to obtain the first probability corresponding to each mode.

8. The object recognition method for a vehicle-mounted infrared system based on image recognition as described in claim 1, characterized in that, The method for constructing the association includes obtaining a first probability and a probability threshold, comparing the first probability with the probability threshold, and jumping to the next modality when the first probability is less than or equal to the probability threshold. When the first probability is greater than the probability threshold, the mode is marked as a high-confidence mode; When all the first probabilities are traversed and no high-confidence mode is found, the probability threshold is lowered by 0.05 and the comparison is repeated until at least one high-confidence mode is found, at which point the probability threshold is stopped from being lowered. Obtain each high-confidence mode and its corresponding first probability; When there are two or more identical permutations in a high-confidence mode, i.e., the corresponding A... w B u and C h When there are two or more identical permutations and combinations in the classification of the statistical characteristics represented, the first probabilities of the high-confidence modes are compared, the high-confidence modes with the highest first probability are retained, and the remaining high-confidence modes with the highest first probability are deleted. When there are no two or more identical permutations in the high-confidence mode, i.e., the corresponding A... w B u and C h When there are no two or more identical permutations or combinations of the statistical characteristics represented, the high-confidence mode is retained. Obtain the retained high-confidence modes and construct the mapping relationship between the permutations and combinations corresponding to the target features and the permutations and combinations corresponding to the background features and noise features; Construct the association between the permutations and combinations corresponding to the target features and the item names; Acquire the infrared image to be identified, and extract the background and noise components based on the infrared image model; Extract the statistical features of the background and noise parts, and obtain the permutations and combinations corresponding to the background features and noise features based on the statistical features of the background and noise parts. Retrieve the mapping relationship, input the permutations and combinations corresponding to the background features and noise features into the mapping relationship, and obtain the permutations and combinations corresponding to the target features corresponding to the permutations and combinations corresponding to the background features and noise features. Retrieve the association relationships, input the permutations and combinations corresponding to the target features into the association relationships, obtain the item names, and output the item names.

9. An image recognition-based object recognition system for a vehicle-mounted infrared system, the system being used to execute the image recognition-based object recognition method for a vehicle-mounted infrared system as described in claim 1, characterized in that, It includes a construction module, an association module, and an identification module; The construction module models historical infrared images to obtain infrared image models; The association module constructs the association relationship between target features, background features, and noise features based on the infrared image model; The recognition module identifies objects based on association relationships and outputs the recognition results.

Images