Search

CN-122024221-A - Vehicle-mounted infrared system object identification method and system based on image identification

CN122024221ACN 122024221 ACN122024221 ACN 122024221ACN-122024221-A

Abstract

The invention discloses a vehicle-mounted infrared system object recognition method and system based on image recognition, which relate to the technical field of image recognition and comprise the following steps of step S100, modeling historical infrared images, and obtaining an infrared image model, constructing an association relation among the target feature, the background feature and the noise feature according to the infrared image model, and identifying an object and outputting an identification result based on the association relation, in step S200. According to the invention, through modeling and feature association, targets, backgrounds and noise are effectively distinguished, false detection rate is reduced, modeling based on historical data can be adapted to infrared image changes in different time, weather and seasons, noise features are clearly separated, stability of a system in a complex environment is improved, continuous updating along with data accumulation is supported, incremental learning of new target types is supported, infrared imaging does not depend on visible light, and reliable identification under severe conditions such as night, foggy days and strong light is realized.

Inventors

  • YANG SHIYONG
  • LI FANGLING

Assignees

  • 东莞市科谱达光电科技有限公司

Dates

Publication Date
20260512
Application Date
20260403

Claims (9)

  1. 1. The vehicle-mounted infrared system object identification method based on image identification is characterized by comprising the following steps of modeling a historical infrared image to obtain an infrared image model, wherein the step S100 is performed on the historical infrared image; step S200, constructing association relations among target features, background features and noise features according to the infrared image model; Step S300, identifying the object based on the association relation and outputting an identification result; the expression of the infrared image model is that, ; Wherein, the Represented as the kth historical infrared image, (i, j) is the pixel coordinate location in the historical infrared image, Represented as the intensity of the infrared feedback signal at the (i, j) position of the pixel area after the first mark in the kth historical infrared image, Represented as the (i, j) position of the background portion in the kth historical infrared image, Expressed as the infrared feedback signal strength belonging to the noise portion (i, j) position in the kth historical infrared image.
  2. 2. The method for recognizing an object in a vehicle-mounted infrared system based on image recognition according to claim 1, wherein the step S100 comprises the following sub-steps, step S101, obtaining a vehicle model, and calling a history infrared image corresponding to the vehicle model; Step S102, preprocessing is carried out on a history infrared image, wherein the preprocessing comprises non-local mean filtering noise reduction processing and CLAHE contrast enhancement processing; step S103, selecting any preprocessed historical infrared image, extracting shape feature quantity of the historical infrared image, and recording the shape feature quantity as a first feature quantity; Step S104, calling an infrared recognition object standard image library, selecting any standard image in the infrared recognition object standard image library, extracting the shape characteristic quantity of the standard image, and marking the shape characteristic quantity as a second characteristic quantity; Step S105, calculating the similarity between the first characteristic quantity and the second characteristic quantity through a cosine similarity formula, setting a first value as a similarity threshold value, and comparing the similarity with the first value; When the similarity is greater than or equal to the first value, step S106 is skipped; When the similarity is smaller than the first value, step S104 is skipped, wherein the selected standard image does not include the already selected standard image; Step S106, setting the object name corresponding to the standard image as the object name corresponding to the historical infrared image, carrying out first marking on the pixel part of the object name in the historical infrared image, and setting the object name as the label of the pixel area after the first marking; step S107, setting pixels of the history infrared image which are not marked as a background part, calculating an average gray value of the background part, setting a second value as a gray error value, and identifying a noise part according to the average gray value and the second value; and S108, constructing an infrared image model according to the background part, the noise part and the first marked pixel area.
  3. 3. The method for recognizing an object in a vehicle-mounted infrared system based on image recognition according to claim 2, wherein step S107 further comprises the substeps of step S1071, obtaining an average gray value of a background portion of any one of the history infrared images; Step S1072, a gray value lookup table is called, the average gray value is input into the gray value lookup table, and a temperature value corresponding to the average gray value is obtained; step S1073, calculating a second value according to the temperature value and the basic temperature fluctuation noise formula, and setting the second value as a gray level error value; step S1074, calculating a first sum of the average gray value and the second value, calculating a first difference value of the average gray value and the second value, and constructing a gray scale interval by taking the first sum as the upper limit of the gray scale interval and the first difference value as the lower limit of the gray scale interval; step S1075, respectively comparing the gray value of each pixel of the background part of the history infrared image with the gray interval; When the gray value of the pixel is distributed in the gray interval, skipping to the next pixel, and repeating the comparison process of the gray value of the pixel and the gray interval, wherein the distribution is represented by that the gray value of the pixel is smaller than or equal to the upper limit of the gray interval and larger than or equal to the lower limit of the gray interval; when the gray value of a pixel is not distributed in the gray interval, the pixel is set as a noise part, wherein the non-distribution is represented by that the gray value of the pixel is larger than the upper limit of the gray interval or the gray value of the pixel is smaller than the lower limit of the gray interval.
  4. 4. The method for recognizing an object in a vehicle-mounted infrared system based on image recognition according to claim 2, wherein the background portion is automatically updated to a background portion remaining after the noise portion is removed before the infrared image model is constructed.
  5. 5. The method for recognizing an object in a vehicle-mounted infrared system based on image recognition according to claim 1, wherein a new history infrared image is called, and a background portion, a noise portion and a first marked pixel region in the new history infrared image are extracted according to an infrared image model; Respectively extracting the statistical characteristics of the background part, the statistical characteristics of the noise part and the statistical characteristics of the first marked pixel region in the new historical infrared image, and respectively setting the statistical characteristics of the background part, the statistical characteristics of the noise part and the statistical characteristics of the first marked pixel region as background characteristics, noise characteristics and target characteristics; the statistical characteristics comprise gray variance and gray standard deviation; Setting a fourth value and a fifth value as variance boundary values, setting a sixth value and a seventh value as standard deviation boundary values, and classifying each statistical characteristic according to the variance boundary values to obtain a first class variance, a second class variance, a third class variance, a first class standard deviation, a second class standard deviation and a third class standard deviation; Constructing a mode according to the arrangement and combination method and the classified statistical characteristics; under each mode, counting the probability of linkage occurrence of the statistical characteristics of the background part, the statistical characteristics of the noise part and the statistical characteristics of the pixel area after the first mark, and marking the probability as a first probability; setting the eighth value as a probability threshold, and constructing the association relation among the target feature, the background feature and the noise feature according to the probability threshold and the first probability.
  6. 6. The method for recognizing an object in a vehicle-mounted infrared system based on image recognition according to claim 5, wherein the method for classifying each statistical characteristic according to the variance demarcation value comprises obtaining a target feature, a background feature, a noise feature, a variance demarcation value and a standard deviation demarcation value; comparing the gray variance in the target feature with the variance demarcation value, and comparing the gray standard deviation in the target feature with the standard deviation demarcation value; comparing the gray variance in the background feature with the variance demarcation value, and comparing the gray standard deviation in the background feature with the standard deviation demarcation value; Comparing the gray variance in the noise characteristic with a variance demarcation value, and comparing the gray standard deviation in the noise characteristic with a standard deviation demarcation value; the comparison process of the gray variance and the gray standard deviation corresponding to the target feature, the background feature and the noise feature is logically the same, and the comparison process is expressed as a feature comparison process.
  7. 7. The image recognition-based vehicle-mounted infrared system object recognition method according to claim 6, wherein the feature comparison process includes classifying the gray variance as a first type variance when the gray variance is equal to or smaller than a fourth value; When the gray variance is larger than the fourth value and smaller than or equal to the fifth value, classifying the gray variance as a second class variance; Classifying the gray variance as a third class variance when the gray variance is greater than the fifth value; When the gray standard deviation is smaller than or equal to the sixth value, classifying the gray variance as a first type standard deviation; when the gray standard deviation is larger than the sixth value and smaller than or equal to the seventh value, classifying the gray variance as a second class standard deviation; classifying the gray scale variance as a third class of standard deviation when the gray scale standard deviation is greater than the seventh value; According to the permutation and combination method and the classified statistical characteristics, the method for constructing the mode comprises the following steps that the permutation and combination corresponding to the background characteristics, the permutation and combination corresponding to the noise characteristics and the permutation and combination corresponding to the target characteristics are respectively expressed as A w 、B u and C h ; Wherein w, u and h represent a w-th permutation and combination corresponding to the background feature, a u-th permutation and combination corresponding to the noise feature and a h-th permutation and combination corresponding to the target feature; The arrangement and combination are firstly combined according to the ascending order of the numbers of the variance classification, and then the ascending order of the numbers of the standard deviation classification; The modality is denoted as a w B u C h ; the first probability calculating method comprises the steps of calculating the number of new historical infrared images and recording the number as a first number; acquiring any mode, counting the number of new historical infrared images conforming to the mode, and recording the number as a second number; calculating the ratio of the second quantity to the first quantity, and recording the ratio of the second quantity to the first quantity as a first probability; Traversing each mode to obtain a first probability corresponding to each mode.
  8. 8. The method for recognizing the vehicle-mounted infrared system object based on the image recognition according to claim 1, wherein the method for constructing the association relation comprises the steps of obtaining a first probability and a probability threshold, comparing the first probability with the probability threshold, and skipping to the next mode when the first probability is smaller than or equal to the probability threshold; when the first probability is greater than the probability threshold, marking the modality as a high confidence modality; when all the first probabilities are traversed and no high confidence modes exist, the probability threshold is adjusted down by 0.05 and then is compared again until at least one high confidence mode is screened out, and the probability threshold is stopped being adjusted down; acquiring each high confidence coefficient mode and a corresponding first probability thereof; When two or more permutation and combination are the same in the high confidence modes, namely the corresponding classification of the statistical characteristics represented by A w 、B u and C h is the same, comparing the first probability of the high confidence modes, reserving the high confidence mode with the highest first probability, and deleting the rest high confidence modes with the highest first probability; when two or more permutation and combination are not identical in the high confidence mode, namely the classification of the statistical characteristics represented by the corresponding A w 、B u and C h is not identical, the high confidence mode is reserved; acquiring a reserved high confidence coefficient mode, and constructing a mapping relation between an arrangement combination corresponding to the target feature, an arrangement combination corresponding to the background feature and an arrangement combination corresponding to the noise feature; Constructing an association relationship between the permutation and combination corresponding to the target features and the names of the articles; acquiring an infrared image to be identified, and extracting a background part and a noise part according to an infrared image model; Extracting the statistical characteristics of the background part and the statistical characteristics of the noise part, obtaining an arrangement combination corresponding to the background characteristic and an arrangement combination corresponding to the noise characteristic according to the statistical characteristics of the background part and the statistical characteristics of the noise part, calling a mapping relation, and inputting the arrangement combination corresponding to the background characteristic and the arrangement combination corresponding to the noise characteristic into the mapping relation to obtain an arrangement combination corresponding to the target characteristic corresponding to the arrangement combination corresponding to the background characteristic and the arrangement combination corresponding to the noise characteristic; And calling the association relation, inputting the permutation and combination corresponding to the target feature into the association relation to obtain the article name, and outputting the article name.
  9. 9. An image recognition-based vehicle-mounted infrared system object recognition system for executing the image recognition-based vehicle-mounted infrared system object recognition method according to claim 1, characterized by comprising a construction module, a correlation module and a recognition module; the construction module models the historical infrared image to obtain an infrared image model; The association module constructs association relation among the target feature, the background feature and the noise feature according to the infrared image model; The recognition module recognizes the object based on the association relation and outputs a recognition result.

Description

Vehicle-mounted infrared system object identification method and system based on image identification Technical Field The invention relates to the technical field of image recognition, in particular to a vehicle-mounted infrared system object recognition method and system based on image recognition. Background In recent years, the object recognition technology of a vehicle-mounted infrared system is rapidly developed to achieve an intelligent sensing system, a deep learning framework such as a YOLO series algorithm and a fast R-CNN is adopted to achieve real-time detection and classification, biological recognition technologies such as gait analysis, body temperature distribution, profile characteristics and the like are fused, the misjudgment rate is reduced through a continuous learning optimization algorithm, infrared sensors and other sensors are fused, the problem of complex environment sensing failure is solved, the recognition precision is improved by means of an AI algorithm, and the reliability of environment sensing is enhanced by means of multi-sensor fusion. At present, in the Chinese patent of the invention with the publication number of CN119821153A, a method and a system for controlling the running early warning and braking of a vehicle based on visible light and infrared are disclosed, bimodal data are collected through a visible light camera and an infrared thermal imager, the bimodal data are preprocessed, the preprocessed bimodal data are subjected to feature fusion, a YOLOV model is input for training, a Yolo-DMFF-detect model is obtained, a Yolo DMFF DETECT model is used for detecting a target object and measuring the distance, an emergency braking module is controlled to send out voice early warning and emergency braking according to the detected target object and the distance, the object recognition and the vehicle early warning braking function are fused through fusion recognition and ranging technology, and the function connection of the vehicle control from the target is realized, but the object recognition process is more complex by taking visible light and auxiliary features as feature references in the related technology, the adaptability to different scenes is not facilitated, the key recognition area of the object is not screened and searched according to the related features of the vehicle-mounted infrared image, and the quick response and the limitation to the object recognition is overcome. Disclosure of Invention The method solves the technical problems that in the related art, the visible light and the auxiliary features are used as feature references, so that the object identification process is more complex, the calibration difficulty is greatly improved, the adaptability to different scenes is not facilitated, the key identification area of the object is not screened and searched according to the related features of the vehicle-mounted infrared image, the quick response and the simplicity of the object identification are not facilitated, and certain limitations exist. In order to solve the technical problems, the invention provides a technical scheme that the vehicle-mounted infrared system object identification method based on image identification in the first aspect comprises the following steps of modeling a historical infrared image to obtain an infrared image model in step S100; step S200, constructing association relations among target features, background features and noise features according to the infrared image model; Step S300, identifying the object based on the association relation and outputting an identification result; the expression of the infrared image model is that, ; Wherein, the Represented as the kth historical infrared image, (i, j) is the pixel coordinate location in the historical infrared image,Represented as the intensity of the infrared feedback signal at the (i, j) position of the pixel area after the first mark in the kth historical infrared image,Represented as the (i, j) position of the background portion in the kth historical infrared image,Expressed as the infrared feedback signal strength belonging to the noise portion (i, j) position in the kth historical infrared image. As the preferable scheme of the vehicle-mounted infrared system object identification method based on image identification, the step S100 comprises the following substeps that step S101, a vehicle model is obtained, and a historical infrared image corresponding to the vehicle model is called; Step S102, preprocessing is carried out on a history infrared image, wherein the preprocessing comprises non-local mean filtering noise reduction processing and CLAHE contrast enhancement processing; step S103, selecting any preprocessed historical infrared image, extracting shape feature quantity of the historical infrared image, and recording the shape feature quantity as a first feature quantity; Step S104, calling an infrared recognition object standard image library, selec