An intelligent recognition method for road-air collaborative traffic scenes

By calculating the consistency of the movement trajectory of the target object in the traffic scene and the grayscale of pixel points in the outline reference neighborhood, the significance is optimized and the accurate significant map is generated, and the inaccurate identification problem caused by insufficient grayscale value in traditional methods is solved. Improved safety risk assessment for traffic scene identification.

CN118366108BActive Publication Date: 2024-11-01嘉兴南湖区路空协同立体交通产业研究院
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410549113.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-06
Publication Date
2024-11-01
Estimated Expiration
2044-05-06

AI Technical Summary

Technical Problem

In traffic scene recognition, traditional significance detection methods rely on grayscale values ​​and cannot effectively resist dust, fog and light pollution, resulting in inaccurate identification results and reducing the accuracy of security risks of traffic scene recognition.

Method used

By obtaining the consistency of the motion trajectory of the target object in the multi-frame image and the grayscale of pixel points in the outline reference neighborhood, the need for significant adjustment of each pixel point is calculated, and the significance is optimized to generate more accurate significant maps for traffic scenes Identify.

Benefits of technology

It improves the accuracy of safety risks in traffic scene identification, highlights the target object, and enhances the accuracy and effectiveness of identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118366108B_ABST
    Figure CN118366108B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of image processing technology, and in particular to a method for intelligent recognition of road-air coordinated traffic scenes, comprising: obtaining a plurality of target objects in a plurality of frames of grayscale images of traffic scenes to be recognized; obtaining the consistency of the motion trajectory of each target object according to the position change of each target object in all frames of grayscale images of traffic scenes to be recognized; obtaining the necessity of significant adjustment of each pixel point in the reference neighborhood of the contour of each target object according to the consistency of the motion trajectory of each target object and the grayscale of the pixel point in the reference neighborhood of the contour; further obtaining the optimized significance of each pixel point in the reference neighborhood of the contour of each target object; obtaining a significant map according to the optimized significance; and performing traffic scene recognition according to the significant map. The present invention improves the accuracy of traffic scene recognition safety risks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an intelligent recognition method for road-to-air collaborative traffic scenes. Background Art

[0002] Video surveillance systems play an important role in intelligent recognition of road-air collaborative traffic scenes. Through real-time detection, early warning and identification of safety risks at traffic scenes, they help improve the safety management level of traffic areas, reduce the risk of accidents, and promote safety and efficiency in the traffic process. In the process of saliency detection, there may be a large amount of dust and fog suspended in the air at the traffic scene recognition site, as well as light pollution and other influences, which will cause blurred vision and low visibility. This will affect the observation range of monitoring equipment and personnel, and reduce the accuracy and effectiveness of safety monitoring. For this reason, it is necessary to perform saliency detection on video images. However, in the traditional saliency detection process, the saliency size is more expressed by the difference between grayscale values. However, in the scene of traffic scene recognition, due to the influence of dust, fog and light pollution factors, only grayscale values ​​are used to evaluate the saliency size, resulting in inaccurate detection results, which in turn reduces the accuracy of traffic scene recognition safety risks. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a method for intelligently identifying road-to-air coordinated traffic scenes, the method comprising:

[0004] Acquire several target objects in several frames of grayscale images of traffic scenes to be identified;

[0005] According to the position change of each target object in the grayscale image of the traffic scene to be identified in all frames, the consistency of the motion trajectory of each target object is obtained;

[0006] Record any frame of grayscale image of traffic scene to be identified as target grayscale image; obtain the contour reference neighborhood of each target object in the target grayscale image; obtain the significant adjustment necessity of each pixel in the contour reference neighborhood of each target object according to the consistency of motion trajectory of each target object and the grayscale of the pixel in the contour reference neighborhood; obtain the optimized significance of each pixel in the contour reference neighborhood of each target object according to the significant adjustment necessity;

[0007] A saliency map is obtained according to the optimized saliency; and traffic scene recognition is performed according to the saliency map.

[0008] Preferably, the method of obtaining the consistency of the motion trajectory of each target object according to the position change of each target object in the grayscale image of the traffic scene to be identified in all frames includes the following specific methods:

[0009] For any target object, obtain a position change trajectory sequence of the target object and a trajectory length of an element in the position change trajectory sequence;

[0010] Acquire a temporal saliency value sequence of the target object;

[0011] The calculation method for obtaining the consistency of the motion trajectory of the i-th target object is:

[0012]

[0013] In the formula, SY i represents the consistency of the motion trajectory of the i-th target object; S i represents the temporal saliency value sequence of the i-th target object; M i represents the position change trajectory sequence of the i-th target object; I i represents the total number of all elements in the position change trajectory sequence of the i-th target object; l i,j represents the trajectory length of the jth element in the position change trajectory sequence of the i-th target object; dtw(S i ,M i ) indicates S i and M i The DTW distance between them; exp() represents an exponential function with a natural constant as the base.

[0014] Preferably, the obtaining of the position change trajectory sequence of the target object and the trajectory length of the elements in the position change trajectory sequence includes the following specific methods:

[0015] The SIFT algorithm is used to obtain the entire area of ​​the target object in each frame of the grayscale image of the traffic scene to be identified; the coordinate position sequence composed of the centroid position of the entire area of ​​the target object in all frames of the grayscale image of the traffic scene to be identified is recorded as the position change trajectory sequence of the target object; for any element in the position change trajectory sequence recorded as the target object, the Euclidean distance between the element and the next element is used as the trajectory length of the element.

[0016] Preferably, the obtaining of the temporal saliency value sequence of the target object comprises the following specific methods:

[0017] An image saliency detection algorithm is used to obtain the average saliency value of the target object in each frame of the grayscale image of the traffic scene to be identified, and a saliency value sequence composed of the average saliency values ​​of the target object in all frames of the grayscale image of the traffic scene to be identified is recorded as the temporal saliency value sequence of the target object.

[0018] Preferably, the specific method of obtaining the reference neighborhood of the contour of each target object in the target grayscale image includes:

[0019] For the i1th target object in the target grayscale image, all the pixels constituting the i1th target object are recorded as the first pixel; the minimum circumscribed rectangular area of ​​the i1th target object in the target grayscale image is recorded as the reference area of ​​the i1th target object in the target grayscale image; the area composed of all the pixels except the first pixel in the reference area of ​​the i1th target object in the target grayscale image is recorded as the reference neighborhood of the contour of the i1th target object.

[0020] Preferably, the method of obtaining the significant adjustment necessity of each pixel in the reference neighborhood of the contour of each target object according to the consistency of the motion trajectory of each target object and the grayscale of the pixel in the reference neighborhood of the contour includes the following specific methods:

[0021] For the j1th pixel point in the reference neighborhood of the contour of the i1th target object in the target grayscale image, obtain the centroid position vector and motion trajectory vector of the j1th pixel point;

[0022] Obtain a significant adjustment factor of the j1-th pixel point according to the centroid position vector and motion trajectory vector of the j1-th pixel point;

[0023] The calculation method for obtaining the necessity of significant adjustment of the j1-th pixel point in the reference neighborhood of the contour of the i1-th target object is:

[0024]

[0025] Where, XT i1,j1 Indicates the necessity of significant adjustment of the j1th pixel point in the reference neighborhood of the contour of the i1th target object; SY i1 Indicates the consistency of the motion trajectory of the i1th target object; coss i1,j1 represents the significant adjustment factor of the j1th pixel in the reference neighborhood of the i1th target object’s contour; g i1,j1 Represents the gray value of the j1th pixel in the reference neighborhood of the contour of the i1th target object; Represents the mean grayscale value of all pixels in the reference neighborhood of the contour of the i1th target object; || represents the absolute value.

[0026] Preferably, the method of obtaining the centroid position vector and motion trajectory vector of the j1-th pixel point includes:

[0027] The straight line segment between the j1-th pixel point and the centroid position of the i1-th target object in the target grayscale image is recorded as the first straight line; the angle between the first straight line and the horizontal line is recorded as the first direction; the unit vector in the first direction is used as the centroid position vector of the j1-th pixel point;

[0028] The next frame of the target grayscale image of the traffic scene to be identified is recorded as the reference grayscale image; the straight line segment between the centroid position of the i1th target object in the target grayscale image and the centroid position of the i1th target object in the reference grayscale image is recorded as the second straight line; the angle between the second straight line and the horizontal line is recorded as the second direction; the unit vector in the second direction is used as the motion trajectory vector of the i1th target object.

[0029] Preferably, the obtaining of the significant adjustment factor of the j1-th pixel point according to the centroid position vector and the motion trajectory vector of the j1-th pixel point comprises the following specific methods:

[0030] The cosine similarity between the centroid position vector of the j1th pixel and the motion trajectory vector of the i1th target object is used as the significant adjustment factor of the j1th pixel.

[0031] Preferably, the method of obtaining the optimized saliency of each pixel in the reference neighborhood of the contour of each target object according to the necessity of saliency adjustment includes the following specific methods:

[0032] For the j1-th pixel point in the reference neighborhood of the contour of the i1-th target object in the target grayscale image, the image saliency detection algorithm is used to obtain the saliency of the j1-th pixel point; the calculation method for obtaining the optimized saliency of the j1-th pixel point in the reference neighborhood of the contour of the i1-th target object is:

[0033]

[0034] In the formula, α' i1,j1 represents the optimized significance of the j1th pixel in the reference neighborhood of the i1th target object’s contour; α i1,j1 W represents the significance of the j1th pixel in the reference neighborhood of the i1th target object’s contour; i1 Indicates the number of all pixels in the reference neighborhood of the contour of the i1th target object; XT i1,j1 Indicates the necessity of significant adjustment of the j1th pixel point in the reference neighborhood of the i1th target object’s contour; XT i1,k Indicates the necessity of significant adjustment of the k-th pixel in the reference neighborhood of the contour of the i1-th target object.

[0035] Preferably, the obtaining of a saliency map according to the optimized saliency; and performing traffic scene recognition according to the saliency map include the following specific methods:

[0036] All pixels in the reference neighborhood of the contours of all target objects in the target grayscale image are recorded as target pixels of the target grayscale image; the optimized saliency of each target pixel in the target grayscale image is used as the saliency of each target pixel in the target grayscale image and recorded as the optimized target grayscale image; the optimized target grayscale image is input into the image saliency detection algorithm to obtain a saliency map;

[0037] A saliency threshold T1 is preset, and the pixels with saliency values ​​greater than T1 in the saliency map are recorded as focus pixels, and the multiple closed areas formed by all the focus pixels are recorded as traffic scene focus areas.

[0038] The beneficial effects of the technical solution of the present invention are as follows: the present invention obtains the necessity of saliency adjustment of each pixel point in the contour reference neighborhood of each target object according to the consistency of the motion trajectory of each target object and the grayscale of the pixel points in the contour reference neighborhood; obtains the optimized saliency of each pixel point in the contour reference neighborhood of each target object according to the necessity of saliency adjustment; thereby optimizing the saliency according to the relative distance and trajectory change between different frame images of the target object in the grayscale image of the traffic scene to be identified, thereby obtaining a saliency map; obtaining the saliency map according to the optimized saliency, the saliency map can more accurately highlight the target object during traffic scene identification, thereby improving the accuracy of traffic scene identification safety risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 This is a flowchart of the steps of an intelligent identification method for road-air coordinated traffic scene of the present invention;

[0041] Figure 2 This is a feature relationship flow chart of an intelligent identification method for road-to-air collaborative traffic scenes of the present invention. DETAILED DESCRIPTION

[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation method, structure, characteristics and effects of the intelligent recognition method of road-air collaborative traffic scenes proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0043] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0044] The specific scheme of the intelligent identification method of road-to-air collaborative traffic scene provided by the present invention is described in detail below with reference to the accompanying drawings.

[0045] See also Figure 1 , which shows a flowchart of a method for intelligently identifying road-to-air coordinated traffic scenes provided by an embodiment of the present invention, the method comprising the following steps:

[0046] Step S001: Acquire several target objects in several frames of grayscale images of traffic scenes to be identified.

[0047] Specifically, it is necessary to first collect several target objects in several frames of grayscale images of traffic scenes to be identified. The specific process is as follows:

[0048] A plurality of frames of traffic scene images to be identified are obtained through traffic monitoring video images; for any frame of traffic scene images to be identified, the frame of traffic scene images to be identified is subjected to median filtering, denoising and graying operations to obtain the frame of traffic scene graying images to be identified; and then a plurality of frames of traffic scene graying images to be identified are obtained. Among them, the median filtering and graying operations are prior arts, and are not described in detail in this embodiment.

[0049] Furthermore, for any frame of grayscale image of the traffic scene to be identified, the frame of grayscale image of the traffic scene to be identified is input into the trained neural network to obtain several target objects in the frame of grayscale image of the traffic scene to be identified; wherein, the neural network used in this embodiment is Resnet50, and the method for obtaining the data set for training the neural network is: collecting a large number of frames of grayscale images of the traffic scene to be identified, artificially marking the positions of people and vehicles in each frame of grayscale image of the traffic scene to be identified, and recording them as the positions of target objects, that is, the position of the target object in the grayscale image of the traffic scene to be identified is marked as 1, and the non-target object is marked as 0, and this marking result is recorded as the label of each frame of grayscale image of the traffic scene to be identified; collecting a large number of frames of grayscale images of the traffic scene to be identified and their corresponding labels to form a data set; using the data set to train the neural network, and the loss function used in the training process is the cross loss function; wherein, the specific training process is a well-known content of the neural network, and this embodiment will not repeat the specific training process.

[0050] So far, several target objects in several frames of grayscale images of traffic scenes to be identified are obtained through the above method.

[0051] Step S002: Obtain the consistency of the motion trajectory of each target object according to the position change of each target object in the grayscale image of the traffic scene to be identified in all frames.

[0052] It should be noted that the edge contours of the target object may not be clear enough due to the influence of environmental factors, that is, the edge contours of people and vehicle targets can often only represent the approximate area. The consistency of the corresponding changes of the overall target object in the time-series multi-frame images can be used to analyze the stability and consistency of the grayscale features of the pixels inside and outside the target object in the time series and during the changes in the motion trajectory, reflecting the stability of the impact of environmental factors on the edge of the target object, which is convenient for the subsequent adjustment of significance.

[0053] Specifically, for any target object, the SIFT (Scale-invariant feature transform) algorithm is used to obtain the entire area of ​​the target object in each frame of the grayscale image of the traffic scene to be identified; the coordinate position sequence composed of the centroid position of the entire area of ​​the target object in all frames of the grayscale image of the traffic scene to be identified is recorded as the position change trajectory sequence of the target object; for any element in the position change trajectory sequence recorded as the target object, the Euclidean distance between the element and the next element is taken as the trajectory length of the element.

[0054] An image saliency detection algorithm is used to obtain the average saliency value of the target object in each frame of the grayscale image of the traffic scene to be identified, and a saliency value sequence composed of the average saliency values ​​of the target object in all frames of the grayscale image of the traffic scene to be identified is recorded as the temporal saliency value sequence of the target object.

[0055] Among them, the SIFT algorithm and the image saliency detection algorithm are existing technologies, and are not described in detail in this embodiment.

[0056] It should be noted that, since the change in the average saliency value of each target object in the grayscale image of the traffic scene to be identified in the corresponding frame is associated with the change in the elements in the position change trajectory sequence, the stronger the correlation between the change in the position trajectory of the target object and the change in the average saliency value, the more stable the analysis of the pixels within the target object in both multi-scale and time series, that is, the smaller the change in the relative grayscale characteristics of the internal and external pixels of the target object, the higher the credibility of the result after the saliency adjustment.

[0057] Specifically, the consistency of the motion trajectory of each target object is obtained according to the position change of each target object in the grayscale image of the traffic scene to be identified in all frames.

[0058] As an example, the calculation method for obtaining the consistency of the motion trajectory of the i-th target object is:

[0059]

[0060] In the formula, SY i represents the consistency of the motion trajectory of the i-th target object; S i represents the temporal saliency value sequence of the i-th target object; M i represents the position change trajectory sequence of the i-th target object; I i represents the total number of all elements in the position change trajectory sequence of the i-th target object; l i,j represents the trajectory length of the jth element in the position change trajectory sequence of the i-th target object; dtw(S i ,M i ) indicates S i and M i DTW distance between; exp() represents an exponential function with a natural constant as the base. The embodiment adopts the exp(-x) model to present the inverse proportional relationship. x is the input of the model. The implementer can select the inverse proportional function according to the actual situation.

[0061] So far, the consistency of the motion trajectory of each target object is obtained through the above method.

[0062] Step S003: According to the consistency of the motion trajectory of each target object and the grayscale of the pixels in the contour reference neighborhood, the necessity of significant adjustment of each pixel in the contour reference neighborhood of each target object is obtained; and then the optimized significance of each pixel in the contour reference neighborhood of each target object is obtained.

[0063] It should be noted that since the saliency value adjustment of the image saliency detection algorithm is based on the Gaussian distribution matrix, in the scenario of precise intelligent recognition of traffic scenes, the focus on moving target objects is higher. This is because the angle of the camera equipment when collecting road traffic images is relatively fixed, so the movement tendency of the target object is often the area that needs more attention; and since the image saliency detection algorithm needs to use the saliency threshold to extract the focus area during the processing process, under normal circumstances, the focus area is located in the middle of the acquired image, and in the scenario of road-air coordinated traffic condition analysis, the focus area is usually each moving target object, and the closer the relative position relationship of the target object is, the more refined the pixel points at the edge of the region need to be analyzed, and the necessity of saliency adjustment of the regional pixels is constructed to obtain accurate saliency detection results, so as to facilitate the subsequent precise intelligent recognition of traffic scenes.

[0064] Specifically, any frame of grayscale image of a traffic scene to be identified is recorded as a target grayscale image. For the i1th target object in the target grayscale image, all pixels constituting the i1th target object are recorded as the first pixel. The minimum circumscribed rectangular area of ​​the i1th target object in the target grayscale image is recorded as the reference area of ​​the i1th target object in the target grayscale image. The area composed of all pixels except the first pixel in the reference area of ​​the i1th target object in the target grayscale image is recorded as the reference neighborhood of the contour of the i1th target object.

[0065] 1. Obtain the significant adjustment necessity of each pixel point in the reference neighborhood of the contour of each target object.

[0066] Specifically, for the j1th pixel point in the contour reference neighborhood of the i1th target object in the target grayscale image, the straight line segment between the j1th pixel point and the center of mass position of the i1th target object in the target grayscale image is recorded as the first straight line; the angle between the first straight line and the horizontal line is recorded as the first direction; and the unit vector in the first direction is used as the center of mass position vector of the j1th pixel point.

[0067] The next frame of the target grayscale image of the traffic scene to be identified is recorded as the reference grayscale image; the straight line segment between the centroid position of the i1th target object in the target grayscale image and the centroid position of the i1th target object in the reference grayscale image is recorded as the second straight line; the angle between the second straight line and the horizontal line is recorded as the second direction; the unit vector in the second direction is used as the motion trajectory vector of the i1th target object.

[0068] The cosine similarity between the centroid position vector of the j1th pixel and the motion trajectory vector of the i1th target object is used as the significant adjustment factor of the j1th pixel.

[0069] According to the consistency of the motion trajectory of each target object and the grayscale of the pixels in the reference neighborhood of the contour, the necessity of significant adjustment of each pixel in the reference neighborhood of the contour of each target object is obtained.

[0070] As an example, a calculation method for obtaining the significant adjustment necessity of the j1-th pixel point in the reference neighborhood of the contour of the i1-th target object is:

[0071]

[0072] Where, XT i1,j1 Indicates the necessity of significant adjustment of the j1th pixel point in the reference neighborhood of the contour of the i1th target object; SY i1 Indicates the consistency of the motion trajectory of the i1th target object; coss i1,j1represents the significant adjustment factor of the j1th pixel in the reference neighborhood of the i1th target object’s contour; g i1,j1 Represents the gray value of the j1th pixel in the reference neighborhood of the contour of the i1th target object; Represents the mean grayscale value of all pixels in the reference neighborhood of the contour of the i1th target object; || represents the absolute value.

[0073] At this point, the necessity of significant adjustment of each pixel point in the reference neighborhood of the contour of each target object is obtained.

[0074] 2. Obtain the optimized saliency of each pixel in the reference neighborhood of the contour of each target object.

[0075] Specifically, for the j1-th pixel point in the reference neighborhood of the contour of the i1-th target object in the target grayscale image, the image saliency detection algorithm is used to obtain the saliency of the j1-th pixel point; and the optimized saliency of each pixel point in the reference neighborhood of the contour of each target object is obtained according to the necessity of saliency adjustment.

[0076] As an example, the calculation method for obtaining the optimized saliency of the j1-th pixel point in the reference neighborhood of the contour of the i1-th target object is:

[0077]

[0078] In the formula, α' i1,j1 represents the optimized significance of the j1th pixel in the reference neighborhood of the i1th target object’s contour; α i1,j1 W represents the significance of the j1th pixel in the reference neighborhood of the i1th target object’s contour; i1 Indicates the number of all pixels in the reference neighborhood of the contour of the i1th target object; XT i1,j1 Indicates the necessity of significant adjustment of the j1th pixel point in the reference neighborhood of the i1th target object’s contour; XT i1,k Indicates the necessity of significant adjustment of the k-th pixel in the reference neighborhood of the contour of the i1-th target object.

[0079] At this point, the optimized saliency of each pixel in the reference neighborhood of the contour of each target object is obtained through the above method.

[0080] Step S004: obtaining a saliency map according to the optimized saliency; and performing traffic scene recognition according to the saliency map.

[0081] Specifically, all pixels in the reference neighborhood of the contours of all target objects in the target grayscale image are recorded as target pixels of the target grayscale image; the optimized saliency of each target pixel in the target grayscale image is used as the saliency of each target pixel in the target grayscale image and recorded as the optimized target grayscale image; the optimized target grayscale image is input into the image saliency detection algorithm to obtain a saliency map; wherein the saliency map contains multiple pixels, and each pixel corresponds to a saliency value.

[0082] Furthermore, a saliency value threshold T1 is preset, wherein this embodiment is described using T1=0.8 as an example, and this embodiment is not specifically limited, wherein T1 may be determined according to the specific implementation situation; pixel points in the saliency map whose saliency values ​​are greater than T1 are recorded as focus pixels, and multiple closed areas formed by all focus pixels are recorded as traffic scene focus areas.

[0083] See also Figure 2 , which shows a feature relationship flow chart of an intelligent identification method for road-air collaborative traffic scenes.

[0084] At this point, this embodiment is completed.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligent recognition of road-air collaborative traffic scenes, characterized in that: The method comprises the following steps: Acquire several target objects in several frames of grayscale images of traffic scenes to be identified; According to the position change of each target object in the grayscale image of the traffic scene to be identified in all frames, the consistency of the motion trajectory of each target object is obtained; The specific method of obtaining the consistency of the motion trajectory of each target object according to the position change of each target object in the grayscale image of the traffic scene to be identified in all frames is as follows: For any target object, obtain a position change trajectory sequence of the target object and a trajectory length of an element in the position change trajectory sequence; Acquire a temporal saliency value sequence of the target object; Get the The calculation method for the consistency of the motion trajectory of the target object is: ; In the formula, Indicates The consistency of the motion trajectory of the target object; Indicates A temporal saliency value sequence of target objects; Indicates A sequence of position change trajectories of target objects; Indicates The total number of all elements in the position change trajectory sequence of the target object; Indicates The first The length of the trajectory of the element; express and The DTW distance between them; represents an exponential function with a natural constant as base; Record any frame of the grayscale image of the traffic scene to be identified as the target grayscale image; obtain the reference neighborhood of the contour of each target object in the target grayscale image; The method of obtaining the significant adjustment necessity of each pixel in the reference neighborhood of the contour of each target object according to the consistency of the motion trajectory of each target object and the grayscale of the pixel in the reference neighborhood of the contour includes: For the target grayscale image The contour of each target object is referenced in the neighborhood pixels, get the The centroid position vector and motion trajectory vector of each pixel; According to The centroid position vector and motion trajectory vector of the pixel point are obtained. Significant adjustment factors for each pixel; Get the The contour of the target object is referenced by the first The calculation method for the necessity of significant adjustment of pixels is: ; In the formula, Indicates The contour of the target object is referenced by the first The need for significant adjustments to pixels; Indicates The consistency of the motion trajectory of the target object; Indicates The contour of the target object is referenced by the first The significant adjustment factor of each pixel; Indicates The contour of the target object is referenced by the first The gray value of each pixel; Indicates The mean of the grayscale values ​​of all pixels in the reference neighborhood of the target object’s contour; Indicates taking the absolute value; According to the consistency of the motion trajectory of each target object and the grayscale of the pixel points in the contour reference neighborhood, the necessity of significant adjustment of each pixel point in the contour reference neighborhood of each target object is obtained; according to the necessity of significant adjustment, the optimized significance of each pixel point in the contour reference neighborhood of each target object is obtained; The specific method of obtaining the optimized saliency of each pixel point in the reference neighborhood of the contour of each target object according to the necessity of saliency adjustment is as follows: For the target grayscale image The contour of the target object is referenced by the pixels, and use the image saliency detection algorithm to obtain the The saliency of the pixel; get the The contour of the target object is referenced by the first The calculation method of the optimized saliency of each pixel is: ; In the formula, Indicates The contour of the target object is referenced by the first The optimized saliency of pixels; Indicates The contour of the target object is referenced by the first The saliency of each pixel; Indicates The number of all pixels in the reference neighborhood of the target object’s contour; Indicates The contour of the target object is referenced by the first The need for significant adjustments to pixels; The table indicates The contour of the target object is referenced by the first The need for significant adjustments to pixels; A saliency map is obtained according to the optimized saliency; and traffic scene recognition is performed according to the saliency map.

2. According to claim 1, a method for intelligently identifying road-air coordinated traffic scenes is characterized in that: The specific method of obtaining the position change trajectory sequence of the target object and the trajectory length of the elements in the position change trajectory sequence includes: The SIFT algorithm is used to obtain the entire area of ​​the target object in each frame of the grayscale image of the traffic scene to be identified; the coordinate position sequence composed of the centroid position of the entire area of ​​the target object in all frames of the grayscale image of the traffic scene to be identified is recorded as the position change trajectory sequence of the target object; for any element in the position change trajectory sequence recorded as the target object, the Euclidean distance between the element and the next element is used as the trajectory length of the element.

3. According to claim 1, a method for intelligently identifying road-air coordinated traffic scenes is characterized in that: The specific method of obtaining the temporal saliency value sequence of the target object includes: An image saliency detection algorithm is used to obtain the average saliency value of the target object in each frame of the grayscale image of the traffic scene to be identified, and a saliency value sequence composed of the average saliency values ​​of the target object in all frames of the grayscale image of the traffic scene to be identified is recorded as the temporal saliency value sequence of the target object.

4. According to claim 1, a method for intelligently identifying road-air coordinated traffic scenes is characterized in that: The specific method of obtaining the reference neighborhood of the contour of each target object in the target grayscale image includes: For the target grayscale image target objects, which will form the All the pixels of the target object are recorded as the first pixel; The minimum bounding rectangle area of ​​the target object in the target grayscale image is recorded as A reference area of ​​a target object in the target grayscale image; The first The area consisting of all pixels except the first pixel in the reference area of ​​the target object in the target grayscale image is recorded as The reference neighborhood of the contours of the target objects.

5. According to claim 1, a method for intelligently identifying road-air coordinated traffic scenes is characterized in that: The acquisition The centroid position vector and motion trajectory vector of each pixel point include the following specific methods: The first Pixels and The straight line segment of the target object at the centroid position in the target grayscale image is recorded as the first straight line; the angle between the first straight line and the horizontal line is recorded as the first direction; the unit vector in the first direction is recorded as the first The centroid position vector of pixels; The next frame of the target grayscale image of the traffic scene to be identified is recorded as the reference grayscale image; The centroid position of the target object in the target grayscale image is The straight line segment of the target object at the centroid position in the reference grayscale image is recorded as the second straight line; the angle between the second straight line and the horizontal line is recorded as the second direction; the unit vector in the second direction is recorded as the second The motion trajectory vector of the target object.

6. According to claim 1, a method for intelligently identifying road-air coordinated traffic scenes is characterized in that: According to the The centroid position vector and motion trajectory vector of the pixel point are obtained. The significant adjustment factors of pixels include: The first The centroid position vector of the pixel is The cosine similarity between the motion trajectory vectors of the target objects is used as the The significant adjustment factor for each pixel.

7. According to claim 1, a method for intelligently identifying road-air coordinated traffic scenes is characterized in that: The method of obtaining a saliency map according to the optimized saliency and identifying a traffic scene according to the saliency map includes the following specific methods: All pixels in the reference neighborhood of the contours of all target objects in the target grayscale image are recorded as target pixels of the target grayscale image; the optimized saliency of each target pixel in the target grayscale image is used as the saliency of each target pixel in the target grayscale image, and recorded as the optimized target grayscale image; Input the optimized target grayscale image into the image saliency detection algorithm to obtain a saliency map; A saliency threshold T1 is preset, and the pixels with saliency values ​​greater than T1 in the saliency map are recorded as focus pixels, and the multiple closed areas formed by all the focus pixels are recorded as traffic scene focus areas.