Target credibility evaluation method and system based on fuzzy space-time semantic understanding

Abstract

The invention belongs to the technical field of computer vision, and provides a target credibility evaluation method and system based on fuzzy time-space semantic comprehension, and the method comprises the steps: inputting a current frame image and historical data into an environment semantic map construction module, and outputting a plurality of semantic quantitative indexes through the environment semantic map construction module; inputting each semantic quantitative index into a fuzzy reasoning system, and outputting defuzzification environment semantic credibility by the fuzzy reasoning system; and fusing the environmental semantic credibility with the detection confidence and the tracking confidence to obtain the final credibility of the target. According to the invention, false alarm and leak detection in a complex environment can be significantly reduced, and the target can be accurately captured even in complex environments such as shielding and illumination variation.

Description

Technical Field

[0001] This application belongs to the field of computer vision technology, and in particular relates to a target credibility assessment method and system based on fuzzy spatiotemporal semantic understanding. Background Technology

[0002] Object detection is a crucial task in computer vision, playing a key role in many applications. The rise of deep learning technology has brought significant breakthroughs to object detection, especially in the remarkable performance improvements achieved by deep learning-based object detectors in simple real-world environments. However, deep learning-based object detectors still face key technical challenges in complex real-world environments, such as:

[0003] 1. False alarm problem: False alarms are easily generated in areas with complex backgrounds (such as swaying leaves or reflections on water).

[0004] 2. Missed Detection Issues: Real targets are easily missed under conditions such as target occlusion and changes in lighting.

[0005] 3. Poor adaptability: The fixed detection threshold cannot adapt to changes in the characteristics of different environments. Summary of the Invention

[0006] This application provides a target credibility assessment method and system based on fuzzy spatiotemporal semantic understanding to solve at least one of the above-mentioned technical problems in the prior art.

[0007] The first aspect of this application provides a target credibility assessment method based on fuzzy spatiotemporal semantic understanding, including:

[0008] The current frame image and historical data are input into the environmental semantic map construction module, which then outputs multiple semantic quantification indicators.

[0009] Each semantic quantification index is input into the fuzzy inference system, and the fuzzy inference system outputs the defuzzified environmental semantic credibility.

[0010] By fusing environmental semantic confidence with detection confidence and tracking confidence, the final confidence of the target is obtained.

[0011] In some preferred embodiments, the current frame image and historical data are input into the environmental semantic map construction module, and the environmental semantic map construction module outputs multiple semantic quantification indicators, including background complexity.

[0012] The environmental semantic map construction module calculates the background complexity based on the current frame image, specifically including the following:

[0013] Preprocess the current frame image;

[0014] Based on the preprocessed image, texture features and edge features are calculated separately, and gray-level co-occurrence matrix entropy and edge density are obtained accordingly.

[0015] The gray-level co-occurrence matrix entropy and edge density are respectively normalized and then weighted and fused to obtain the background complexity.

[0016] In some preferred embodiments, the current frame image and historical data are input to the environmental semantic map construction module, and the environmental semantic map construction module outputs multiple semantic quantification indicators, including the target motion continuity.

[0017] The environmental semantic map construction module calculates the continuity of target movement based on historical data, specifically including the following:

[0018] Extract all historical trajectory data from historical data;

[0019] Calculate directional consistency and velocity consistency based on historical trajectory data;

[0020] By comprehensively calculating the consistency of direction and velocity, the continuity of motion is obtained.

[0021] In some preferred embodiments, the current frame image and historical data are input into the environmental semantic map construction module, and the environmental semantic map construction module outputs multiple semantic quantification indicators, including a priori heatmap.

[0022] The environmental semantic map construction module calculates a priori heatmap based on historical data, which specifically includes the following:

[0023] Extract historically confirmed target locations from historical data;

[0024] Based on the historically confirmed true target locations, kernel density estimation is performed to obtain the prior probability of the target's appearance;

[0025] The prior probability of the target's occurrence is visualized to obtain a prior heatmap.

[0026] In some preferred embodiments, the current frame image and historical data are input into the environmental semantic map construction module, and the environmental semantic map construction module outputs multiple semantic quantification indicators, including trend consistency.

[0027] The environmental semantic map construction module calculates trend consistency based on the prior heatmap and the current frame image, specifically including the following:

[0028] By combining the prior heatmap and the current frame image, the overall probability distribution of the target at a specific location is calculated;

[0029] Based on the overall probability distribution of the target at a specific location, the short-term and long-term probability distributions of the target at that specific location are obtained, and the KL divergence is calculated based on the short-term and long-term probability distributions.

[0030] The degree of matching between the current environmental state and historical experience is calculated based on KL divergence.

[0031] By using the correlation coefficient method to process the degree of matching between the current environmental state and historical experience, trend consistency is obtained.

[0032] In some preferred embodiments, the step of inputting various semantic quantification indicators into the fuzzy inference system and having the fuzzy inference system output the defuzzified environmental semantic credibility specifically includes the following:

[0033] Each semantic quantification index is fuzzified to obtain the degree of membership of the fuzzy set to which each semantic quantification index belongs.

[0034] The weight of the rule set to which each semantic quantification indicator belongs is obtained based on the degree of membership of the fuzzy set to which it belongs and the preset fuzzy rule base.

[0035] By combining the rule set weights for fuzzy reasoning and defuzzification, the credibility of the defuzzified environmental semantics is obtained.

[0036] In some preferred embodiments, the final confidence of the target is obtained by fusing environmental semantic confidence with detection confidence and tracking confidence, and the calculation formula is as follows:

[0037] ;

[0038] in: The ultimate credibility of the target; : Original detection confidence of the detection model; : Detection model's original tracking confidence; : Environmental semantic credibility; α, β, γ: Adaptive weight coefficients, satisfying α+β+γ=1.

[0039] In some preferred embodiments, the gray-level co-occurrence matrix entropy and edge density are respectively normalized and weighted and fused to obtain the background complexity, the calculation formula of which is:

[0040] C_complexity=w1×Norm(GLCM_entropy)+w2×Norm(Edge_density);

[0041] Where: C_complexity: represents the background complexity;

[0042] w1: is the weight coefficient of the gray-level co-occurrence matrix entropy (GLCM_entropy);

[0043] w2: is the weight coefficient of edge density, and satisfies w1+w2=1;

[0044] Norm(): is the normalization function.

[0045] In some preferred embodiments, the motion continuity is obtained by comprehensively calculating the consistency of direction and velocity, and the calculation formula is as follows:

[0046] ;

[0047] in : Indicates the continuity of motion;

[0048] : is the weighting coefficient of the directional consistency related terms;

[0049] : is the weighting coefficient of the speed consistency related terms, and satisfies ;

[0050] : is the normalization function;

[0051] Standard deviation of the orientation angle;

[0052] : Standard deviation of velocity.

[0053] In some preferred embodiments, the kernel density estimation is performed based on the historically confirmed true target locations to obtain the prior probability of the target's occurrence, and the calculation formula is as follows:

[0054] ;

[0055] in: : Indicates position The prior probability of the target appearing at the location;

[0056] The number of historically confirmed actual target locations;

[0057] K(⋅): Kernel function (e.g., Gaussian kernel function);

[0058] h: Bandwidth parameter;

[0059] : The coordinates of the location where the prior probability is to be estimated, i.e., the coordinates of the point on the heat map where the probability is to be calculated;

[0060] : The true location of the i-th historically confirmed target;

[0061] Position (x,y) and historical position The distance between them.

[0062] As can be seen from the above, the beneficial effects of the embodiments of this application compared with the prior art include at least the following:

[0063] This application embodiment integrates historical data and real-time information (current frame image) through an environmental semantic map construction module, and outputs semantic quantification indicators. That is, it can construct a dynamic semantic map to capture the contextual data of the target and its current environment, overcome the limitation of the prior art in processing target attributes in isolation, and thus not only significantly reduce false alarms and missed detections in complex environments, but also improve detection accuracy through semantic understanding.

[0064] Meanwhile, by implementing a fuzzy inference mechanism on each semantic quantification indicator, adaptive threshold adjustment can be achieved, overcoming the problem of poor environmental adaptability of fixed thresholds.

[0065] In addition, by fusing multi-source information, stability and reliability can be ensured in complex environments such as shading and changes in lighting.

[0066] A second aspect of this application provides a target credibility assessment system based on fuzzy spatiotemporal semantic understanding, comprising:

[0067] The camera input module is used to provide the current frame image and historical data to the semantic map module, and at the same time, to the detection and tracking module.

[0068] The semantic map module is used to input the current frame image and historical data into the environmental semantic map construction module, which then outputs multiple semantic quantification indicators.

[0069] The detection and tracking module is used to perform target detection and tracking based on the current frame image, calculate detection confidence and tracking confidence, and at the same time, pass various semantic quantization indicators to the fuzzy inference system.

[0070] The fuzzy inference system is used to perform fuzzy inference calculations based on various semantic quantification indicators and output the defuzzified environmental semantic credibility.

[0071] The fusion decision module is used to fuse environmental semantic confidence with detection confidence and tracking confidence to obtain the final confidence of the target.

[0072] A third aspect of this application provides a terminal including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the first aspect.

[0073] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0074] The fifth aspect of this application provides a computer program product that, when run on a terminal, causes the terminal to perform the steps of the method described in the first aspect.

[0075] It should be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0077] Figure 1 This is a flowchart illustrating a target credibility assessment method based on fuzzy spatiotemporal semantic understanding provided in one embodiment of this application;

[0078] Figure 2 This is a flowchart illustrating a target credibility assessment method based on fuzzy spatiotemporal semantic understanding, provided in another embodiment of this application.

[0079] Figure 3 yes Figure 2 A schematic diagram illustrating the calculation process of background complexity;

[0080] Figure 4 This is a timing diagram of the target credibility assessment system based on fuzzy spatiotemporal semantic understanding provided in the embodiments of this application;

[0081] Figure 5 This is a structural diagram of a terminal provided in an embodiment of this application. Detailed Implementation

[0082] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0083] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0084] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. 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.

[0085] It should be understood that the term "and / or" as used in this application specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0086] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0087] It should be understood that the sequence number of each step in this embodiment does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application embodiment.

[0088] To illustrate the technical solution described in this application, specific embodiments are provided below.

[0089] Research has shown that the root cause of the technical problems in the background section lies in the fact that existing deep learning-based object detectors treat the "confidence" of an object as an isolated attribute, determined solely by the detection score of the current frame and the correlation between multiple frames. In other words, it is determined only by the detection confidence output by the object detection module and the tracking confidence output by the object tracking module, completely ignoring the following key factors:

[0090] 1. The inherent semantic features of the target's environment;

[0091] 2. Historical behavioral patterns of the environment;

[0092] 3. Information on trends in environmental change.

[0093] Based on this, such as Figure 1 As shown in the embodiments of this application, the target credibility assessment method based on fuzzy spatiotemporal semantic understanding includes the following steps:

[0094] Step 1: Input the current frame image and historical data into the environmental semantic map construction module, which then outputs multiple semantic quantification indicators.

[0095] Step 2: Input each semantic quantification index into the fuzzy inference system, and the fuzzy inference system outputs the defuzzified environmental semantic credibility.

[0096] Step 3: Integrate the environmental semantic confidence with the detection confidence and tracking confidence to obtain the final confidence of the target.

[0097] Obviously, the embodiments of this application integrate historical data and real-time information (current frame image) through the environmental semantic map construction module, and output semantic quantification indicators. That is, it can construct a dynamic semantic map to capture the contextual data of the target and its current environment, overcome the limitation of the prior art in processing target attributes in isolation, and thus not only significantly reduce false alarms and missed detections in complex environments, but also improve detection accuracy through semantic understanding.

[0098] Meanwhile, by implementing a fuzzy inference mechanism on each semantic quantification indicator, adaptive threshold adjustment can be achieved, overcoming the problem of poor environmental adaptability of fixed thresholds.

[0099] In addition, by fusing multi-source information, stability and reliability can be ensured in complex environments such as shading and changes in lighting.

[0100] In some embodiments, such as Figure 2 As shown, in step 1, the current frame image and historical data are input into the environmental semantic map construction module, and the environmental semantic map construction module outputs multiple semantic quantification indicators, which can be background complexity, motion continuity, prior heatmap and trend consistency.

[0101] In the embodiments of this application, such as Figure 3 As shown, the environmental semantic map construction module calculates the background complexity based on the current frame image, which may include the following:

[0102] Step 101: Preprocess the current frame image;

[0103] Step 102: Based on the preprocessed image, perform texture feature calculation and edge feature calculation respectively to obtain the gray-level co-occurrence matrix entropy and edge density.

[0104] Step 103: Perform feature normalization on the gray-level co-occurrence matrix entropy and edge density respectively, and then perform weighted fusion to obtain the background complexity.

[0105] Clearly, this embodiment comprehensively evaluates background complexity by combining texture complexity and edge complexity, making the background complexity quantification more objective, comprehensive, and realistic, thereby accurately reflecting the current environmental state of the target.

[0106] For example, step 101, preprocessing the current frame image, may include the following:

[0107] Step 1011: Divide the current frame image into several uniform grids, for example, each grid is 10×10 pixels;

[0108] Step 1012: Extract RGB image patches for each grid.

[0109] Step 1013: Convert each RGB image block to a grayscale image.

[0110] Clearly, the above image preprocessing facilitates the calculation of texture features and edge features, thereby improving the accuracy of the calculation results.

[0111] For example, in step 102, based on the preprocessed image, texture feature calculation and edge feature calculation are performed respectively to obtain the gray-level co-occurrence matrix entropy and edge density, which may include the following:

[0112] Specifically, based on the preprocessed image, texture features are calculated to obtain the gray-level co-occurrence matrix entropy, which can be:

[0113] Calculate the probability of two pixels with a specific spatial relationship appearing adjacently based on the grayscale image.

[0114] The gray-level co-occurrence matrix entropy is obtained by calculating the probability value and the preset first formula.

[0115] More specifically, the probability values ​​of two pixels with a specific spatial relationship appearing adjacently, calculated based on a grayscale image, can include the following:

[0116] (1) Gray-level quantization

[0117] First, the grayscale image's grayscale range is compressed (quantized) into L levels (for example, if the original image has 0-255 grayscale levels, it can be quantized into 8, 16, or fewer levels) to reduce computational complexity.

[0118] (2) Define “spatial relationship”

[0119] Choose the offset (Δx, Δy) to specify the relative positional relationship between the two pixels (i.e., "who coexists with whom").

[0120] The offset can be:

[0121] Horizontal direction: Δx=1, Δy=0 (right adjacent pixel);

[0122] Vertical direction: Δx=0, Δy=1 (adjacent pixels below);

[0123] Diagonal direction: Δx=1, Δy=1 (bottom right adjacent pixel);

[0124] The offset "distance" is usually set to 1 (to analyze the co-occurrence of adjacent pixels).

[0125] (3) Count the number of occurrences N of gray-level pairs. , )

[0126] Iterate through all pixel pairs in the image that satisfy the "offset (Δx, Δy)":

[0127] For each pixel (x, y), find its offset pixel (x+Δx, y+Δy), and record the grayscale values ​​of these two pixels. (Original pixel) and (Offset pixels).

[0128] Statistics of all such ( , The number of occurrences is denoted as N( , ).

[0129] in:

[0130] N( , ): Indicates that the grayscale value is and The number of times a pair of pixels (satisfying a spatial positional relationship with a specific offset (Δx, Δy)) appears in the image;

[0131] : The grayscale value of the original pixel (position (x,y));

[0132] : The grayscale value of the pixel after offset (at position (x+Δx, y+Δy);

[0133] (x,y): The coordinates of a pixel in the image.

[0134] (4) Normalize to obtain probability values

[0135] Number N ( , Divide by the total number of pixel pairs that satisfy the offset condition, N. total The joint probability value is obtained, and the calculation formula is as follows:

[0136]

[0137] Where: N( , ): Indicates that the grayscale value is and The number of times a pair of pixels (satisfying a spatial positional relationship with a specific offset (Δx, Δy)) appears in the image;

[0138] The total number of pixel pairs that satisfy the offset condition (i.e., conform to the spatial positional relationship specified by (Δx,Δy));

[0139] P( , This indicates that under "Specified Offset (Direction + Distance)", the grayscale value is... and grayscale The probability of adjacent occurrences.

[0140] More specifically, the gray-level co-occurrence matrix entropy is obtained based on the probability value and the preset first formula, whereby the preset first formula is as follows:

[0141] ;

[0142] Where: GLCM_entropy: represents the entropy of the gray-level co-occurrence matrix, which is an indicator used to measure the complexity of image texture. The larger the entropy value, the more complex and irregular the texture usually is.

[0143] : Represents the number of gray levels in an image, that is, the total number of different gray values ​​in the image;

[0144] and : These are the row and column indices in the gray-level co-occurrence matrix, corresponding to the gray values ​​of two pixels in the image that have a specific spatial relationship;

[0145] : Position in the gray-level co-occurrence matrix The probability value at a given location describes the probability of two pixels in an image having a specific spatial relationship, each with a different grayscale value. and The joint probability of "".

[0146] The Gray-Level Co-occurrence Matrix (GLCM) describes the joint distribution of gray levels in an image in terms of spatial relationships: it statistically analyzes the distribution of gray levels of two pixels with a specific spatial relationship. and The probability of "", and P( , That is the joint probability.

[0147] Among them, edge feature calculation is performed based on the preprocessed image to obtain the corresponding edge density, which can be specifically as follows:

[0148] The Canny edge detection algorithm is used to detect the grayscale image and obtain the number of edge pixels.

[0149] The edge density is calculated based on the number of edge pixels and the total number of pixels in the grid, and the calculation formula is as follows:

[0150] Edge_density=N_edges / N_total

[0151] Where: N_edges is the number of edge pixels detected by the Canny edge detection algorithm;

[0152] N_total is the total number of pixels in the grid, which is the total number of pixels in the analyzed area;

[0153] Edge_density: Represents edge density, used to measure the density of edge pixels in an image.

[0154] Clearly, the Canny edge detection algorithm can extract edges in an image quite accurately because it ensures the accuracy of edge density.

[0155] For example, in step 103, the gray-level co-occurrence matrix entropy and edge density are respectively normalized and then weighted and fused to obtain the background complexity, the calculation formula of which is shown below:

[0156] C_complexity=w1×Norm(GLCM_entropy)+w2×Norm(Edge_density);

[0157] Where: C_complexity: represents the background complexity, which is obtained by combining the two features of gray-level co-occurrence matrix entropy and edge density, after weighting and normalization, and is used to comprehensively evaluate the complexity of the image background;

[0158] w1: is the weighting coefficient of the gray-level co-occurrence matrix entropy (GLCM_entropy) in the calculation of overall complexity;

[0159] w2: is the weighting coefficient of edge density in the calculation of overall complexity, and satisfies w1+w2=1. It is used to adjust the contribution of gray-level co-occurrence matrix entropy and edge density to overall complexity.

[0160] Norm() is a normalization function that maps the values ​​of the gray-level co-occurrence matrix entropy (GLCM_entropy) and edge density (Edge_density) to a uniform range so that a weighted summation can be performed to calculate the overall complexity.

[0161] In this embodiment of the application, the environmental semantic map construction module calculates the target motion continuity based on historical data, which may specifically include the following:

[0162] Step 111: Extract all historical trajectory data from historical data;

[0163] Step 112: Calculate directional consistency and velocity consistency based on historical trajectory data.

[0164] Step 113: Based on the preset second formula, perform a comprehensive calculation on the consistency of direction and the consistency of velocity to obtain the motion continuity.

[0165] Clearly, motion continuity reflects the regularity and predictability of target motion within a region. By comprehensively evaluating motion continuity through directional consistency and velocity consistency, the quantitative results of motion continuity become more accurate and reliable.

[0166] For example, in step 111, extract all historical trajectory data from the historical data, where:

[0167] Historical trajectory data can be represented as {T1,T2,...,T...} n};

[0168] Each trajectory can be represented as: T i ={P1,P2,...,P m}, where P j =(x j ,y j ,t j ,v j ,θ j );

[0169] Where: n: is the number of targets, T i : The i-th individual trajectory, P j : No. There are several points in motion, x j The goal is in the first place. The x-coordinate of each point in motion, y j The goal is in the first place. The ordinate of each motion state point, t j The target is at the first At any moment of a motion state point, v j The goal is in the first place. The velocity of each moving point, θ j The goal is in the first place. The direction angle of each moving point;

[0170] For example, in step 112, calculate the directional consistency and velocity consistency based on the historical trajectory data. The directional consistency can be calculated using the following formula:

[0171] ;

[0172] Where: σ θ This represents the "standard deviation of the direction angle" (the subscript θ corresponds to the angle symbol theta).

[0173] std(): is an abbreviation for "standard deviation", used to calculate the dispersion of data;

[0174] {θ1,θ2,…,θ n}: Represents a sequence of direction angles for a set of trajectories (e.g., the direction angles of the trajectory at the 1st, 2nd, ..., nth time step);

[0175] Speed ​​consistency can be calculated using the following formula:

[0176] ;

[0177] Where: σ v : indicates the "standard deviation of velocity" (the subscript v corresponds to "velocity");

[0178] std(): is an abbreviation for "standard deviation", used to quantify the dispersion of data;

[0179] {v1,v2,…,v n}: Represents the velocity sequence of the trajectory at different times (e.g., velocity values ​​from time 1 to time n);

[0180] For example, in step 113, based on the preset second formula, a comprehensive calculation is performed on the consistency of direction and the consistency of velocity to obtain the motion continuity, wherein the preset second formula is as follows:

[0181] ;

[0182] in : Represents motion continuity, used to measure the overall continuity of the target motion. The larger the value, the better the motion continuity.

[0183] : is the weighting coefficient of the directional consistency related terms, used to adjust the contribution ratio of directional consistency in the comprehensive continuity calculation. It is a value between and .

[0184] : This is the weighting coefficient for speed consistency-related terms, used to adjust the contribution ratio of speed consistency in the overall continuity calculation. It also ranges from 0 to 1 and satisfies the following condition: ;

[0185] : is the normalization function, its function is to normalize the standard deviation of the direction angle. and speed standard deviation The values ​​are mapped to a uniform range in order to perform weighted calculations to synthesize continuity;

[0186] The standard deviation of the orientation angle measures the dispersion of a set of trajectory orientation angle data. The smaller the value, the higher the consistency in direction. The larger the value, the greater the contribution to the overall continuity;

[0187] The standard deviation of velocity measures the dispersion of trajectory velocity data. The smaller the value, the higher the speed consistency. The larger the value, the greater its contribution to the continuity of the synthesis.

[0188] In this embodiment of the application, the environmental semantic map construction module calculates a priori heatmap based on historical data, which may specifically include the following:

[0189] Step 121: Extract the historically confirmed true target locations from historical data;

[0190] Step 122: Based on the historically confirmed true target locations, perform kernel density estimation to obtain the prior probability of the target's appearance;

[0191] Step 123: Visualize the prior probability of the target's appearance to obtain a prior heatmap.

[0192] Clearly, by establishing a prior probability distribution of target occurrence based on historical real target locations and visualizing it, a prior heatmap can be obtained. The prior heatmap can intuitively quantify the probability of a target appearing at a certain location.

[0193] For example, in step 121, the historically confirmed true target locations are extracted from historical data, where the historically confirmed true target locations can be represented as: {L1, L2, ..., L...} n Each location contains: coordinates (x, y), timestamp t, and target type c.

[0194] For example, in step 122, based on the historically confirmed true target locations, kernel density estimation is performed to obtain the prior probability of the target's appearance. The calculation formula can be expressed as:

[0195] ;

[0196] in: : Indicates position The prior probability of a target appearing at a location, that is, the probability of a target appearing at that location based on historical data; : The number of historically confirmed true target locations, which is the total number of samples used for kernel density estimation; K(⋅): Kernel function (such as Gaussian kernel function); h: Bandwidth parameter (controls the influence range of the kernel function); : The coordinates of the location where the prior probability is to be estimated, i.e., the coordinates of the point on the heat map where the probability is to be calculated; : The true location of the i-th historically confirmed target, each Including coordinates Information such as timestamps and target types are used, but in kernel density estimation, the coordinate information is mainly used. Position (x,y) and historical position The distance between them.

[0197] In this embodiment of the application, the environmental semantic map construction module calculates trend consistency based on the prior heatmap and the current frame image, which may specifically include the following:

[0198] Step 131: Combine the prior heatmap and the current frame image to calculate the overall probability distribution of the target at a specific location;

[0199] Step 132: Based on the overall probability distribution of the target at a specific location, obtain the short-term probability distribution and long-term probability distribution of the target at the specific location, and calculate the KL divergence based on the short-term probability distribution and long-term probability distribution.

[0200] Step 133: Calculate the degree of matching between the current environmental state and historical experience based on the KL divergence.

[0201] Step 134: The correlation coefficient method is used to process the degree of matching between the current environmental state and historical experience to obtain trend consistency.

[0202] Clearly, by quantifying trend consistency, we can reflect the degree of matching between the current environmental state and historical experience, thereby facilitating the improvement of the accuracy of target detection results.

[0203] For example, in step 131, the overall probability distribution of the target at a specific location is calculated by combining the prior heatmap and the current frame image.

[0204] The current frame image is detected using YOLO (a real-time object detection algorithm system), resulting in several object detection boxes;

[0205] For each target box (x1, y1, x2, y2), increment the value of all pixels (or grid cells) within the box on the prior heatmap by 1 to obtain the updated prior heatmap;

[0206] The target probability distribution is obtained by normalizing the updated prior heatmap, and its calculation formula is shown below:

[0207] ;

[0208] in, The original value at a location in the heatmap is the initial heat value obtained by summing the values ​​of all pixels (or grid cells) within the box.

[0209] After normalization, the position The probability distribution value at a given location indicates the probability that the target appears in the entire detection result. The relative frequency;

[0210] All locations on the heat map The sum of the original values ​​is used to... Normalize it to transform it into a probability distribution.

[0211] For example, in step 132, based on the overall probability distribution of the target at a specific location, the short-term and long-term probability distributions of the target at that specific location are obtained, and the KL divergence is calculated based on the short-term and long-term probability distributions. The calculation formula can be as follows:

[0212] ;

[0213] in: KL divergence measures the difference between the "short-term distribution" and the "long-term distribution" (the smaller the difference, the higher the consistency).

[0214] Map short (i, ): Position (i, The short-term probability distribution of ).

[0215] Map long (i, ): Position (i, The long-term probability distribution of ).

[0216] For all positions (i, Sum the calculation results of ) to obtain the overall difference;

[0217] : is a distribution mapping ( and The row and column indices in the map are used to traverse each position in the map, thereby calculating the KL divergence of the entire distribution;

[0218] For example, in step 133, the degree of matching between the current environmental state and historical experience is calculated based on the KL divergence. The calculation formula can be as follows:

[0219] ;

[0220] in: The trend consistency index measures the degree of matching between the current environmental state and historical experience. The higher the value, the higher the degree of matching and the better the trend consistency.

[0221] The attenuation coefficient is used to adjust the degree of influence of KL divergence on the trend consistency index. It is a parameter that is set manually.

[0222] For example, in step 134, the correlation coefficient method is used to process the degree of matching between the current environmental state and historical experience to obtain trend consistency. The calculation formula can be as follows:

[0223] r = corr(Map_short, Map_long);

[0224] S_consistency=(r+1) / 2 / / Mapped to the interval [0,1];

[0225] Where: corr: represents the Pearson correlation coefficient, a statistical indicator used to measure the degree of linear correlation between two variables;

[0226] Map_short: Represents a short-term distribution mapping, which can be understood as the distribution of a certain state, probability, etc. in the current or short term;

[0227] Map_long: Represents a long-term distribution mapping, that is, the distribution of states, probabilities, etc. in the long history;

[0228] S_consistency: This is a consistency index. It is calculated by (r+1) / 2 and maps the correlation coefficient to the interval [0,1]. It is used to measure the degree of consistency between Map_short and Map_long. The closer the value is to 1, the better the consistency between the two.

[0229] Obviously, the embodiments of this application can improve the accuracy of target detection results by innovatively quantifying background complexity, motion continuity, prior heatmap and trend consistency, and by extracting multi-dimensional features (such as gray-level co-occurrence matrix entropy and edge density) to establish a dynamic environment model.

[0230] In some embodiments, step 2 involves inputting various semantic quantification indicators into a fuzzy inference system, which then outputs the defuzzified environmental semantic credibility. This may specifically include the following:

[0231] Step 21: Fuzzify each semantic quantification index to obtain the degree of membership of the fuzzy set to which each semantic quantification index belongs.

[0232] Step 22: Based on the degree of membership of each semantic quantification index to its fuzzy set and the preset fuzzy rule base, obtain the weight of the rule set to which it belongs;

[0233] Step 23: Combine the rule set weights to perform fuzzy reasoning and defuzzification to obtain the defuzzified environmental semantic credibility.

[0234] Clearly, by fuzzifying each semantic quantification index, applying membership functions, and using fuzzy rule bases, adaptive reasoning under uncertain environments can be achieved. For example, trapezoidal and sigmoid functions can be used to process complex inputs.

[0235] For example, in step 21, each semantic quantization index is fuzzified to obtain the degree of membership of the fuzzy set to which each semantic quantization index belongs;

[0236] Specifically, the background complexity is fuzzified to obtain the membership degree of the fuzzy set to which the background complexity belongs.

[0237] Background complexity fuzzy set: {low, medium, high};

[0238] Background complexity membership function: trapezoidal function, which can be expressed by the following formula;

[0239] μ_bg_low(x)=max(0,min(1,(0.3-x) / 0.2));

[0240] μ_bg_medium(x)=max(0,min(1,(x-0.2) / 0.3,(0.8-x) / 0.3));

[0241] μ_bg_high(x)=max(0,min(1,(x-0.6) / 0.3));

[0242] Where: μ_bg_low(x): represents the membership function of the background complexity to the fuzzy set "low", x is the input variable value of the background complexity, and the result of the function is the degree of membership of the input value x to the "low" fuzzy set, which is in the range of [0,1].

[0243] μ_bg_medium(x): This function represents the membership degree of the background complexity to the fuzzy set "medium". x is the input variable value of the background complexity. It calculates the degree of membership of the input value x to the fuzzy set "medium".

[0244] μ_bg_high(x): This function represents the membership degree of the background complexity to the fuzzy set "high". x is the input variable value of the background complexity. It calculates the degree of membership of the input value x to the fuzzy set "high".

[0245] Specifically, the motion continuity is fuzzified to obtain the membership degree of the fuzzy set to which the motion continuity belongs.

[0246] Fuzzy set of motion continuity: {low, high};

[0247] The membership function for motion continuity is the Sigmoid function, which can be expressed as follows:

[0248] μ_mc_low(x)=1 / (1+exp(-k×(x-0.5)));

[0249] μ_mc_high(x)=1-μ_mc_low(x);

[0250] Where: k: is the parameter of the Sigmoid function, used to adjust the shape of the function; x: is the input variable value of motion continuity; μ_mc_low(x): represents the membership function of motion continuity belonging to the "low" fuzzy set, used to adjust the shape of the function, the function calculates the degree of membership of the input value x to the "low" fuzzy set; μ_mc_high(x): represents the membership function of motion continuity belonging to the "high" fuzzy set, calculated by 1-μ_mc_low(x), that is, the degree of membership of the input value x to the "high" fuzzy set is the complement of the degree of membership of the "low" fuzzy set.

[0251] Specifically, the prior probability is fuzzified to obtain the membership degree of the fuzzy set to which the prior probability belongs.

[0252] The fuzzy set of prior probabilities: {low, medium, high};

[0253] Membership function of prior probability: trigonometric function.

[0254] Specifically, the trend consistency is fuzzified to obtain the membership degree of the fuzzy set to which the trend consistency belongs.

[0255] A fuzzy set of trends with consistency: {low, medium, high};

[0256] Membership function for trend consistency: Gaussian function.

[0257] For example, in step 22, based on the degree of membership of each semantic quantification index to its fuzzy set and the preset fuzzy rule base, the weight of the rule set to which it belongs is obtained. The preset fuzzy rule base includes the following:

[0258] (1) Rule structure: IF [precondition] THEN [conclusion] WITH [weight];

[0259] (2) Core rule set:

[0260] High credibility rules:

[0261] IF background complexity IS low AND motion continuity IS high THEN credibility IS extremely high WITH 0.9;

[0262] IF prior probability IS high AND trend consistency IS high THEN confidence IS high WITH 0.8;

[0263] Low credibility rule:

[0264] IF background complexity IS high AND prior probability IS low THEN confidence IS low WITH 0.7;

[0265] IF trend consistency IS low AND motion continuity IS low THEN reliability IS extremely low WITH 0.6;

[0266] Medium confidence level rule:

[0267] IF background complexity IS AND prior probability IS THEN confidence IS WITH 0.5;

[0268] IF trend consistency IS low THEN confidence IS with 0.4 / / conservative strategy;

[0269] For example, in step 23, fuzzy reasoning and defuzzification are performed by combining the rule set weights to obtain the defuzzified environmental semantic credibility, which may include the following:

[0270] Step 231: Calculate the trigger strength of the fuzzy rule by combining the rule set weights. The calculation formula can be as follows:

[0271] ;

[0272] Where: w j : Trigger strength of the j-th fuzzy rule (0~1, the larger the value, the more the rule is satisfied);

[0273] x1,x2,…,x n Input clear values ​​(such as quantized background complexity Ccomplexity, motion continuity, etc.);

[0274] A 1, A2,…,A n : Fuzzy set of rule premises (such as "low background complexity" and "high motion continuity");

[0275] μ Ai (x i ): Input x i Belongs to fuzzy set A i The rule set weight (0.4-0.9, the larger the value, the higher the matching degree);

[0276] Reason for taking the minimum: The premise is an "AND" relationship, and the trigger strength is determined by the premise that is least satisfied.

[0277] Step 232: Based on the triggering strength of the fuzzy rules, calculate the membership function of the final output fuzzy set after aggregation. The calculation formula can be as follows:

[0278] ;

[0279] in: The final output fuzzy set membership function after aggregation;

[0280] w j The trigger strength of the j-th fuzzy rule;

[0281] B j : The fuzzy set of the conclusions of the j-th rule (e.g., "extremely high credibility", "low credibility", etc.);

[0282] μ Bj (y): The output variable y belongs to the conclusion fuzzy set B. j Membership degree;

[0283] Take the maximum value of the results of all rules and merge them into a unified output fuzzy set.

[0284] Step 232: Based on the centroid method, defuzzify the membership function of the final output fuzzy set after aggregation. The defuzzified environmental semantic credibility can be calculated using the following formula:

[0285] ;

[0286] Where: C semantic : Environmental semantic credibility obtained after deblurring (clarity value, used for subsequent fusion);

[0287] y: The universe of discourse element of the output variable (confidence);

[0288] : Membership function of the aggregated output fuzzy set;

[0289] Integration: Iterate through all possible values ​​of the output variable y and calculate the "centroid" of the fuzzy set as the clear output;

[0290] dy is the differential symbol, which represents a small change in the output variable y.

[0291] Clearly, by using fuzzy reasoning to defuzzify (such as the centroid method) and trend consistency quantification (such as KL divergence), the system can be adjusted online in complex scenarios.

[0292] In some embodiments, step 3, fusing the environmental semantic confidence with the detection confidence and tracking confidence, yields the final confidence of the target, which can be calculated using the following formula:

[0293] ;

[0294] in: : The final credibility of the target (final output result); : The original detection confidence of the detection model (such as the YOLO detection score); : Detection model original tracking confidence, that is: confidence based on multi-frame tracking (the more stable the trajectory, the higher the value); : Environmental semantic credibility; α, β, γ: Adaptive weight coefficients (satisfying α+β+γ=1, allocating the contribution ratio of each credibility).

[0295] Clearly, by combining detection confidence, tracking confidence, and environmental semantic credibility based on a multi-source information fusion mechanism, and making decisions through adaptive weights (α, β, γ), the robustness of the final credibility can be improved, ensuring stability in scenarios such as occlusion and changes in illumination.

[0296] In some embodiments, step 4 is further included, continuously receiving target location information to iteratively update historical data in order to support the dynamic optimization of subsequent semantic indicators.

[0297] Based on the same inventive concept, embodiments of this application also provide a target credibility assessment system based on fuzzy spatiotemporal semantic understanding, such as... Figure 4 As shown, it includes:

[0298] The camera input module is used to provide the current frame image and historical data to the semantic map module, and at the same time, to the detection and tracking module.

[0299] The semantic map module is used to input the current frame image and historical data into the environmental semantic map construction module, which then outputs multiple semantic quantification indicators.

[0300] The detection and tracking module is used to perform target detection and tracking based on the current frame image, calculate detection confidence and tracking confidence, and at the same time, pass various semantic quantization indicators to the fuzzy inference system.

[0301] The fuzzy inference system is used to perform fuzzy inference calculations based on various semantic quantification indicators and output the defuzzified environmental semantic credibility.

[0302] The fusion decision module is used to fuse environmental semantic confidence with detection confidence and tracking confidence to obtain the final confidence of the target.

[0303] Obviously, the embodiments of this application integrate historical data and real-time information (current frame image) through the environmental semantic map construction module, and output semantic quantification indicators. That is, it can construct a dynamic semantic map to capture the contextual data of the target and its current environment, overcome the limitation of the prior art in processing target attributes in isolation, and thus not only significantly reduce false alarms and missed detections in complex environments, but also improve detection accuracy through semantic understanding.

[0304] Meanwhile, by implementing a fuzzy inference mechanism on each semantic quantification indicator, adaptive threshold adjustment can be achieved, overcoming the problem of poor environmental adaptability of fixed thresholds.

[0305] In addition, by fusing multi-source information, stability and reliability can be ensured in complex environments such as shading and changes in lighting.

[0306] In some embodiments, the detection and tracking module also transmits target location information to the camera input module, which updates and iterates historical data based on the target location information to support dynamic optimization of subsequent semantic indicators.

[0307] Figure 5 This is a structural diagram of a terminal provided in an embodiment of this application. As shown in the figure, the terminal 5 of this embodiment includes: at least one processor 50 ( Figure 5 (Only one is shown in the diagram), memory 51, and computer program 52 stored in said memory 51 and executable on said at least one processor 50, wherein said processor 50 executes said computer program 52 to implement the steps in any of the above method embodiments.

[0308] The terminal 5 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The terminal 5 may include, but is not limited to, a processor 50 and a memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of terminal 5 and does not constitute a limitation on terminal 5. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal may also include input / output devices, network access devices, buses, etc.

[0309] The processor 50 can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0310] The memory 51 can be an internal storage unit of the terminal 5, such as a hard disk or memory of the terminal 5. The memory 51 can also be an external storage device of the terminal 5, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc., equipped on the terminal 5. Furthermore, the memory 51 can include both internal storage units and external storage devices of the terminal 5. The memory 51 is used to store the computer program and other programs and data required by the terminal. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0311] In specific implementations, the terminals described in the embodiments of this application include, but are not limited to, other portable devices such as mobile phones, laptop computers, or tablet computers with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads). It should also be understood that in some embodiments, the device is not a portable communication device, but a desktop computer with touch-sensitive surfaces (e.g., touchscreen displays and / or touchpads).

[0312] Furthermore, the terminal supports a variety of applications, such as one or more of the following: drawing applications, presentation applications, word processing applications, website creation applications, disc burning applications, spreadsheet applications, game applications, telephone applications, video conferencing applications, email applications, instant messaging applications, exercise support applications, photo management applications, digital camera applications, digital camcorder applications, web browsing applications, digital music player applications, and / or digital video player applications.

[0313] Various applications that can run on a terminal can use at least one common physical user interface device, such as a touch-sensitive surface. One or more functions of the touch-sensitive surface and the corresponding information displayed on the terminal can be adjusted and / or changed between and / or within applications. In this way, the terminal's common physical architecture (e.g., the touch-sensitive surface) can support various applications with user interfaces that are intuitive and transparent to the user.

[0314] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0315] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0316] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0317] In the embodiments provided in this application, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0318] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0319] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0320] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory, random access memory, electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0321] The methods described in this application can be implemented in whole or in part by a computer program product. When the computer program product is run on a terminal, the terminal executes the steps in the various method embodiments described above.

[0322] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A target credibility evaluation method based on fuzzy space-time semantic understanding, characterized in that, The application relates to a method for constructing an environment semantic map, comprising the following steps: inputting a current frame image and historical data into an environment semantic map construction module, outputting a plurality of semantic quantization indexes by the environment semantic map construction module; inputting each semantic quantization index into a fuzzy inference system, outputting an environment semantic credibility after defuzzification by the fuzzy inference system; fusing the environment semantic credibility with detection confidence and tracking confidence to obtain a final credibility of a target.

2. The target credibility assessment method based on fuzzy spatio-temporal semantic understanding according to claim 1, characterized in that, In the step of inputting the current frame image and the historical data into the environment semantic map construction module, the semantic quantization indexes comprise background complexity. The environment semantic map construction module calculates the background complexity according to the current frame image, and the calculation specifically comprises the following steps: preprocessing the current frame image; respectively performing texture feature calculation and edge feature calculation based on the preprocessed image, and correspondingly obtaining a gray level co-occurrence matrix entropy and an edge density; respectively performing feature normalization processing on the gray level co-occurrence matrix entropy and the edge density, and performing weighted fusion to obtain the background complexity. 3.The target credibility evaluation method based on fuzzy space-time semantic understanding according to claim 1, characterized in that, In the step of inputting the current frame image and the historical data into the environment semantic map construction module, the semantic quantization indexes comprise target motion continuity. The environment semantic map construction module calculates the target motion continuity according to the historical data, and the calculation specifically comprises the following steps: extracting all historical trajectory data from the historical data; respectively calculating direction consistency and speed consistency according to the historical trajectory data; performing comprehensive calculation on the direction consistency and the speed consistency to obtain the motion continuity.

4. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 1, characterized in that, In the step of inputting the current frame image and the historical data into the environment semantic map construction module, the semantic quantization indexes comprise a prior heat map. The environment semantic map construction module calculates the prior heat map according to the historical data, and the calculation specifically comprises the following steps: extracting a historical confirmed real target position from the historical data; performing kernel density estimation according to the historical confirmed real target position to obtain a prior probability of target appearance; performing visual processing on the prior probability of target appearance to obtain the prior heat map.

5. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 1, characterized in that, In the step of inputting the current frame image and the historical data into the environment semantic map construction module, the semantic quantization indexes comprise trend consistency. The environment semantic map construction module calculates the trend consistency according to the prior heat map and the current frame image, and the calculation specifically comprises the following steps: combining the prior heat map and the current frame image to calculate an overall probability distribution of the target at a specific position; obtaining a short-term probability distribution and a long-term probability distribution of the target at the specific position according to the overall probability distribution of the target at the specific position, and calculating a KL divergence according to the short-term probability distribution and the long-term probability distribution; calculating a matching degree of a current environment state and historical experience according to the KL divergence; processing the matching degree of the current environment state and the historical experience by a correlation coefficient method to obtain the trend consistency.

6. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 1, characterized in that, The various semantic quantitative indicators are input into the fuzzy reasoning system, and the environment semantic credibility after defuzzification is output by the fuzzy reasoning system, and the specific contents include the following: The various semantic quantitative indicators are respectively subjected to fuzzy processing, and the membership degrees of the fuzzy sets to which the various semantic quantitative indicators belong are obtained; The membership degrees of the fuzzy sets to which the various semantic quantitative indicators belong are obtained, and the rule set weight is obtained according to the preset fuzzy rule base; The fuzzy reasoning and defuzzification processing are performed in combination with the rule set weight, and the environment semantic credibility after defuzzification is obtained.

7. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 1, characterized in that, The environment semantic credibility is fused with the detection confidence and the tracking confidence, and the final credibility of the target is obtained, and the calculation formula is: ； wherein: : final credibility of the target; : original detection confidence of the detection model; : original tracking confidence of the detection model; : environmental semantic credibility; a, b, g: adaptive weight coefficients, and satisfy a+b+g=1.

8. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 2, characterized in that, The gray level co-occurrence matrix entropy and the edge density are respectively subjected to feature normalization processing, and are fused by weighting, and the background complexity is obtained, and the calculation formula is: C_complexity=w1×Norm(GLCM_entropy)+w2×Norm(Edge_density); Wherein: C_complexity represents the background complexity; w1 is the weight coefficient of the gray level co-occurrence matrix entropy (GLCM_entropy); w2 is the weight coefficient of the edge density (Edge_density), and satisfies w1+w2=1; Norm() is a normalization function.

9. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 3, characterized in that, The direction consistency and the speed consistency are comprehensively calculated, and the motion continuity is obtained, and the calculation formula is: ； wherein : indicates motion continuity; : is the weight coefficient of the direction consistency related term; : is the weight coefficient of the speed consistency related item, and satisfies ; : is a normalization function; : standard deviation of the directional angle; : Standard deviation of speed.

10. The target trustworthiness evaluation method based on fuzzy space-time semantic understanding according to claim 4, characterized in that, The kernel density estimation is performed according to the historical confirmed real target position, and the prior probability of target appearance is obtained, and the calculation formula is: ； wherein: : represents the prior probability of the target appearing at the location ; : number of historically confirmed true target locations; K(·): kernel function; h: bandwidth parameter; : position coordinates of which the prior probability is to be estimated, i.e. coordinates of the point in the heat map for which the probability is to be calculated; : ith historically confirmed true target position; : distance between position (x,y) and historical position .

11. A target credibility assessment system based on fuzzy spatio-temporal semantic understanding, characterized in that, It includes: The camera input module is used to provide the current frame image and historical data to the semantic map module, and at the same time, the current frame image is provided to the detection and tracking module; The semantic map module is used to input the current frame image and historical data into the environment semantic map construction module, and a plurality of semantic quantitative indicators are output by the environment semantic map construction module; The detection and tracking module is used to perform target detection and tracking according to the current frame image, calculate the detection confidence and the tracking confidence, and at the same time, the various semantic quantitative indicators are transmitted to the fuzzy reasoning system; The fuzzy reasoning system is used to develop fuzzy reasoning calculation according to the various semantic quantitative indicators, and output the environment semantic credibility after defuzzification; The fusion decision module is used to fuse the environment semantic credibility with the detection confidence and the tracking confidence, and obtain the final credibility of the target.

Images