Multi-sensor fusion method and system based on multi-dimensional attribute correlation analysis

Abstract

The application provides a multi-sensor fusion method and system based on multi-dimensional attribute correlation analysis, and relates to the technical field of multi-sensor data fusion.The application firstly acquires consistency features and stability features of sensor data time series;and acquires correlation features of the sensor data time series based on an improved shape distance algorithm;then calculates the reliability of the sensor data based on the consistency features, the stability features and the correlation features;finally, the multi-sensor data is weighted and fused based on the reliability of each sensor.The application solves the problem that the relative error is large and the precision is low when most of the sensor data deviates, improves the accuracy of multi-sensor data fusion in different scenes, and provides accurate guidance for the driving decision of the autonomous vehicle.

Description

Technical Field

[0001] This invention relates to multi-sensor data fusion technology, specifically to a multi-sensor fusion method and system based on multidimensional attribute correlation analysis. Background Technology

[0002] Multi-sensor data fusion is a crucial method for autonomous vehicles to perceive their environment. Higher accuracy in data fusion leads to more precise descriptions of target locations, which is essential for building environmental models and making accurate driving decisions. In the process of fusing observations of targets from multiple radar sensors to obtain target locations and construct environmental models, multi-radar sensor data fusion can be viewed as a homogeneous time-series data fusion problem, typically solved using weighted fusion methods.

[0003] Currently, the common approach to weighted fusion of multi-sensor data is to define sensor reliability using data consistency and stability characteristics before performing weighted fusion. This method works well when sensors do not exhibit large-area deviations. However, in reality, due to vehicle vibrations or other environmental factors, most sensors may deviate simultaneously. When most sensor data deviates, especially in the same direction, the sensor weights defined solely by data consistency and stability cannot accurately reflect sensor reliability. This leads to relatively high data fusion errors and low accuracy.

[0004] Therefore, there is an urgent need to propose a new multi-sensor data fusion technology to overcome the problem that the data fusion has a relatively large error and low accuracy when most sensor data are biased. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a multi-sensor fusion method and system based on multidimensional attribute correlation analysis, which solves the problem of large relative error and low accuracy in existing multi-sensor data fusion technologies when most sensor data show deviations.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0009] In a first aspect, this invention proposes a multi-sensor fusion method based on multidimensional attribute correlation analysis, the method comprising:

[0010] To obtain the consistency and stability characteristics of sensor data time series;

[0011] The distance between sensor data time series is obtained based on the slope of the change in sensor data time series and the standard Euclidean distance, and the correlation characteristics of sensor data time series are obtained based on the distance between sensor data time series.

[0012] The reliability of sensor data is calculated based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor.

[0013] The multi-sensor data is weighted and fused based on the reliability of each sensor.

[0014] Preferably, the consistency and stability characteristics of the acquired sensor data include:

[0015] Consistency characteristics of sensor data are obtained based on the support method; stability characteristics of sensor data are obtained based on the support variance.

[0016] Preferably, the step of obtaining the distance between sensor data time series based on the slope of the change in the sensor data time series and the standard Euclidean distance, and obtaining the correlation characteristics of the sensor data time series based on the distance between the sensor data time series, includes:

[0017] S21. Standardize the time series data of the sensor;

[0018] S22. Obtain the standard Euclidean distance between sensor data time series based on the standardized sensor data time series;

[0019] S23. Obtain the difference between adjacent time points in the standardized sensor data time series, obtain the slope of the sensor data time series change based on the difference, and determine the change state pattern of each data time series based on the amount of change of the slope.

[0020] S24. Based on the changing state pattern and the standard Euclidean distance, obtain the correlation characteristics between any different attribute parameters in the sensor data.

[0021] Preferably, the method further includes:

[0022] When performing weighted fusion of multi-sensor data based on the reliability of each sensor, adjustment factors and contribution factors are set and adjusted. The adjustment factors include support adjustment factors, stability adjustment factors, and correlation adjustment factors. The contribution factors include support contribution factors, stability contribution factors, and correlation contribution factors.

[0023] Preferably, the method further includes:

[0024] A multi-sensor dataset is acquired and preprocessed to obtain a complete sensor data time series.

[0025] Secondly, the present invention also proposes a multi-sensor fusion system based on multidimensional attribute correlation analysis, the system comprising:

[0026] The basic feature acquisition module acquires the consistency and stability features of the time series of sensor data;

[0027] The correlation feature acquisition module obtains the distance between sensor data time series based on the slope of the change in sensor data time series and the standard Euclidean distance, and obtains the correlation features of sensor data time series based on the distance between sensor data time series.

[0028] The reliability acquisition module calculates the reliability of sensor data based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor.

[0029] The data weighted fusion module performs weighted fusion of multi-sensor data based on the reliability of each sensor.

[0030] Preferably, the consistency and stability features of the sensor data acquired by the basic feature acquisition module include:

[0031] Consistency characteristics of sensor data are obtained based on the support method; stability characteristics of sensor data are obtained based on the support variance.

[0032] Preferably, the correlation feature acquisition module obtains the distance between sensor data time series based on the slope of the change in the sensor data time series and the standard Euclidean distance, and obtains the correlation features of the sensor data time series based on the distance between the sensor data time series, including:

[0033] S21. Standardize the time series data of the sensor;

[0034] S22. Obtain the standard Euclidean distance between sensor data time series based on the standardized sensor data time series;

[0035] S23. Obtain the difference between adjacent time points in the standardized sensor data time series, obtain the slope of the sensor data time series change based on the difference, and determine the change state pattern of each data time series based on the amount of change of the slope.

[0036] S24. Based on the changing state pattern and the standard Euclidean distance, obtain the correlation characteristics between any different attribute parameters in the sensor data.

[0037] Preferably, the system further includes:

[0038] The parameter setting and adjustment module is used to set and adjust adjustment factors and contribution factors when performing weighted fusion of multi-sensor data based on the reliability of each sensor. The adjustment factors include support adjustment factors, stability adjustment factors, and correlation adjustment factors; the contribution factors include support contribution factors, stability contribution factors, and correlation contribution factors.

[0039] Preferably, the system further includes:

[0040] The data acquisition and processing module acquires a multi-sensor dataset and preprocesses the multi-sensor dataset to obtain a complete sensor data time series.

[0041] (III) Beneficial Effects

[0042] This invention provides a multi-sensor fusion method and system based on multidimensional attribute correlation analysis. Compared with existing technologies, it has the following advantages:

[0043] 1. This invention first obtains the consistency and stability characteristics of sensor data time series; then, it obtains the correlation characteristics of sensor data time series based on an improved shape distance algorithm; next, it calculates the reliability of sensor data based on the consistency, stability, and correlation characteristics; finally, it performs weighted fusion of multi-sensor data based on the reliability of each sensor. This invention solves the problem of large relative errors and low accuracy in multi-sensor data fusion when most sensor data exhibit deviations, improves the accuracy of multi-sensor data fusion in different scenarios, and provides precise guidance for driving decisions of autonomous vehicles.

[0044] 2. This invention uses an improved shape distance method to calculate the correlation between observation parameters of the same sensor, which can effectively measure the changing trend of sensor data time series and thus calculate the distance between sensor data time series. Compared with traditional statistical data association analysis methods, it can accurately describe the correlation characteristics of the changing trend between parameters within a certain time window, i.e. when the amount of sensor data is limited. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1This is a flowchart of a multi-sensor fusion method based on multidimensional attribute correlation analysis in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram showing the situation where most sensors shift in the same direction in an embodiment of the present invention;

[0048] Figure 3 a is a schematic diagram comparing the changing trends of x and y coordinate values ​​when there is no abnormal data in an embodiment of the present invention;

[0049] Figure 3 b is a schematic diagram comparing the changing trends of x-coordinate and y-coordinate values ​​when there is abnormal data in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] This application provides a multi-sensor fusion method and system based on multidimensional attribute correlation analysis, which solves the problem of large relative error and low accuracy in existing multi-sensor data fusion technologies when most sensor data show deviations.

[0052] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:

[0053] To address the issue of large relative errors and low accuracy in multi-sensor data fusion when most sensor data exhibits bias, this invention adds the correlation characteristics between sensor observation parameters as an indicator. It calculates the reliability of sensor data by combining the consistency and stability characteristics of the sensor data time series, and then performs weighted fusion of multi-sensor data based on the reliability of each sensor. This weighted fusion technique significantly improves the accuracy of multi-sensor data fusion in various scenarios.

[0054] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0055] Example 1:

[0056] Firstly, this invention proposes a multi-sensor fusion method based on multidimensional attribute correlation analysis, the method comprising:

[0057] S1. Obtain the consistency and stability characteristics of the time series of sensor data;

[0058] S2. Obtain the distance between sensor data time series based on the slope of the change in sensor data time series and the standard Euclidean distance, and obtain the correlation characteristics of sensor data time series based on the distance between sensor data time series;

[0059] S3. Calculate the reliability of sensor data based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor.

[0060] S4. Perform weighted fusion of multi-sensor data based on the reliability of each sensor.

[0061] As can be seen, this invention first obtains the consistency and stability characteristics of sensor data time series; then, it obtains the correlation characteristics of sensor data time series based on an improved shape distance algorithm; next, it calculates the reliability of sensor data based on the consistency, stability, and correlation characteristics; finally, it performs weighted fusion of multi-sensor data based on the reliability of each sensor. This invention solves the problem of large relative errors and low accuracy in multi-sensor data fusion when most sensor data exhibit deviations, improves the accuracy of multi-sensor data fusion in different scenarios, and provides precise guidance for driving decisions of autonomous vehicles.

[0062] The following section uses a multi-radar sensor as an example, along with an explanation of the specific steps, to detail the implementation process of an embodiment of the present invention.

[0063] S1. Obtain the consistency and stability characteristics of the time series of sensor data.

[0064] Before obtaining the consistency and stability characteristics of the sequence, it is necessary to acquire a multi-sensor dataset and preprocess it (including handling missing data) to obtain the complete sensor data time series. Specifically,

[0065] This example obtains multi-radar sensor data from Nuscenes, a large-scale autonomous driving dataset established by an autonomous driving company. Nuscenes is a typical multimodal dataset for autonomous driving, containing mainstream vehicle sensor data such as LiDAR, millimeter-wave radar, and image data. Furthermore, during the environmental perception process of radar sensors, due to complex road conditions such as occlusion and the impact on the accuracy of target recognition algorithms, data gaps are easily encountered during information fusion. Directly ignoring missing data will reduce the accuracy of the fusion result, especially when data from highly weighted sensors is missing. Therefore, it is necessary to process the missing data. Since the sensor time series is a fixed-range time series, the mean of the data before and after the missing value can be used to impute the missing value, obtaining a complete sensor data time series.

[0066] After obtaining the complete sensor data time series, consistency and stability characteristics are obtained based on the sensor data time series.

[0067] 1) Obtain consistency characteristics of sensor data time series based on support method.

[0068] Data consistency is an indicator used to express how closely a sensor's data is similar to that of all other sensor data. High consistency indicates high reliability, while low consistency indicates low reliability. Methods such as belief entropy, information entropy, probability distribution, preference relationships, and support can be used to describe sensor data consistency. To make the calculation results of data consistency more accurate, this invention prioritizes the use of the support method, defined by an exponential decay function, to describe data consistency. Support not only describes the differences between data points but also amplifies the impact of small differences on the fusion results, making it more suitable for practical applications. Extending this support from the differences between two sensor data points to the differences between a single sensor and multiple other sensors effectively expresses the differences between a particular sensor and other sensor data at the same time, thus allowing the weight of that sensor to be obtained during data fusion based on the support.

[0069] When using support to characterize the consistency of sensor data, support is defined exponentially as the distance between data points, as follows:

[0070]

[0071] Where, r A,B This refers to the support between A and B, where μ is the support adjustment factor.

[0072] The support definition indicates that when the distance between data points is greater than a certain preset value, the support of the data is considered to be very low.

[0073] Assume sensor S i and sensor S j The observations of the same object at time t are respectively... and In the absence of observation errors and The values ​​should be equal. If there is observation error, if and If the values ​​differ little, it indicates a high degree of mutual support between the two sensors when observing the x-coordinate; if and If the values ​​of the two sensors differ significantly, it indicates low mutual support when observing the x-coordinate. According to the definition of mutual support, at time t, sensor S... i and sensor S j The mutual support of the x-coordinate observations at time t It is expressed as follows:

[0074]

[0075] Where μ is the support adjustment factor.

[0076] Similarly, at time t, the sensor S i and sensor S j The mutual support of the observations about the y-coordinate at time t It is expressed as follows:

[0077]

[0078] Suppose there are n sensors {S1, S2, ..., S...} n} Observations of the same object are performed separately, and the results of the observations at time t are respectively and If sensor S i If the cross-support between the observations of sensor S and other sensor observations is small, it indicates that sensor S... i The value of sensor S is relatively inaccurate; if sensor S i If the cross-support between the observations of sensor S and other sensor observations is large, it indicates that sensor S... i The value is relatively accurate. Taking the observation of the x-coordinate as an example, the mutual support is extended to the relationship between one sensor and all other sensors, using sensor S at time t. i Support for observation of x-coordinate This indicates the proximity of the sensor's data to data from all other sensors, specifically defined as follows:

[0079]

[0080] It can be seen that, The larger the value, the closer the sensor's value is to the values ​​of all other sensors when observing the x-coordinate at time t, thus the sensor's value is considered more accurate. The smaller the value, the less similar the sensor's value is to all other sensors when observing the x-coordinate at time t, indicating that the sensor's value is relatively inaccurate.

[0081] Similarly, at time t, sensor S i Support for observation of the y-coordinate The definition is as follows:

[0082]

[0083] The larger the value, the closer the sensor's value is to the values ​​of all other sensors when observing the y-coordinate at time t, thus the sensor's value is considered more accurate. The smaller the value, the less similar the sensor's value is to all other sensors when observing the y-coordinate at time t, indicating that the sensor's value is relatively inaccurate.

[0084] 2) Obtain the stability characteristics of sensor data time series based on support variance.

[0085] Data stability characterizes a sensor's performance within a certain time window. For time-series data, the data at the current moment is influenced by data from previous moments. Therefore, a sensor with high reliability in historical moments will also perform better in the current moment; that is, a sensor with higher data stability is more reliable. However, since the observed object in a driving environment is constantly in motion, the stability of the observed data cannot directly reflect the stability of the sensor. Based on this consideration, this invention uses the variance of sensor reliability as the basis for measuring sensor stability. Similar to support, if the variance is greater than a certain value, the reliability of a stability-based sensor can be considered very low; if the correlation variation is greater than a certain value, the reliability of a sensor based on correlation variation can be considered very low.

[0086] Taking the x-coordinate value as an example, assuming that within the time window l, at time t, sensor S i The obtained support sequence is as follows:

[0087]

[0088] Then the average support of this sequence The calculation is as follows:

[0089]

[0090] Considering that the new data has higher reference value, an exponential time decay factor is set, where the time decay factor η at time q is... q for:

[0091] η q =e -λ(t-q)

[0092] Where λ is the attenuation parameter.

[0093] Therefore, within the time window l, for the observation of the x-coordinate at time t, the sensor S i variance Represented as:

[0094]

[0095] Similarly, within the time window l, for the observation of the y-coordinate at time t, sensor S i variance Represented as:

[0096]

[0097] S2. Obtain the distance between sensor data time series based on the slope of the change in sensor data time series and the standard Euclidean distance, and obtain the correlation characteristics of sensor data time series based on the distance between sensor data time series.

[0098] When most sensor data in a sensor system deviates, especially when it deviates in the same direction, simply using data consistency based on support and variance based on support variation cannot describe the reliability of multi-sensor data. This is because when most sensor data deviates, the deviated data may actually have greater data support, and variance based on support cannot reflect changes in the stability of the data. See also Figure 2 As shown in the figure, sensor 1 is the most reliable sensor, while sensors 2 and 3 are offset in the same direction. This results in sensor 1 having the lowest support, the highest variance, and the worst stability, thus exhibiting the lowest reliability, which is inconsistent with reality. Therefore, a new metric is needed to characterize the reliability of multi-sensor data.

[0099] Sensor observations possess multidimensional attributes, and these attributes are correlated. For example, when radar simultaneously observes multiple parameters of the same target, such as the x-axis, y-axis, and z-axis values ​​representing the target's position, as well as its offset angle and velocity, these parameters are not independent but correlated. Considering that the observed object does not undergo sudden changes over a certain time span, when a sensor is reliable, all parameters observed by that sensor should change with the same or similar trends. Therefore, the correlation between parameter sequences remains constant. Figure 3As shown in Figure a, when there are no data anomalies, the relationship between the time series of parameter x and the time series of parameter y remains stable. Based on this, this invention uses the correlation between parameters (i.e., correlation) as one of the standards for measuring sensor reliability. For a specific sensor, the smaller the change in the correlation between parameters, the higher the sensor's reliability. In particular, this invention uses a shape distance-based method to perform more accurate segmentation based on pattern distance, which can effectively analyze the correlation of short sequences. The following example, using the correlation between the coordinate x and coordinate y parameters when a radar sensor observes a target, illustrates the specific process of obtaining the correlation between parameters based on the improved shape distance method.

[0100] Shape distance methods are effective at measuring the changing trends of sequences, thus allowing the calculation of distances between sequences. However, traditional shape distance methods focus on describing relationships between line segments, and have limitations in describing relationships between points. For example... Figure 3 As shown in b, at t=7, the x-axis data experiences an anomalous abrupt change, which can be well described using the shape distance method. However, the data at t=8 is also considered anomalous due to the slope of the abrupt change dropping. Therefore, this technical solution improves the shape distance method by incorporating standard Euclidean distance between auxiliary data points to jointly describe the distance changes between time series of different parameters at each time point. The specific steps are as follows:

[0101] ① To eliminate the influence of data scaling on the results, the data is first standardized. Generally, max-min scaling is used to standardize the data, obtaining the standard parameter time series X' of the x-axis. i The standard parameter time series Y' of the y-axis i ,in:

[0102]

[0103]

[0104] in, and Let x and y represent the standardized x and y coordinates at time t, respectively.

[0105] ② At time t, calculate the distance between time series with different parameters within the time window l. Calculate the difference between data points at adjacent time points, and segment the series according to time to obtain:

[0106]

[0107]

[0108]

[0109]

[0110] in, and These represent the sets of differences between adjacent time points on the x-axis and y-axis, respectively. and These represent the differences between adjacent time points on the x-axis and y-axis, respectively.

[0111] Calculate the slope k of the data change based on the differences between the data points. t This determines the pattern for each data segment.

[0112]

[0113]

[0114] Using the traditional shape distance method, the sequence states are categorized into accelerating descent, horizontal descent, decelerating descent, unchanged, decelerating ascent, horizontal ascent, and accelerating ascent, described by the pattern M = {-3, -2, -1, 0, 1, 2, 3}, with a threshold th used for pattern classification. The threshold th is one of the criteria used to determine whether the sequence state has changed. To prevent the influence of minor jitter in the sequence on the distance analysis, th is generally set to a positive number close to 0. This value can be obtained experimentally, and is typically set to 0.2.

[0115] When k t When <-th, the sequence is in a downward trend. The change in slope is used to classify the downward trend.

[0116] When Δk t When m = 0, the sequence state is horizontally decreasing. t =-2;

[0117] When Δk t When m < 0, the sequence state is in accelerated descent, and m is defined as... t =-3;

[0118] When Δk t When m > 0, the sequence state is decelerating downwards, and m is defined as... t =-1.

[0119] When k t When the value is greater than th, the sequence state is ascending.

[0120] When Δk t When m = 0, the sequence state is horizontally decreasing. t =2;

[0121] When Δk t When < 0, the sequence state is horizontally decreasing, and m is defined as follows:t =1;

[0122] When Δk t When m > 0, the sequence state is horizontally decreasing, and m is defined as... t =3.

[0123] When -th≤k t When m ≤ th, the sequence is considered approximately invariant, and m is defined as follows: t =0. The specific criteria for determining the sequence state are shown in Table 1 below.

[0124] Table 1 Sequence State Determination Table

[0125]

[0126] In addition, to better characterize the distance between sequences, the standard Euclidean distance d between corresponding time points within the time window is calculated. t :

[0127]

[0128] in,

[0129]

[0130]

[0131] and Represents sequence X i and Y i Mean, S x and S y Represents sequence X i and Y i The standard deviation.

[0132] Finally, at time t, sequence X i and Y i Distance D between t for:

[0133]

[0134] Where, m xn The mode value represents the x-coordinate sequence at time n; m yn The mode value represents the y-coordinate sequence at time n; d n It represents the Euclidean standard distance between the x and y sequences at time n.

[0135] ③ Calculate the change in correlation D at time t. t Change in correlation with time t-1, D t-1 The difference

[0136]

[0137] S3. Calculate the reliability of sensor data based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor.

[0138] After obtaining three data characteristics—consistency, stability, and correlation—of multi-sensor data, the reliability of the sensors is calculated based on these three aspects.

[0139] The above analysis shows that the higher the support of a sensor describing data consistency, the closer its data is to that of other sensors, and the higher its data reliability. The smaller the support variance describing sensor stability, the higher the observation stability of the sensor within that time window, and the higher its data reliability. The smaller the difference in the magnitude of the correlation changes between multiple parameters observed by the sensor, the more stable the evolution trend of the correlation between the parameters remains, and the higher its data reliability.

[0140] To ensure that the three indicators have the same calculation scale, the reliability of the sensor can be expressed by the following formula:

[0141]

[0142]

[0143] in, and Representing the sensor S at time t respectively i When observing the x and y coordinates, (α,β) are the adjustment factors of sensor stability and correlation change (i.e., correlation), respectively, and (a,b,c) are the contribution factors of sensor consistency, stability and correlation change to sensor reliability. and Representing the sensor S at time t respectively i Support when observing x and y coordinates; and The values ​​of sensor S and sensor Y represent the x and y coordinates of the target object observed by the sensor, respectively. i The variance; D represents the change in the correlation at time t. t Change in correlation with time t-1, D t-1 The difference.

[0144] To achieve more accurate data fusion, when weighting and fusing multi-sensor data based on the reliability of each sensor, adjustment factors and contribution factors are set and adjusted. The adjustment factors include a support adjustment factor μ, a stability adjustment factor α, and a correlation adjustment factor β; the contribution factors include a support contribution factor a, a stability contribution factor b, and a correlation contribution factor c.

[0145] S4. Perform weighted fusion of multi-sensor data based on the reliability of each sensor.

[0146] The weight of a sensor is positively correlated with its reliability; that is, the higher the reliability of a sensor, the greater the weight of its data during data fusion.

[0147] At time t, sensor S i Weight when observing the x-coordinate The definition is as follows:

[0148]

[0149] Where n represents the number of sensors to be fused. This represents the weight assigned to the k-th sensor when observing the x-coordinate at time t.

[0150] Similarly, at time t, sensor S i Weight when observing the y-coordinate The definition is as follows:

[0151]

[0152] Where n represents the number of sensors to be fused. This represents the weight assigned to the k-th sensor when observing the y-coordinate at time t.

[0153] Therefore, the result of fusing data from n sensors at time t is obtained. and They are represented as follows:

[0154]

[0155]

[0156] This completes the entire process of the multi-sensor fusion method based on multidimensional attribute correlation analysis of this invention.

[0157] To verify the effectiveness of the method of the present invention, the experimental results of the present method (SSR) were compared with those of the mean method (AVG) and the method considering data consistency and stability (SS). The fusion results of x-axis and y-axis values ​​were compared under the following scenarios: all cases (Situation 1), cases with differences in data from most sensors (Situation 2), cases with differences in data from a few sensors (Situation 3), and cases with no significant differences in data from any sensor (Situation 4). The fusion accuracy of different methods under different scenarios was summarized, and the results are shown in Table 2 below.

[0158] Table 2 Comparison of Fusion Results in Different Scenarios

[0159]

[0160] The results above show that the averaging method (AVG) for information fusion does not consider the impact of significant offsets on the fusion results, leading to noticeable deviations. Methods that only consider data consistency and stability show significantly improved accuracy compared to the direct averaging method, but the accuracy is low when most data exhibit significant offsets, especially in the same direction. In contrast, the data fusion method proposed in this invention (SSR) significantly improves accuracy across different scenarios compared to the other two methods, particularly when most sensors exhibit differences, resulting in a marked improvement in fusion results.

[0161] Example 2:

[0162] Secondly, the present invention also provides a multi-sensor fusion system based on multidimensional attribute correlation analysis, the system comprising:

[0163] The basic feature acquisition module acquires the consistency and stability features of the time series of sensor data;

[0164] The correlation feature acquisition module obtains the distance between sensor data time series based on the slope of the change in sensor data time series and the standard Euclidean distance, and obtains the correlation features of sensor data time series based on the distance between sensor data time series.

[0165] The reliability acquisition module calculates the reliability of sensor data based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor.

[0166] The data weighted fusion module performs weighted fusion of multi-sensor data based on the reliability of each sensor.

[0167] Optionally, the basic feature acquisition module acquires the consistency and stability features of the sensor data, including:

[0168] Consistency characteristics of sensor data are obtained based on the support method; stability characteristics of sensor data are obtained based on the support variance.

[0169] Optionally, the correlation feature acquisition module obtains the distance between sensor data time series based on the slope of the change in the sensor data time series and the standard Euclidean distance, and obtains the correlation features of the sensor data time series based on the distance between the sensor data time series, including:

[0170] S21. Standardize the time series data of the sensor;

[0171] S22. Obtain the standard Euclidean distance between sensor data time series based on the standardized sensor data time series;

[0172] S23. Obtain the difference between adjacent time points in the standardized sensor data time series, obtain the slope of the sensor data time series change based on the difference, and determine the change state pattern of each data time series based on the amount of change of the slope.

[0173] S24. Based on the changing state pattern and the standard Euclidean distance, obtain the correlation characteristics between any different attribute parameters in the sensor data.

[0174] Optionally, the system further includes:

[0175] The parameter setting and adjustment module is used to set and adjust adjustment factors and contribution factors when performing weighted fusion of multi-sensor data based on the reliability of each sensor. The adjustment factors include support adjustment factors, stability adjustment factors, and correlation adjustment factors; the contribution factors include support contribution factors, stability contribution factors, and correlation contribution factors.

[0176] Optionally, the system further includes:

[0177] The data acquisition and processing module acquires a multi-sensor dataset and preprocesses the multi-sensor dataset to obtain a complete sensor data time series.

[0178] It is understood that the multi-sensor fusion system based on multidimensional attribute correlation analysis provided in this embodiment of the invention corresponds to the multi-sensor fusion method based on multidimensional attribute correlation analysis described above. The explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding content in the multi-sensor fusion method based on multidimensional attribute correlation analysis, and will not be repeated here.

[0179] In summary, compared with existing technologies, it has the following beneficial effects:

[0180] 1. This invention first obtains the consistency and stability characteristics of sensor data time series; then, it obtains the correlation characteristics of sensor data time series based on an improved shape distance algorithm; next, it calculates the reliability of sensor data based on the consistency, stability, and correlation characteristics; finally, it performs weighted fusion of multi-sensor data based on the reliability of each sensor. This invention solves the problem of large relative errors and low accuracy in multi-sensor data fusion when most sensor data exhibit deviations, improves the accuracy of multi-sensor data fusion in different scenarios, and provides precise guidance for driving decisions of autonomous vehicles.

[0181] 2. This invention uses an improved shape distance method to calculate the correlation between observation parameters of the same sensor, which can effectively measure the changing trend of sensor data time series and thus calculate the distance between sensor data time series. Compared with traditional statistical data association analysis methods, it can accurately describe the correlation characteristics of the changing trend between parameters within a certain time window, i.e. when the amount of sensor data is limited.

[0182] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0183] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to 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 the present invention.

Claims

1. A multi-sensor fusion method based on multidimensional attribute correlation analysis, characterized in that, The method includes: To obtain the consistency and stability characteristics of sensor data time series; The distance between sensor data time series is obtained based on the slope of the change in sensor data time series and the standard Euclidean distance, and the correlation characteristics of sensor data time series are obtained based on the distance between sensor data time series. The reliability of sensor data is calculated based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor. The multi-sensor data is weighted and fused based on the reliability of each sensor. The sensor time series is a fixed-interval time series; The method of obtaining the distance between sensor data time series based on the slope of the time series changes and the standard Euclidean distance, and obtaining the correlation characteristics of the sensor data time series based on the distance between the sensor data time series, includes: S21. Standardize the time series data of the sensor; S22. Obtain the standard Euclidean distance between sensor data time series based on the standardized sensor data time series; S23. Obtain the difference between adjacent time points in the standardized sensor data time series, obtain the slope of the sensor data time series change based on the difference, and determine the change state pattern of each data time series based on the amount of change of the slope. S24. Based on the changing state pattern and the standard Euclidean distance, obtain the correlation characteristics between any different attribute parameters in the sensor data.

2. The method as described in claim 1, characterized in that, The consistency and stability characteristics of the acquired sensor data include: Consistency characteristics of sensor data are obtained based on the support method; stability characteristics of sensor data are obtained based on the support variance.

3. The method as described in claim 1, characterized in that, The method further includes: When performing weighted fusion of multi-sensor data based on the reliability of each sensor, adjustment factors and contribution factors are set and adjusted. The adjustment factors include a support adjustment factor, a stability adjustment factor, and a correlation adjustment factor. The contribution factors include a support contribution factor, a stability contribution factor, and a correlation contribution factor.

4. The method as described in claim 1, characterized in that, The method further includes: A multi-sensor dataset is acquired and preprocessed to obtain a complete sensor data time series.

5. A multi-sensor fusion system based on multidimensional attribute correlation analysis, characterized in that, The system includes: The basic feature acquisition module acquires the consistency and stability features of the sensor data time series. The correlation feature acquisition module obtains the distance between sensor data time series based on the slope of the change in sensor data time series and the standard Euclidean distance, and obtains the correlation features of sensor data time series based on the distance between sensor data time series. The reliability acquisition module calculates the reliability of sensor data based on the consistency characteristics, stability characteristics, and correlation characteristics of the sensor. The data weighted fusion module performs weighted fusion of multi-sensor data based on the reliability of each sensor. The sensor time series is a fixed-interval time series; The correlation feature acquisition module obtains the distance between sensor data time series based on the slope of the change in the sensor data time series and the standard Euclidean distance, and obtains the correlation features of the sensor data time series based on the distance between the sensor data time series, including: S21. Standardize the time series data of the sensor; S22. Obtain the standard Euclidean distance between sensor data time series based on the standardized sensor data time series; S23. Obtain the difference between adjacent time points in the standardized sensor data time series, obtain the slope of the sensor data time series change based on the difference, and determine the change state pattern of each data time series based on the amount of change of the slope. S24. Based on the changing state pattern and the standard Euclidean distance, obtain the correlation characteristics between any different attribute parameters in the sensor data.

6. The system as described in claim 5, characterized in that, The basic feature acquisition module acquires the consistency and stability features of the sensor data, including: Consistency characteristics of sensor data are obtained based on the support method; stability characteristics of sensor data are obtained based on the support variance.

7. The system as described in claim 5, characterized in that, The system also includes: The parameter setting and adjustment module is used to set and adjust adjustment factors and contribution factors when performing weighted fusion of multi-sensor data based on the reliability of each sensor. The adjustment factors include support adjustment factors, stability adjustment factors, and correlation adjustment factors; the contribution factors include support contribution factors, stability contribution factors, and correlation contribution factors.

8. The system as described in claim 5, characterized in that, The system also includes: The data acquisition and processing module acquires a multi-sensor dataset and preprocesses the multi-sensor dataset to obtain a complete sensor data time series.

Images