Target tracking method and device, equipment and storage medium
By acquiring sensor data, determining the fusion weight and fusing the Gaussian component, the problem of low target tracking accuracy in intelligent driving is solved, and the accuracy of bicycle path planning and driving safety are improved.
Patent Information
- Application Number
- CN202510090552.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-20
- Publication Date
- 2025-06-03
AI Technical Summary
In the field of intelligent driving technology, the existing target tracking methods are low, which affects the accuracy of the bicycle planning path, thereby affecting driving safety and may even lead to safety accidents.
By obtaining the Gaussian components to be fused and their probability data corresponding to several sensors at the current time, the fusion weights are determined, and these weights are used to fuse the Gaussian components to obtain the fusion Gaussian component as the target tracking result.
Improve the accuracy of target tracking, enhance the accuracy of bicycle path planning, and thus improve driving safety.
Smart Images
Figure CN120088757A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of intelligent driving, and particularly to a target tracking method, device, equipment, and storage medium. Background Art
[0002] In the technical field of intelligent driving, it is usually necessary to detect the state data of target obstacles around the host vehicle that affect the driving of the host vehicle, and track the motion state, motion trajectory, etc. of the target obstacles based on the state data of the target obstacles, so as to plan the driving path of the host vehicle.
[0003] However, if the accuracy of target tracking is low, it may affect the accuracy of the path planned by the host vehicle, thereby affecting the driving safety of the host vehicle, and even leading to vehicle safety accidents. Therefore, how to accurately track the target obstacles around the host vehicle has become a technical problem to be solved urgently. Summary of the Invention
[0004] The main technical problem to be solved by the present application is to provide a target tracking method, device, equipment, and computer-readable storage medium, which can improve the accuracy of target tracking.
[0005] To solve the above technical problem, a technical solution adopted by the present application is: to provide a target tracking method, the method includes: obtaining the to-be-fused Gaussian components respectively corresponding to a plurality of sensors at the current moment; obtaining the probability data respectively corresponding to each to-be-fused Gaussian component at the current moment; wherein, the probability data includes at least one of a first probability that the obstacle label is a target obstacle and a second probability that the obstacle label is an interference obstacle; determining the fusion weight respectively corresponding to each to-be-fused Gaussian component at the current moment based on the probability data of each to-be-fused Gaussian component at the current moment; fusing each to-be-fused Gaussian component at the current moment by using each fusion weight to obtain the fused Gaussian component at the current moment; taking the fused Gaussian component at the current moment as the target tracking result at the current moment.
[0006] Optionally, the probability data includes the first probability, and obtaining the probability data respectively corresponding to each to-be-fused Gaussian component at the current moment includes: performing kernel density estimation by using a first sample data set to obtain a first probability density function; wherein, the first sample data set includes a plurality of first sample obstacle detection probabilities with the obstacle labels all being target obstacles; obtaining the obstacle detection probabilities respectively corresponding to each to-be-fused Gaussian component at the current moment; for each to-be-fused Gaussian component at the current moment, determining the first probability of the to-be-fused Gaussian component by using the first probability density function and the obstacle detection probability of the to-be-fused Gaussian component.
[0007] Optionally, the probability data includes a second probability. Obtaining the probability data corresponding to each Gaussian component to be fused at the current moment includes: performing kernel density estimation using a second sample data set to obtain a second probability density function; wherein, the second sample data set includes a plurality of second sample obstacle detection probabilities with the obstacle labels all being the target obstacle; obtaining the obstacle detection probabilities corresponding to each Gaussian component to be fused at the current moment; for each Gaussian component to be fused at the current moment, determining the second probability of the Gaussian component to be fused by using the second probability density function and the obstacle detection probability of the Gaussian component to be fused.
[0008] Optionally, based on the probability data of each Gaussian component to be fused at the current moment, determining the fusion weights corresponding to each Gaussian component to be fused at the current moment includes: using the probability data of each Gaussian component to be fused at the current moment to determine the first weight corresponding to each Gaussian component to be fused at the current moment; using the fused Gaussian component at the previous moment to predict the predicted Gaussian component at the current moment; using the difference between each Gaussian component to be fused at the current moment and the predicted Gaussian component to determine the second weight corresponding to each Gaussian component to be fused at the current moment; respectively using the first weight and the second weight of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment.
[0009] Optionally, the probability data includes a first probability and a second probability; using the probability data of each Gaussian component to be fused at the current moment to determine the first weight corresponding to each Gaussian component to be fused at the current moment includes: for each Gaussian component to be fused at the current moment, determining the probability ratio of the first probability to the second probability of the Gaussian component to be fused; using the probability ratios of each Gaussian component to be fused at the current moment to determine the first weight corresponding to each Gaussian component to be fused at the current moment.
[0010] Optionally, using the difference between each Gaussian component to be fused at the current moment and the predicted Gaussian component to determine the second weight corresponding to each Gaussian component to be fused at the current moment includes: respectively determining the KL divergence (Kullback–Leibler Divergence) between each Gaussian component to be fused at the current moment and the predicted Gaussian component; using the KL divergences of each Gaussian component to be fused at the current moment to determine the second weight corresponding to each Gaussian component to be fused at the current moment.
[0011] Optionally, respectively using the first weight and the second weight of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment includes: for each Gaussian component to be fused at the current moment, determining the weight product of the first weight and the second weight of the Gaussian component to be fused; using the weight products of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment.
[0012] To solve the above technical problems, another technical solution adopted by this application is: to provide an object tracking device, which includes: a Gaussian component acquisition module for acquiring the Gaussian components to be fused corresponding to several sensors at the current moment; a probability acquisition module for acquiring the probability data corresponding to each Gaussian component to be fused at the current moment; wherein the probability data includes at least one of a first probability that the obstacle label is the target obstacle and a second probability that the obstacle label is the interference obstacle; a weight determination module for determining the fusion weights corresponding to each Gaussian component to be fused at the current moment based on the probability data of each Gaussian component to be fused at the current moment; a fusion module for fusing each Gaussian component to be fused at the current moment by using each fusion weight to obtain the fused Gaussian component at the current moment; and an object tracking module for using the fused Gaussian component at the current moment as the object tracking result at the current moment.
[0013] Optionally, the probability data includes the first probability. The probability acquisition module is used to perform kernel density estimation by using the first sample data set to obtain the first probability density function; wherein the first sample data set includes several first sample obstacle detection probabilities with the obstacle labels all being the target obstacle; acquire the obstacle detection probabilities corresponding to each Gaussian component to be fused at the current moment; and for each Gaussian component to be fused at the current moment, determine the first probability of the Gaussian component to be fused by using the first probability density function and the obstacle detection probability of the Gaussian component to be fused.
[0014] Optionally, the probability data includes the second probability. The probability acquisition module is used to perform kernel density estimation by using the second sample data set to obtain the second probability density function; wherein the second sample data set includes several second sample obstacle detection probabilities with the obstacle labels all being the target obstacle; acquire the obstacle detection probabilities corresponding to each Gaussian component to be fused at the current moment; and for each Gaussian component to be fused at the current moment, determine the second probability of the Gaussian component to be fused by using the second probability density function and the obstacle detection probability of the Gaussian component to be fused.
[0015] Optionally, the weight determination module is used to determine the first weights corresponding to each Gaussian component to be fused at the current moment by using the probability data of each Gaussian component to be fused at the current moment; predict the predicted Gaussian component at the current moment by using the fused Gaussian component at the previous moment; determine the second weights corresponding to each Gaussian component to be fused at the current moment by using the difference between each Gaussian component to be fused at the current moment and the predicted Gaussian component; and determine the fusion weights corresponding to each Gaussian component to be fused at the current moment by using the first weights and the second weights of each Gaussian component to be fused at the current moment respectively.
[0016] Optionally, the probability data includes a first probability and a second probability. The weight determination module is configured to determine, for each Gaussian component to be fused at the current moment, the probability ratio of the first probability to the second probability of the Gaussian component to be fused; and determine the first weight of each Gaussian component to be fused at the current moment by using the probability ratios of the Gaussian components to be fused at the current moment.
[0017] Optionally, the weight determination module is configured to separately determine the KL divergence between each Gaussian component to be fused and the predicted Gaussian component at the current moment; and determine the second weight corresponding to each Gaussian component to be fused at the current moment by using the KL divergences of the Gaussian components to be fused at the current moment.
[0018] Optionally, the weight determination module is configured to determine, for each Gaussian component to be fused at the current moment, the weight product of the first weight and the second weight of the Gaussian component to be fused; and determine the fusion weight corresponding to each Gaussian component to be fused at the current moment by using the weight products of the Gaussian components to be fused at the current moment.
[0019] To solve the above technical problem, another technical solution adopted by this application is: to provide an electronic device, including a memory and a processor coupled to each other, where the memory stores program instructions; the processor is configured to execute the program instructions stored in the memory to implement the above target tracking method.
[0020] To solve the above technical problem, another technical solution adopted by this application is: to provide a computer-readable storage medium, which is configured to store program instructions that can be executed by a processor to implement the above target tracking method.
[0021] In the above solution, obtain the Gaussian components to be fused corresponding to several sensors at the current moment and the probability data corresponding to each Gaussian component to be fused, where the probability data includes at least one of a first probability that the obstacle label is the target obstacle and a second probability that the obstacle label is the interfering obstacle; determine the fusion weight corresponding to each Gaussian component to be fused at the current moment based on the probability data of the Gaussian components to be fused at the current moment; fuse each Gaussian component to be fused at the current moment by using each fusion weight to obtain the fused Gaussian component at the current moment; and use the fused Gaussian component at the current moment as the target tracking result at the current moment. Since the probability data corresponding to the Gaussian components to be fused of different sensors at the current moment may be different, the fusion weights corresponding to each Gaussian component to be fused at the current moment determined based on the probability data of the Gaussian components to be fused at the current moment are relatively accurate, so that the fusion accuracy of the Gaussian components to be fused of each sensor can be improved, and further the accuracy of target tracking can be improved. Description of the Drawings
[0022] Figure 1It is a schematic flowchart of an embodiment of the target tracking method provided by this application;
[0023] Figure 2 It is a schematic flowchart of another embodiment of the target tracking method provided by this application;
[0024] Figure 3 It is a schematic framework diagram of an embodiment of the target tracking device provided by this application;
[0025] Figure 4 It is a schematic framework diagram of an embodiment of the electronic device provided by this application;
[0026] Figure 5 It is a schematic framework diagram of an embodiment of the computer-readable storage medium provided by this application. Detailed implementation manners
[0027] To make the objectives, technical solutions and effects of this application clearer and more definite, the following further describes this application in detail with reference to the accompanying drawings and by way of examples.
[0028] In the following description, specific details such as specific system architectures, interfaces, and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand this application.
[0029] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, "multiple" in this article means two or more than two. In addition, the term "at least one" in this article means any one of multiple or any combination of at least two of multiple. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set composed of A, B, and C. "Several" means at least one. The terms "first", "second", etc. in the description, claims, and above-mentioned drawings of this article are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence.
[0030] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of an embodiment of the target tracking method provided by this application. It should be noted that if there are substantially the same results, the method of this application is not limited to Figure 1 the flow order shown. As Figure 1 shown, the method includes the following steps:
[0031] S11: Obtain the Gaussian components to be fused corresponding to several sensors at the current moment.
[0032] Among them, several sensors are sensors installed on the vehicle for detecting target obstacles. Exemplarily, several sensors are radars.
[0033] Let the number of several sensors be n k pieces, then there are n k Gaussian components to be fused at the current moment. The Gaussian components to be fused at the current moment include the mean state to be fused and the covariance to be fused at the current moment. In this embodiment, the Gaussian components to be fused at the current moment are the results after single-target Kalman filtering.
[0034] Specifically, the mean state to be fused corresponding to the Gaussian component to be fused at the current moment is determined by the following formula:
[0035]
[0036] Where represents the mean state to be fused corresponding to the i-th Gaussian component to be fused at the current moment, represents the predicted state corresponding to the i-th Gaussian component to be fused at the current moment, K represents the Kalman gain, z represents the measurement state of the sensor corresponding to the i-th Gaussian component to be fused at the current moment, and H represents the measurement matrix. And K are determined by the following formulas respectively:
[0037]
[0038] Where F represents the state transition matrix, represents the fused mean state corresponding to the i-th Gaussian component to be fused at the previous moment, represents the predicted covariance corresponding to the i-th Gaussian component to be fused at the current moment, and R represents the measurement noise covariance. is determined by the following formula:
[0039]
[0040] Where represents the fused covariance corresponding to the i-th Gaussian component to be fused at the previous moment.
[0041] Specifically, the covariance to be fused corresponding to the Gaussian component to be fused at the current moment is determined by the following formula:
[0042]
[0043] S12: Obtain the probability data corresponding to each Gaussian component to be fused at the current moment.
[0044] Among them, the probability data includes at least one of a first probability that the obstacle label is the target obstacle and a second probability that the obstacle label is the interfering obstacle. That is, the probability data may only include the first probability, or only include the second probability, or include both the first probability and the second probability at the same time.
[0045] In one embodiment, obtaining the first probability corresponding to each Gaussian component to be fused at the current moment includes the following steps:
[0046] Step 1: Perform kernel density estimation using the first sample data set to obtain the first probability density function.
[0047] Among them, the first sample data set includes several first sample obstacle detection probabilities where the obstacle labels are all target obstacles. The first sample obstacle detection probability represents the probability of detecting an obstacle corresponding to the sample, and this obstacle may be a target obstacle affecting vehicle driving or an interfering obstacle not affecting vehicle driving. Exemplarily, the first sample data set can be expressed as [t 1 ,t 2 ,...,t m_t . The first sample data set can be obtained based on the historical detection data of the vehicle itself, or can also be obtained based on big data statistics. This embodiment does not make specific limitations on this. The acquisition method of the first sample obstacle detection probability is the same as the acquisition method of the obstacle detection probability in the following text, and specific reference can be made to the following explanation.
[0048] Specifically, the first probability density function is represented by the following formula:
[0049]
[0050] Among them, m_t represents the number of first sample obstacle detection probabilities, h represents the bandwidth, t i represents the i-th first sample obstacle detection probability, and K(x) represents the kernel function. Exemplarily, if K(x) selects the Gaussian kernel function, then K(x) can be represented by the following formula:
[0051]
[0052] Step 2: Obtain the obstacle detection probability corresponding to each Gaussian component to be fused at the current moment.
[0053] In Step 2, the obstacle detection probability corresponding to the Gaussian component to be fused at the current moment is determined by the height of the obstacle detected by the corresponding sensor at the current moment or the motion state of the obstacle detected by the sensor.
[0054] In one case, if the height of the obstacle is within the normal passage height threshold range, the obstacle is regarded as impassable (i.e., the obstacle is a non-high-altitude target and a non-low and passable target). At this time, the obstacle detection probability can be determined according to the height of the obstacle and the first mapping relationship. Among them, the normal passage height threshold range can be set according to actual needs. The first mapping relationship is the mapping relationship between the height of the obstacle and the obstacle detection probability, and the higher the height of the obstacle, the greater the corresponding obstacle detection probability, and the lower the height of the obstacle, the lower the corresponding obstacle detection probability. Exemplarily, the value range of the obstacle detection probability in the first mapping relationship is 60% - 100%.
[0055] In another case, if the height of the obstacle is greater than the high-altitude target threshold value, the obstacle is regarded as a high-altitude target. At this time, the obstacle detection probability can be determined according to the height of the obstacle and the second mapping relationship. Among them, the high-altitude target threshold value can be set according to actual needs. The second mapping relationship is also the mapping relationship between the height of the obstacle and the obstacle detection probability, and the higher the height of the obstacle, the lower the corresponding obstacle detection probability, and the lower the height of the obstacle, the higher the corresponding obstacle detection probability. Exemplarily, the value range of the obstacle detection probability in the second mapping relationship is 0% - 60%.
[0056] In yet another case, if the height of the obstacle is less than the low target threshold value, the obstacle is regarded as a low and passable target. At this time, the obstacle detection probability can be determined according to the height of the obstacle and the third mapping relationship. Among them, the low target threshold value can be set according to actual needs. The third mapping relationship is also the mapping relationship between the height of the obstacle and the obstacle detection probability, and the higher the height of the obstacle, the higher the corresponding obstacle detection probability, and the lower the height of the obstacle, the lower the corresponding obstacle detection probability. Exemplarily, the value range of the obstacle detection probability in the third mapping relationship is 0% - 60%.
[0057] Exemplarily, the first mapping relationship, the second mapping relationship, and the third mapping relationship are all linear mapping relationships.
[0058] In yet another case, if the obstacle detected by the sensor is in a moving state, or changes from a moving state to a stopped state, the corresponding obstacle detection probability is a preset probability value. The preset probability value is a relatively high probability value, and the preset probability value is not less than 60%. Exemplarily, the preset probability value is 60%.
[0059] Step 3: For each Gaussian component to be fused at the current moment, use the first probability density function and the obstacle detection probability of the Gaussian component to be fused to determine the first probability of the Gaussian component to be fused.
[0060] Specifically, the first probability of the Gaussian component to be fused at the current moment can be determined by the following formula:
[0061]
[0062] Among them, f t (o i ) represents the first probability of the i-th Gaussian component to be fused at the current moment, and o i represents the obstacle detection probability corresponding to the i-th Gaussian component to be fused at the current moment.
[0063] In one embodiment, obtaining the second probability corresponding to each Gaussian component to be fused at the current moment includes the following steps:
[0064] Step 1, perform kernel density estimation using the second sample dataset to obtain the second probability density function.
[0065] Among them, the second sample dataset includes several second sample obstacle detection probabilities with the obstacle labels all being target obstacles. The second sample obstacle detection probability represents the probability that the sample detects an obstacle, and this obstacle may be a target obstacle affecting vehicle driving or an interference obstacle not affecting vehicle driving. Exemplarily, the second sample dataset can be expressed as [c 1 , c 2 ,..., c m_c . The second sample dataset can be obtained based on the historical detection data of the vehicle itself, or it can also be obtained based on big data statistics. This embodiment does not make specific limitations on this. The obtaining method of the second sample obstacle detection probability is the same as the obtaining method of the obstacle detection probability in the previous text, and specifically refer to the previous explanation.
[0066] Specifically, the second probability density function is represented by the following formula:
[0067]
[0068] Among them, m_c represents the number of second sample obstacle detection probabilities, h represents the bandwidth, c i represents the i-th second sample obstacle detection probability, and K(x) represents the kernel function. Exemplarily, K(x) selects the Gaussian kernel function.
[0069] Step 2, obtain the obstacle detection probability corresponding to each Gaussian component to be fused at the current moment.
[0070] The relevant content can refer to Step 2 in the previous embodiment and will not be elaborated here.
[0071] Step 3, for each Gaussian component to be fused at the current moment, use the second probability density function and the obstacle detection probability of the Gaussian component to be fused to determine the second probability of the Gaussian component to be fused.
[0072] Specifically, the second probability of the Gaussian component to be fused at the current moment can be determined by the following formula:
[0073]
[0074] where f c (o i ) represents the second probability of the i-th Gaussian component to be fused at the current moment, and o i represents the obstacle detection probability corresponding to the i-th Gaussian component to be fused at the current moment.
[0075] There are two common methods for estimating the probability density function of random variables: kernel density estimation and histogram estimation. Among them, the kernel density estimation (KDE) method is essentially a non-parametric estimation method, which can be used to estimate the probability density function of random variables. This method does not require any assumptions about the distribution of data and can directly use the sample data set and kernel function to fit the probability density function of random variables. Compared with histogram estimation, kernel density estimation can provide a smooth and continuous probability density estimation. In the above two embodiments, kernel density estimation is used to model the obstacle probability. Compared with parametric estimation, it avoids assumptions about the model. Compared with histogram estimation, it can provide a smooth and continuous probability density estimation, thereby improving the accuracy of the first probability and the second probability of each Gaussian component to be fused at the current moment determined.
[0076] S13: Based on the probability data of each Gaussian component to be fused at the current moment, determine the fusion weights corresponding to each Gaussian component to be fused at the current moment.
[0077] The fusion weights corresponding to each Gaussian component to be fused at the current moment are respectively used to characterize the credibility of the corresponding Gaussian component to be fused.
[0078] For each Gaussian component to be fused at the current moment, the fusion weight of the Gaussian component to be fused is positively correlated with the first probability of the Gaussian component to be fused and negatively correlated with the second probability of the Gaussian component to be fused. That is, the higher the first probability of the Gaussian component to be fused at the current moment, the higher the corresponding fusion weight; the lower the first probability of the Gaussian component to be fused at the current moment, the lower the corresponding fusion weight. The higher the first probability of the Gaussian component to be fused at the current moment, the lower the corresponding fusion weight; the lower the first probability of the Gaussian component to be fused at the current moment, the higher the corresponding fusion weight.
[0079] S14: Use each fusion weight to fuse each Gaussian component to be fused at the current moment to obtain the fused Gaussian component at the current moment.
[0080] In one embodiment, the fusion of each Gaussian component to be fused at the current moment is achieved through the AA (Arithmetic Average) fusion method. The specific fusion process can be expressed by the following formula:
[0081]
[0082]
[0083] In another embodiment, the fusion of each Gaussian component to be fused at the current moment is achieved through the GA (Geometric Average) fusion method. The specific fusion process can be expressed by the following formula:
[0084]
[0085] In the above two fusion methods, m AA and P AA respectively represent the fusion mean state and fusion covariance of the Gaussian components to be fused after AA fusion at the current moment. m GA and P GA respectively represent the fusion mean state and fusion covariance after GA fusion at the current moment. n represents the number of Gaussian components to be fused at the current moment. m i represents the mean state to be fused of the i-th Gaussian component to be fused at the current moment, and P i represents the covariance to be fused of the i-th Gaussian component to be fused at the current moment. w i represents the fusion weight corresponding to the i-th Gaussian component to be fused at the current moment.
[0086] S15: Use the fused Gaussian component at the current moment as the target tracking result at the current moment.
[0087] In this embodiment, the Gaussian components to be fused corresponding to several sensors at the current moment and the probability data corresponding to each Gaussian component to be fused are obtained, where the probability data includes at least one of the first probability that the obstacle label is the target obstacle and the second probability that the obstacle label is the interfering obstacle; based on the probability data of each Gaussian component to be fused at the current moment, the fusion weights corresponding to each Gaussian component to be fused at the current moment are determined; each Gaussian component to be fused at the current moment is fused using each fusion weight to obtain the fused Gaussian component at the current moment; the fused Gaussian component at the current moment is used as the target tracking result at the current moment. Since the probability data corresponding to the Gaussian components to be fused of different sensors at the current moment may be different, the fusion weights corresponding to each Gaussian component to be fused at the current moment determined based on the probability data of each Gaussian component to be fused at the current moment are relatively accurate, so that the fusion accuracy of the Gaussian components to be fused of each sensor can be improved, and further the accuracy of target tracking can be improved.
[0088] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of another embodiment of the target tracking method provided by this application. It should be noted that if there are substantially the same results, the method of this application is not limited to Figure 2 the flow order shown. As Figure 2 shown, the method includes the following steps:
[0089] S21: Obtain the Gaussian components to be fused corresponding to several sensors at the current moment.
[0090] S22: Obtain the probability data corresponding to each Gaussian component to be fused at the current moment.
[0091] For the relevant content of step S21 and step S22, reference can be made to step S11 and step S12 respectively, and the detailed description is omitted here.
[0092] S23: Use the probability data of each Gaussian component to be fused at the current moment to determine the first weights corresponding to each Gaussian component to be fused at the current moment.
[0093] In an implementation manner, the probability data of each Gaussian component to be fused at the current moment includes the first probability and the second probability. Step S23 includes the following steps:
[0094] Step 1, for each Gaussian component to be fused at the current moment, determine the probability ratio of the first probability of the Gaussian component to be fused to the second probability of the Gaussian component to be fused.
[0095] In step 1, the probability ratio of the i-th Gaussian component to be fused at the current moment can be expressed as f t (o i ) / f c (o i)。
[0096] Step 2: Determine the first weight of each Gaussian component to be fused at the current moment by using the probability ratio of each Gaussian component to be fused at the current moment.
[0097] In one example, directly use the probability ratio of each Gaussian component to be fused at the current moment as the first weight of each Gaussian component to be fused at the current moment.
[0098] In another example, normalize the probability ratio of each Gaussian component to be fused at the current moment to obtain the first weight of each Gaussian component to be fused at the current moment. Specifically, determine the total probability ratio corresponding to the probability ratios of all Gaussian components to be fused at the current moment; for each Gaussian component to be fused at the current moment, use the ratio of the probability ratio of the Gaussian component to be fused to the total probability ratio as the first weight of the Gaussian component to be fused.
[0099] S24: Use the fused Gaussian component at the previous moment to predict the predicted Gaussian component at the current moment.
[0100] The fused Gaussian component at the previous moment includes the fused mean state and fused covariance at the previous moment, and the predicted Gaussian component at the current moment includes the predicted mean state and predicted covariance at the current moment.
[0101] In one embodiment, if the subsequent fusion of each Gaussian component to be fused at the current moment is performed by the GA fusion method, then the predicted Gaussian component at the current moment can be represented by the following formula:
[0102]
[0103]
[0104] Among them, and respectively represent the fused mean state and fused covariance obtained by the GA fusion method at the previous moment, and respectively represent the predicted mean state and predicted covariance of the predicted Gaussian component obtained through the filtering prediction step at the current moment.
[0105] S25: Determine the second weight corresponding to each Gaussian component to be fused at the current moment by using the difference between each Gaussian component to be fused at the current moment and the predicted Gaussian component.
[0106] In one embodiment, step S25 includes the following steps:
[0107] Step 1: Determine the KL divergence between each Gaussian component to be fused at the current moment and the predicted Gaussian component respectively.
[0108] The KL divergence between each Gaussian component to be fused and the predicted Gaussian component at the current moment is used to characterize the difference between each Gaussian component to be fused and the predicted Gaussian component at the current moment. The KL divergence between each Gaussian component to be fused and the predicted Gaussian component at the current moment can be expressed by the following expression:
[0109]
[0110] where represents the KL divergence between the i-th Gaussian component to be fused at the current moment and the predicted Gaussian component at the current moment.
[0111] Step two, use the KL divergence of each Gaussian component to be fused at the current moment to determine the second weight corresponding to each Gaussian component to be fused at the current moment.
[0112] Specifically, normalize the KL divergence of each Gaussian component to be fused at the current moment to obtain the second weight of each Gaussian component to be fused at the current moment. Specifically, determine the total KL divergence corresponding to the KL divergence of all Gaussian components to be fused at the current moment; for each Gaussian component to be fused at the current moment, use the ratio of the KL divergence of the Gaussian component to be fused to the total KL divergence as the second weight of the Gaussian component to be fused.
[0113] S26: Use the first weight and the second weight of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment.
[0114] Step S26 includes the following steps:
[0115] Step one, for each Gaussian component to be fused at the current moment, determine the product of the first weight and the second weight of the Gaussian component to be fused.
[0116] In step one, the product of the first weight and the second weight of the i-th Gaussian component to be fused at the current moment can be expressed as
[0117] Step two, use the product of the weights of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment.
[0118] In an example, directly use the product of the weights of each Gaussian component to be fused at the current moment as the fusion weight of each Gaussian component to be fused at the current moment.
[0119] In another example, the product of the weights of the Gaussian components to be fused at the current moment is normalized to obtain the fusion weights of the Gaussian components to be fused at the current moment. Specifically, the total weight product corresponding to the product of the weights of all Gaussian components to be fused at the current moment is determined; for each Gaussian component to be fused at the current moment, the ratio of the weight product of the Gaussian component to be fused to the total weight product is used as the fusion weight of the Gaussian component to be fused.
[0120] S27: Use the respective fusion weights to fuse the Gaussian components to be fused at the current moment to obtain the fused Gaussian component at the current moment.
[0121] For the relevant content of step S27, reference can be made to the aforementioned step S14 and will not be elaborated here.
[0122] S28: Take the fused Gaussian component at the current moment as the target tracking result at the current moment.
[0123] In the related art, during the process of multi-sensor fusion for target tracking, usually two methods are adopted to determine the fusion weights corresponding to the Gaussian components to be fused by each sensor at the current moment.
[0124] In the first method, the fusion weights of the Gaussian components to be fused at the current moment are determined using the following formula:
[0125]
[0126] In the first method, any attribute information of the Gaussian components to be fused at the current moment is not considered, and the same fusion weight is assigned to each Gaussian component to be fused at the current moment. The result after fusion will not highlight the Gaussian components with good quality.
[0127] In the second method, the KL divergence between the Gaussian components to be fused at the current moment and the predicted Gaussian component is determined; and the KL divergence of the Gaussian components to be fused at the current moment is used to determine the fusion weights of the Gaussian components to be fused at the current moment.
[0128] When there is no interference among the Gaussian components to be fused at the current moment, it is feasible to determine the fusion weights of the Gaussian components to be fused at the current moment through the second method and good results can be obtained. However, the method of determining the fusion weights only relying on the distance dimension is too single. If the distance between the interfering Gaussian component at the current moment and the predicted Gaussian component at the current moment is close enough (i.e., the difference is small), a large fusion weight will be assigned, and the fusion weights of other correct targets will decrease, resulting in a large deviation between the fusion result and the true result.
[0129] In this embodiment, the first weights corresponding to each Gaussian component to be fused at the current moment are determined by using the probability data of each Gaussian component to be fused at the current moment; the second weights corresponding to each Gaussian component to be fused at the current moment are determined by using the differences between each Gaussian component to be fused at the current moment and the predicted Gaussian component; and the fusion weights corresponding to each Gaussian component to be fused at the current moment are determined by using the first weights and the second weights of each Gaussian component to be fused at the current moment respectively. By combining the two dimensions of the probability dimension and the distance dimension, the reliability of each Gaussian component to be fused at the current moment can be reasonably evaluated, which can avoid the problem of large deviation of the fusion result caused by the limitation of a single dimension, has good stability, improves the fusion accuracy of the Gaussian components to be fused by multiple sensors at the current moment, and further improves the accuracy of target tracking.
[0130] Please refer to Figure 3 , Figure 3 which is a schematic framework diagram of an embodiment of the target tracking device provided by this application. In this embodiment, the target tracking device 30 includes a Gaussian component acquisition module 31, a probability acquisition module 32, a weight determination module 33, a fusion module 34, and a target tracking module 35. The Gaussian component acquisition module 31 is used to acquire the Gaussian components to be fused corresponding to several sensors at the current moment. The probability acquisition module 32 is used to acquire the probability data corresponding to each Gaussian component to be fused at the current moment; wherein, the probability data includes at least one of the first probability that the obstacle label is the target obstacle and the second probability that the obstacle label is the interference obstacle. The weight determination module 33 is used to determine the fusion weights corresponding to each Gaussian component to be fused at the current moment based on the probability data of each Gaussian component to be fused at the current moment. The fusion module 34 is used to fuse each Gaussian component to be fused at the current moment by using each fusion weight to obtain the fused Gaussian component at the current moment. The target tracking module 35 is used to use the fused Gaussian component at the current moment as the target tracking result at the current moment.
[0131] Optionally, the probability data includes the first probability. The probability acquisition module 32 is used to perform kernel density estimation by using the first sample data set to obtain the first probability density function; wherein, the first sample data set includes several first sample obstacle detection probabilities with the obstacle labels all being the target obstacles; acquire the obstacle detection probabilities corresponding to each Gaussian component to be fused at the current moment; and for each Gaussian component to be fused at the current moment, determine the first probability of the Gaussian component to be fused by using the first probability density function and the obstacle detection probability of the Gaussian component to be fused.
[0132] Optionally, the probability data includes a second probability. The probability acquisition module 32 is configured to perform kernel density estimation using the second sample data set to obtain a second probability density function; wherein, the second sample data set includes a plurality of second sample obstacle detection probabilities with the obstacle labels all being target obstacles; obtain the obstacle detection probabilities corresponding to each Gaussian component to be fused at the current moment; for each Gaussian component to be fused at the current moment, use the second probability density function and the obstacle detection probability of the Gaussian component to be fused to determine the second probability of the Gaussian component to be fused.
[0133] Optionally, the weight determination module 33 is configured to use the probability data of each Gaussian component to be fused at the current moment to determine the first weight corresponding to each Gaussian component to be fused at the current moment; use the fused Gaussian component at the previous moment to predict the predicted Gaussian component at the current moment; use the difference between each Gaussian component to be fused at the current moment and the predicted Gaussian component to determine the second weight corresponding to each Gaussian component to be fused at the current moment; respectively use the first weight and the second weight of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment.
[0134] Optionally, the probability data includes a first probability and a second probability. The weight determination module 33 is configured to, for each Gaussian component to be fused at the current moment, determine the probability ratio of the first probability of the Gaussian component to be fused to the second probability of the Gaussian component to be fused; use the probability ratios of each Gaussian component to be fused at the current moment to determine the first weight of each Gaussian component to be fused at the current moment.
[0135] Optionally, the weight determination module 33 is configured to respectively determine the KL divergence between each Gaussian component to be fused at the current moment and the predicted Gaussian component; use the KL divergences of each Gaussian component to be fused at the current moment to determine the second weight corresponding to each Gaussian component to be fused at the current moment.
[0136] Optionally, the weight determination module 33 is configured to, for each Gaussian component to be fused at the current moment, determine the weight product of the first weight and the second weight of the Gaussian component to be fused; use the weight products of each Gaussian component to be fused at the current moment to determine the fusion weight corresponding to each Gaussian component to be fused at the current moment.
[0137] It should be noted that the device in this embodiment can execute the steps in the above method. For the detailed description of related content, please refer to the above method part and will not be repeated here.
[0138] Please refer to Figure 4 , Figure 4 which is a schematic framework diagram of an embodiment of the electronic device provided in this application. In this embodiment, the electronic device 40 includes a memory 41 and a processor 42.
[0139] The processor 42 can also be referred to as a CPU (Central Processing Unit). The processor 42 may be an integrated circuit chip with the ability to process signals. The processor 42 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor 42 can also be any conventional processor 42, etc.
[0140] The memory 41 in the electronic device 40 is used to store program instructions required for the operation of the processor 42.
[0141] The processor 42 is used to execute program instructions to implement the target tracking method in this application.
[0142] Please refer to Figure 5 , Figure 5 , which is a schematic framework diagram of an embodiment of the computer-readable storage medium provided in this application. The computer-readable storage medium 50 of the embodiment of this application stores program instructions 51, and when the program instructions 51 are executed, the target tracking method provided in this application is implemented. Among them, the program instructions 51 can form a program file and be stored in the above computer-readable storage medium 50 in the form of a software product, so that a computer device (which can be a personal computer, a server, or a network device, etc.) can execute all or part of the steps of the methods of various embodiments of this application. And the aforementioned computer-readable storage medium 50 includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.
[0143] In the above solution, the to-be-fused Gaussian components corresponding to several sensors at the current moment and the probability data corresponding to each to-be-fused Gaussian component are obtained, where the probability data includes at least one of the first probability that the obstacle label is the target obstacle and the second probability that the obstacle label is the interfering obstacle; based on the probability data of each to-be-fused Gaussian component at the current moment, the fusion weights corresponding to each to-be-fused Gaussian component at the current moment are determined; each to-be-fused Gaussian component at the current moment is fused by using each fusion weight to obtain the fused Gaussian component at the current moment; the fused Gaussian component at the current moment is used as the target tracking result at the current moment. Since the probability data corresponding to the to-be-fused Gaussian components of different sensors at the current moment may be different, the fusion weights corresponding to each to-be-fused Gaussian component at the current moment determined based on the probability data of each to-be-fused Gaussian component at the current moment are relatively accurate, so that the fusion accuracy of the to-be-fused Gaussian components of each sensor can be improved, and further the accuracy of target tracking can be improved.
[0144] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the methods described in the method embodiments above. The specific implementation can refer to the description of the method embodiments above. For the sake of brevity, it will not be repeated here.
[0145] The descriptions of the above embodiments tend to emphasize the differences between the embodiments. The same or similar parts can be referred to each other. For the sake of brevity, they will not be repeated in this article.
[0146] In several embodiments provided by the present application, it should be understood that the disclosed methods, devices and systems can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces. The indirect coupling or communication connection of the device or unit can be in electrical, mechanical or other forms.
[0147] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0148] In addition, in each embodiment of the present application, the functional units may be integrated into one processing unit, may exist physically separately for each unit, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.
[0149] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. With such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods in various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.
[0150] The above are only the embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.
Claims
1. A target tracking method, characterized in that: The method comprises: Obtain the Gaussian components to be fused corresponding to several sensors at the current moment; Obtaining probability data corresponding to each of the Gaussian components to be fused; wherein the probability data includes at least one of a first probability that the obstacle label is a target obstacle and a second probability that the obstacle label is an interference obstacle; Based on the probability data of each Gaussian component to be fused, determining the fusion weights corresponding to each Gaussian component to be fused; Using the fusion weights to fuse the Gaussian components to be fused, to obtain the fused Gaussian components at the current moment; The fused Gaussian component at the current moment is used as the target tracking result at the current moment.
2. The method according to claim 1, characterized in that The probability data includes the first probability, and the acquiring the probability data corresponding to each of the Gaussian components to be fused includes: Performing kernel density estimation using a first sample data set to obtain a first probability density function; wherein the first sample data set includes a number of first sample obstacle detection probabilities whose obstacle labels are all target obstacles; Obtaining obstacle detection probabilities corresponding to the Gaussian components to be fused; For each of the Gaussian components to be fused, the first probability of the Gaussian components to be fused is determined by using the first probability density function and the obstacle detection probability of the Gaussian components to be fused.
3. The method according to claim 1, characterized in that The probability data includes the second probability, and the acquiring the probability data corresponding to each of the Gaussian components to be fused includes: Performing kernel density estimation using the second sample data set to obtain a second probability density function; wherein the second sample data set includes a number of second sample obstacle detection probabilities whose obstacle labels are all target obstacles; Obtaining obstacle detection probabilities corresponding to the Gaussian components to be fused; For each of the Gaussian components to be fused, the second probability of the Gaussian components to be fused is determined by using the second probability density function and the obstacle detection probability of the Gaussian components to be fused.
4. The method according to claim 1, characterized in that: The determining, based on the probability data of each Gaussian component to be fused, the fusion weights corresponding to each Gaussian component to be fused, respectively, includes: Determine first weights respectively corresponding to the Gaussian components to be fused by using the probability data of the Gaussian components to be fused; Using the fused Gaussian component of the previous moment, the predicted Gaussian component of the current moment is predicted; Determine the second weights respectively corresponding to the Gaussian components to be fused by using the differences between the Gaussian components to be fused and the predicted Gaussian components; The fusion weights corresponding to the Gaussian components to be fused are determined by respectively using the first weights and the second weights of the Gaussian components to be fused.
5. The method according to claim 4, characterized in that The probability data includes the first probability and the second probability; and using the probability data of each Gaussian component to be fused to determine the first weights corresponding to each Gaussian component to be fused, respectively, includes: For each of the Gaussian components to be fused, determining a probability ratio of the first probability of the Gaussian component to be fused to the second probability of the Gaussian component to be fused; The first weight of each Gaussian component to be fused is determined by using the probability ratio of each Gaussian component to be fused.
6. The method according to claim 4, characterized in that The determining, by using the difference between each of the Gaussian components to be fused and the predicted Gaussian components, second weights respectively corresponding to each of the Gaussian components to be fused includes: Determine the KL divergence between each of the Gaussian components to be fused and the predicted Gaussian components respectively; The second weights respectively corresponding to the Gaussian components to be fused are determined by using the KL divergence of the Gaussian components to be fused.
7. The method according to claim 4, characterized in that The determining the fusion weights corresponding to the Gaussian components to be fused respectively by using the first weights and the second weights of the Gaussian components to be fused respectively includes: For each of the Gaussian components to be fused, determining a weight product of the first weight and the second weight of the Gaussian component to be fused; The fusion weights corresponding to the Gaussian components to be fused are determined by using the weight products of the Gaussian components to be fused.
8. A target tracking device, characterized in that: The device comprises: A Gaussian component acquisition module, used to obtain the Gaussian components to be fused corresponding to a number of sensors at the current moment; A probability acquisition module, used to acquire probability data corresponding to each of the Gaussian components to be fused; wherein the probability data includes at least one of a first probability that the obstacle label is a target obstacle and a second probability that the obstacle label is an interference obstacle; A weight determination module, used to determine the fusion weights corresponding to the Gaussian components to be fused respectively based on the probability data of the Gaussian components to be fused; A fusion module, used for fusing the Gaussian components to be fused by using the fusion weights to obtain a fused Gaussian component at the current moment; The target tracking module is used to use the fused Gaussian component at the current moment as the target tracking result at the current moment.
9. An electronic device, characterized in that: comprising a memory and a processor coupled to each other, The memory stores program instructions; The processor is used to execute the program instructions stored in the memory to implement the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program instructions, and the program instructions can be executed by a processor to implement the method according to any one of claims 1 to 7.