A weighted evidence fusion processing method, system, storage medium, and vehicle
By combining the ER evidence reasoning algorithm with sensor characteristic parameters, the stability problem of target recognition in intelligent driving vehicles of traditional algorithms is solved, and adaptive adjustment is achieved when sensor characteristics differ and change, thereby improving the adaptation efficiency of the fusion algorithm and driving safety.
Patent Information
- Application Number
- CN202211418284.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-14
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-11-14
AI Technical Summary
Traditional Dempster evidence reasoning algorithms suffer from counterintuitive and combinatorial explosion problems in autonomous vehicles, affecting the stability of target recognition category fusion. Furthermore, existing solutions cannot adaptively adjust weight parameters to adapt to differences and changes in sensor characteristics.
The ER evidence reasoning algorithm is adopted to extract target recognition results from multiple sensors, generate target matching sets and historical matching sets, update sensor weights, and perform information fusion calculation using sensor characteristic parameters and the classification confidence discrete parameters of historical matching sets to obtain the confidence information of each matching target.
It improves the adaptation efficiency and adjustment capability of the fusion algorithm, enhancing vehicle safety, especially the adaptation efficiency when sensor models and types change.
Smart Images

Figure CN118070054B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of autonomous driving technology, and in particular to a weighted evidence fusion processing method, system, storage medium, and vehicle. Background Technology
[0002] Traditional Demspter evidence reasoning algorithms suffer from counterintuitive and combinatorial explosion problems, which affect the stability of target recognition category fusion in intelligent driving vehicle applications and impact vehicle safety.
[0003] To overcome these two problems, the industry has developed the ER evidence reasoning algorithm based on Dempster's combinatorial rule. Its key idea is to allocate residual confidence to each unit and ignorance to the identification frame, and to give each unit a weight parameter. The confidence of a single piece of evidence is adjusted by adjusting the weight parameter. Therefore, choosing a suitable weight parameter is particularly important. A reasonable parameter design will greatly improve the performance of the ER evidence reasoning algorithm.
[0004] In intelligent driving vehicle applications, the differences in characteristics brought about by the selection of different models and brands of sensors for different vehicles, and even the differences in characteristics due to manufacturing processes of the same model of sensor, can affect the weighting parameters and thus the algorithm's adaptation efficiency. Furthermore, when the characteristics of the sensors being fused change frequently, existing solutions lack the ability to adaptively adjust. Summary of the Invention
[0005] The technical problem to be solved by the present invention is to provide a weighted evidence fusion processing method, system, storage medium and vehicle, which can improve the adaptation efficiency and adjustment capability of the fusion algorithm, thereby improving driving safety.
[0006] To address the aforementioned technical problems, as one aspect of the present invention, a weighted evidence fusion processing method is provided, applied to an intelligent driving system, comprising at least the following steps: Extract the current target recognition results from multiple sensors in the intelligent driving system; Read the latest target recognition results from multiple sensors, combine them with the target recognition results from the nearest time, and match the recognition results one by one to generate a target matching set and a historical matching set; Retrieve the target matching set and the historical matching set to obtain the current frame matched targets corresponding to each matching target, as well as the category information of the corresponding historical matching targets, as evidence information corresponding to the matching targets, and assign basic probability values. The weights of each matching target in each sensor are updated based on the sensor characteristic parameters in the pre-generated characteristic parameter matrix of each sensor and the classification confidence dispersion parameter in the historical matching set. The ER evidence reasoning algorithm is used to perform information fusion calculation to obtain the confidence information of each category corresponding to each matching target within the identification framework; Based on the confidence information, the target category fusion result corresponding to each matching target is obtained.
[0007] The multiple sensors include a combination of at least two of a camera, millimeter-wave radar, and lidar; the step of extracting the current target recognition results from the multiple sensors in the intelligent driving system includes: The information collected by each sensor is obtained according to the output frequency of each sensor, forming a corresponding target recognition matrix.
[0008] The specific steps of reading the latest target recognition results from multiple sensors, combining them with the target recognition results from the nearest time, and matching the recognition results one by one to generate a target matching set and a historical matching set are as follows: Read the latest sensor target recognition results from multiple sensors according to the principle of the nearest historical moment; Assuming that the target motion between the nearest neighbor time and the time being read each time conforms to a linear uniform motion model, the multi-sensor target recognition results are matched one by one according to the linear motion model and the Hungarian matching algorithm based on Euclidean distance to generate a target matching set and a historical matching set. Each target matching set includes a matching timestamp, target matching status, target matching identifier, target source sensor identifier, and target recognition matrix of the source sensor; the historical matching set is used to save target matching set information for a predetermined time before the matching timestamp.
[0009] The step of retrieving the target matching set and the historical matching set, obtaining the currently matched targets corresponding to each matching target, and the category information of the corresponding historical matching targets, as evidence information corresponding to the matching targets, and assigning basic probability values, further includes: By retrieving the target matching identifier, we can obtain the original evidence information of the category of the target matched in the current frame and the historical matching targets of the same target in the historical matching set; The target identification information acquired from multiple sensors with different properties is transformed into multiple pieces of evidence, and each fused piece of evidence is treated as a vector (e i And establish an identification framework Θ (θ1, θ2, ..., θ) consisting of N evaluation levels. N ); and for each piece of evidence (e i The basic probability values for the corresponding focal elements are assigned as follows: m i (θ1), m i (θ2), …, m i (θ NThe evaluation level corresponds to the category of the matching target.
[0010] The specific steps for updating the weights corresponding to each matching target in each sensor based on the sensor characteristic parameters in the characteristic parameter matrix of each sensor and the classification confidence dispersion parameter in the historical matching set are as follows: Based on the characteristic parameter matrix of each input sensor, the initial weight parameters of each sensor are assigned values. For each matching target, the current frame's matched targets and the category information of the corresponding historical matched targets are used to calculate the confidence variance and standard deviation of the matched targets; Select a predetermined number of frames within the window period of the historical matching set, find the frame with the smallest variance within the window period, calculate the difference between the confidence level of the state estimate of this frame and the confidence level of the original state output of the sensor at this moment, and use the ratio R of this to the confidence level of the state estimate of this frame as the deviation coefficient. The weight wi for each sensor and each matched target is calculated using the following formula:
[0011] Where w is the initial weight parameter corresponding to the sensor, and sd is the standard deviation of the confidence level of the matched target.
[0012] The step of obtaining the target category fusion result corresponding to each matching target based on the confidence information is as follows: Based on the confidence information of the target classification of each matched target within the calculated identification framework, the classification result with the highest confidence is assigned to the matched target as its target category fusion result.
[0013] The step of using the ER evidence reasoning algorithm to perform information fusion calculation to obtain the confidence information of each category corresponding to each matching target within the identification framework further includes: The combinational probability mass of all pairwise evidence e1,…,en within the identification frame is obtained using the following formula: (3) (4) (5) (6) in, Indicate e1 and e n After combination, it is assigned to proposition θ n The joint probability quality; Indicate e1 and e n The joint probability quality assigned to the identification frame after combination; Indicate e1 and e n The joint probability mass assigned to the power set after combination; K0 represents the normalization coefficient when the two pieces of evidence are combined to ensure that the sum of the probability masses after combination is 1; The corresponding combined confidence level is obtained using the following formula (7):
[0014] (7) The evaluation results corresponding to the evidence are obtained using the confidence distribution form in Equation (8) as follows: (8) Regarding L independent pieces of evidence e i (i=1,…,L), repeat formulas (3) to (7), combine the basic probability mass obtained from the previous combination with the basic probability mass of the third piece of evidence, and repeat in this way to obtain the following final evaluation result of the evaluation scheme: (9) This allows us to obtain the confidence information for each category corresponding to each matching target.
[0015] As another aspect of the present invention, a weighted evidence fusion processing system is also provided, applied to an intelligent driving system, comprising at least: The target recognition result acquisition unit is used to extract the current target recognition results from multiple sensors in the intelligent driving system; The matching set acquisition unit is used to read the latest target recognition results from multiple sensors, combine them with the target recognition results from the nearest time, and match the recognition results one by one to generate a target matching set and a historical matching set. The evidence information extraction unit is used to retrieve the target matching set and the historical matching set, obtain the current frame matched targets corresponding to each matching target, and the category information of the corresponding historical matching targets, as evidence information corresponding to the matching targets, and perform basic probability assignment. The weight allocation unit is used to update the weight of each matching target in each sensor according to the sensor characteristic parameters in the pre-generated characteristic parameter matrix of each sensor and the classification confidence dispersion parameter in the historical matching set. The fusion computing processing unit is used to perform information fusion computing using the ER evidence reasoning algorithm to obtain the confidence information of each category corresponding to each matching target within the identification framework; The result acquisition unit is used to obtain the target category fusion result corresponding to each matching target based on the confidence information.
[0016] As another aspect of the present invention, a computer-readable medium is also provided, on which a computer program is stored, which, when executed by a processor, performs the steps of the method described above.
[0017] As another aspect of the present invention, a vehicle is also provided, wherein an intelligent driving system is provided, the intelligent driving system having multiple types of sensors and integrating the aforementioned weighted evidence fusion processing system.
[0018] Implementing the embodiments of the present invention has the following beneficial effects: This invention provides a weighted evidence fusion processing method, system, storage medium, and vehicle. In this invention, during the multi-sensor target recognition category fusion process, evidence weights are automatically updated by introducing sensor characteristic parameters and the discreteness parameters of classification confidence in historical matching sets. An Evidence Reasoning (ER) algorithm is used for information fusion calculation to obtain the confidence information of each category corresponding to each matched target within the identification framework. Then, the category corresponding to each matched target is determined based on the confidence information. In this embodiment, the weight parameters of the ER evidence fusion algorithm can be automatically adjusted using vehicle sensor characteristic parameters, improving the optimization efficiency of the evidence fusion algorithm parameter adjustment due to changes in vehicle sensor model and type. This improves the adaptability and adjustment capability of the fusion algorithm, thereby enhancing driving safety. Attached Figure Description
[0019] 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, obtaining other drawings based on these drawings without creative effort still falls within the scope of the present invention.
[0020] Figure 1 This is a schematic diagram of the main flow of an embodiment of a weighted evidence fusion processing method provided by the present invention; Figure 2 This is a schematic diagram of an embodiment of a weighted evidence fusion processing system provided by the present invention. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.
[0022] like Figure 1 The diagram illustrates the main flow of an embodiment of a weighted evidence fusion processing method provided by the present invention; in this embodiment, the method includes at least the following steps: Step S10: Extract the current target recognition results from multiple sensors in the intelligent driving system; in this embodiment, the multiple sensors include a combination of at least two of a camera, millimeter-wave radar, and lidar. More specifically, step S10 further includes: The information collected by each sensor is obtained according to its corresponding output frequency, forming a corresponding target recognition matrix. In one example, the vehicle-side target recognition results come from a camera, millimeter-wave radar, and lidar. The camera-recognized target matrix S1(ob) includes information such as the observation time timestamp, observation time position, observation time speed, observation time heading angle, observation time classification, observation time classification confidence, and observation time size. The millimeter-wave radar-recognized target matrix S2(ob) may include information such as the observation time timestamp, observation time position, observation time speed, observation time heading angle, and observation time size. The lidar-recognized target matrix S3(ob) includes information such as the observation time timestamp, observation time position, observation time speed, observation time heading angle, observation time classification, and observation time size.
[0023] Step S11: Obtain the characteristic parameter matrix A of multiple sensors. In this invention, sensor characteristics are represented by sensor accuracy parameters, with the accuracy rates for each target classification being P1, P2, P3…Pk, respectively. Representing the multiple sensors as S1, S2, and S3, the characteristic parameter matrix A of the multiple sensors can be represented as {S1(P1…Pk), S2(P1…Pk), S3(P1…Pk)}.
[0024] Step S12: Read the latest target recognition results from multiple sensors, combine them with the target recognition results from the nearest time, and match the recognition results one by one to generate a target matching set and a historical matching set. Since the output frequencies of multiple sensors are different, in this embodiment of the invention, the latest sensor target recognition results output by multiple sensors are read according to the principle of the nearest historical moment.
[0025] In this specific example, the target recognition output frequency of each sensor is higher than 30Hz. Therefore, it is assumed that the target motion between the nearest neighbor time and the time of the read is a linear uniform motion model. Based on the linear motion model and the Hungarian matching algorithm based on Euclidean distance, the multi-sensor target recognition results are matched one by one to generate a target matching set and a historical matching set. Specifically, each target matching set includes a matching timestamp, target matching status ST, target matching identifier ID, target source sensor identifier ID, and target recognition matrix Sn(ob) of the source sensor; while the historical matching set is used to store target matching set information for a predetermined time (e.g., 1 second) before the matching timestamp.
[0026] Step S13: Retrieve the target matching set and the historical matching set, obtain the current frame matched targets corresponding to each matching target, and the category information of the corresponding historical matching targets, as evidence information corresponding to the matching targets, and assign basic probability values. Specifically, in the target matching set and the historical matching set, the original evidence information of the categories of the matched targets in the current frame and the historical matched targets of the same target in the historical matching set is obtained by retrieving the target matching identifier, and the basic probability of the corresponding evidence focal elements is assigned.
[0027] First, the target identification information acquired from multiple sensors with different properties is transformed into multiple pieces of evidence. Each fused piece of evidence is treated as a vector. Specifically, each piece of evidence is treated as a vector, and the vector of the i-th piece of evidence is denoted by e. i Let i = 1, 2, ..., L; L be the total number of evidence vectors.
[0028] Suppose the evidence from L sensors is e1, e2, ..., e L Assume the identification framework Θ consists of N evaluation levels (θ1, θ2, ..., θ3). N Composed of ) the i-th piece of evidence e i The corresponding basic probability assignments for the focal element are m. i (θ1),m i (θ2),…,m i (θ N If the i-th piece of evidence is given by the following confidence distribution: (1) in, This indicates that the assessment plan is based on evidence. The following was assessed as level θ n The confidence level, and satisfying 0 ≤ ≤1, . This represents global ignorance. For any... , Called One of the focal elements.
[0029] In an embodiment of the present invention, evidence from a camera, a millimeter-wave radar, and a lidar is designated as e1, e2, and e3, respectively. An identification framework Θ (θ1, θ2, ..., θ3) with N evaluation levels is established. NThe evaluation level corresponds to the category of the matched target. For classification fusion, the categories corresponding to the focal elements in the identification framework can be ordinary vehicles, special vehicles, two-wheeled vehicles, other vehicles, dynamic pedestrians, static pedestrians, unknown targets, dynamic targets, static targets, and vehicles. In one example, the identification framework contains 10 focal elements, i.e., N=10.
[0030] Among them, vehicles are equivalent to a combination of ordinary vehicles, special vehicles, two-wheeled vehicles and other vehicles; dynamic targets are equivalent to a combination of dynamic vehicles, dynamic pedestrians and unknown targets; and static targets are equivalent to a combination of static vehicles, static pedestrians and unknown targets.
[0031] For each piece of evidence (e) i The basic probability values for the corresponding focal elements are assigned as follows: m i (θ1), m i (θ2), …, m i (θ N In one example, the focal element probability corresponding to the evidence involving a target matching ID is assigned as e1 (ordinary vehicle) = Sn (ob). Classification confidence level at observation time; e1 (vehicle) = Sn (ob) (1 - classification confidence level at observation time).
[0032] Step S14: Update the weight of each matching target in each sensor according to the sensor characteristic parameters in the pre-generated characteristic parameter matrix of each sensor and the classification confidence dispersion parameter in the historical matching set. In embodiments of the present invention, the weight of each matching target of each sensor is updated by introducing sensor characteristic parameters and the dispersion parameter of classification confidence in historical matching sets.
[0033] Specifically, in one example, step S14 further includes: Considering the characteristics of each sensor, an initial weight parameter needs to be set for each sensor to form a weight vector W{w1, w2, w3}. Specifically, the characteristic parameter matrix A of each input sensor is assigned to W.
[0034] Using the original evidence information of category or size of the matched targets in the current frame of the sensor and the historical matched targets of the same target in the historical matching set extracted in step S13 above, the confidence variance and standard deviation sd of the matched targets are calculated. In one example, a predetermined number of frames (e.g., 10 frames) are selected within the window period of the historical matching set. The frame with the smallest variance is found within the window period. The difference between the confidence level of the state estimate of this frame and the confidence level of the original state output of the sensor at this moment is calculated. The ratio R of this to the confidence level of the state estimate of this frame is used as the deviation coefficient. The purpose is to correct some fusion results with large estimation deviations.
[0035] The weight wi for each sensor and each matched target is calculated using the following formula:
[0036] Where w is the initial weight parameter corresponding to the sensor, sd is the standard deviation of the confidence of the matched target, and R is the deviation coefficient.
[0037] Step S15: Use the Evidence Reasoning (ER) algorithm to perform information fusion calculation to obtain the confidence information of each category corresponding to each matching target within the identification framework.
[0038] In this embodiment of the invention, for evidence e i Its weighted basic probability mass can be expressed as: (2) Then the probability mass of pairwise combinations among all pieces of evidence e1,…,en within this identification framework is: (3) (4) (5) (6) in, Indicate e1 and e n After combination, it is assigned to proposition θ n The joint probability quality; Indicate e1 and e n The joint probability quality assigned to the identification frame after combination; Indicate e1 and e n The joint probability mass assigned to the power set after combination; K0 represents the normalization coefficient when the two pieces of evidence are combined to ensure that the sum of the probability masses after combination is 1; The corresponding combined confidence level is obtained using the following formula (7):
[0039] (7) The evaluation results corresponding to the evidence are obtained using the confidence distribution form in Equation (8) as follows: (8) Regarding L independent pieces of evidence e i (i=1,…,L), repeat formulas (3) to (7), combine the basic probability mass obtained from the previous combination with the basic probability mass of the third piece of evidence, and repeat in this way to obtain the following final evaluation result of the evaluation scheme: (9) That is, to obtain the confidence information of each category corresponding to each piece of evidence (matching target).
[0040] Step S16: Based on the confidence information, obtain the target category fusion result corresponding to each matching target and output it.
[0041] In this invention, based on the confidence information of the target classification of each matching target within the calculated identification framework, the classification result with the highest confidence is assigned to the matching target as its target category fusion result.
[0042] For example, in one case, if the static target has the highest confidence among the matching targets, then the target category fusion result of the matching target is the static target.
[0043] like Figure 2 The diagram illustrates a structural schematic of an embodiment of a weighted evidence fusion processing system provided by the present invention. In this embodiment, the weighted evidence fusion processing system 1 is applied to an intelligent driving system and includes at least: The target recognition result acquisition unit 10 is used to extract the current target recognition results from multiple sensors in the intelligent driving system; The matching set acquisition unit 11 is used to read the latest target recognition results of multiple sensors, combine them with the target recognition results of the nearest time, match the recognition results one by one, and generate a target matching set and a historical matching set. The evidence information extraction unit 12 is used to retrieve the target matching set and the historical matching set, obtain the current frame matched target corresponding to each matching target, and the category information of the corresponding historical matching target, as evidence information corresponding to the matching target, and perform basic probability assignment. The weight allocation unit 13 is used to update the weight of each matching target in each sensor according to the sensor characteristic parameters in the pre-generated characteristic parameter matrix of each sensor and the classification confidence dispersion parameter in the historical matching set. The fusion computing processing unit 14 is used to perform information fusion computing using the ER evidence reasoning algorithm to obtain the confidence information of each category corresponding to each matching target within the identification framework. The result acquisition unit 15 is used to obtain the target category fusion result corresponding to each matching target based on the confidence information.
[0044] For more details, please refer to and combine with the above. Figure 1 The description of that will not be repeated here.
[0045] In another aspect, the present invention provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements as follows: Figure 1 The steps of the described method. For more details, please refer to and combine with the foregoing explanation. Figure 1 The description of that will not be repeated here.
[0046] In another aspect, the present invention provides a vehicle equipped with an intelligent driving system, the intelligent driving system having various types of sensors (such as cameras, millimeter-wave radar, and lidar), and integrating as shown in the appendix. Figure 2 The weighted evidence fusion processing system described above. For more details, please refer to and combine with the aforementioned... Figure 2 The description of that will not be repeated here.
[0047] Implementing the embodiments of the present invention has the following beneficial effects: This invention provides a weighted evidence fusion processing method, system, storage medium, and vehicle. In this invention, during the multi-sensor target recognition category fusion process, evidence weights are automatically updated by introducing sensor characteristic parameters and the discreteness parameters of classification confidence in historical matching sets. An Evidence Reasoning (ER) algorithm is used for information fusion calculation to obtain the confidence information of each category corresponding to each matched target within the identification framework. Then, the category corresponding to each matched target is determined based on the confidence information. In this embodiment, the weight parameters of the ER evidence fusion algorithm can be automatically adjusted using vehicle sensor characteristic parameters, improving the optimization efficiency of the evidence fusion algorithm parameter adjustment due to changes in vehicle sensor model and type. This improves the adaptability and adjustment capability of the fusion algorithm, thereby enhancing driving safety.
[0048] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0049] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0050] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.
Claims
1. A weighted evidence fusion processing method applied to an intelligent driving system, characterized in that, It should include at least the following steps: Extract the current target recognition results from multiple sensors in the intelligent driving system; Read the latest target recognition results from multiple sensors, combine them with the target recognition results from the nearest time, and match the recognition results one by one to generate a target matching set and a historical matching set; Retrieve the target matching set and the historical matching set to obtain the current frame matched targets corresponding to each matching target, as well as the category information of the corresponding historical matching targets, as evidence information corresponding to the matching targets, and assign basic probability values. The steps are as follows: First, the initial weight parameters of each sensor are updated based on the sensor characteristic parameters in the pre-generated characteristic parameter matrix of each sensor and the classification confidence dispersion parameters in the historical matching set. Second, the confidence variance and standard deviation of the matched targets are calculated based on the current frame matched targets and the category information of the corresponding historical matched targets. Third, a predetermined number of frames are selected within the window period of the historical matching set. The frame with the smallest variance is found within the window period. The difference between the confidence of the state estimate of this frame and the state confidence of the original sensor output at this moment is calculated. The ratio of this difference to the confidence of the state estimate of this frame is used as the deviation coefficient. Calculate the weights for each sensor and each matched target; The Evidence Reasoning (ER) algorithm is used to perform information fusion calculations to obtain the confidence information of each category corresponding to each matching target within the identification framework; Based on the confidence information, the target category fusion result corresponding to each matching target is obtained.
2. The method of claim 1, wherein, The multiple sensors include a combination of at least two of cameras, millimeter-wave radar, and lidar; the step of extracting the current target recognition results from the multiple sensors in the intelligent driving system includes: The information collected by each sensor is obtained according to the output frequency of each sensor, forming a corresponding target recognition matrix.
3. The method of claim 1, wherein, The steps of reading the latest target recognition results from multiple sensors, combining them with the target recognition results from the nearest time, and matching the recognition results one by one to generate a target matching set and a historical matching set are as follows: Read the latest sensor target recognition results from multiple sensors according to the principle of the nearest historical moment; Assuming that the target motion between the nearest neighbor time and the time being read each time conforms to a linear uniform motion model, the multi-sensor target recognition results are matched one by one according to the linear motion model and the Hungarian matching algorithm based on Euclidean distance to generate a target matching set and a historical matching set. Each target matching set includes a matching timestamp, target matching status, target matching identifier, target source sensor identifier, and target recognition matrix of the source sensor; the historical matching set is used to save target matching set information for a predetermined time before the matching timestamp.
4. The method of claim 3, wherein, The step of retrieving the target matching set and historical matching set, obtaining the currently matched targets corresponding to each matching target, and the category information of the corresponding historical matching targets, as evidence information corresponding to the matching target, and assigning basic probability values, further includes: In the target matching set and the historical matching set, the original evidence information of the category of the target matched in the current frame and the historical matching target of the same target in the historical matching set is obtained by retrieving the target matching identifier; The identification information of the target obtained by the plurality of sensors with different properties is converted into a plurality of evidences, each fused evidence is regarded as a vector (e i ), and a recognition framework Θ (θ1, θ2, …, θ N ) with N evaluation levels is established; and basic probability assignments of the focal elements corresponding to each evidence (e i ) are respectively: m i (θ1), m i (θ2), …, m i (θ N ); wherein the evaluation levels correspond to the categories of the matched target.
5. The method of claim 4, wherein, The step of using the ER evidence reasoning algorithm to perform information fusion calculation to obtain the confidence information of each category corresponding to each matching target within the identification framework further includes: The combinational probability mass of all pairwise evidence e1,…,en within the identification frame is obtained using the following formula: (3) (4) (5) (6) in, Indicate e1 and e n After combination, it is assigned to proposition θ n The joint probability quality; Indicate e1 and e n The joint probability quality assigned to the identification frame after combination; Indicate e1 and e n The joint probability mass assigned to the power set after combination; K0 represents the normalization coefficient when the two pieces of evidence are combined to ensure that the sum of the probability masses after combination is 1; The corresponding combined confidence level is obtained using the following formula (7): (7) The evaluation results corresponding to the evidence are obtained using the confidence distribution form in Equation (8) as follows: (8) For L pieces of independent evidence e i (i = 1, …, L), repeat formulas (3) to (7) to combine the basic probability mass obtained by the previous combination with the basic probability mass of the third piece of evidence, and cycle in turn to obtain the final evaluation result of the evaluation scheme as follows: (9) This allows us to obtain the confidence information for each category corresponding to each matching target.
6. The method as described in claim 5, characterized in that, The step of obtaining the target category fusion result corresponding to each matching target based on the confidence information is as follows: Based on the confidence information of the target classification of each matched target within the calculated identification framework, the classification result with the highest confidence is assigned to the matched target as its target category fusion result.
7. A weighted evidence fusion processing system, applied in an intelligent driving system, characterized in that, At least including: The target recognition result acquisition unit is used to extract the current target recognition results from multiple sensors in the intelligent driving system; The matching set acquisition unit is used to read the latest target recognition results from multiple sensors, combine them with the target recognition results from the nearest time, and match the recognition results one by one to generate a target matching set and a historical matching set. The evidence information extraction unit is used to retrieve the target matching set and the historical matching set, obtain the current frame matched targets corresponding to each matching target, and the category information of the corresponding historical matching targets, as evidence information corresponding to the matching targets, and perform basic probability assignment. The weight allocation unit is used to update the weight of each matching target in each sensor according to the sensor characteristic parameters in the pre-generated characteristic parameter matrix of each sensor and the classification confidence dispersion parameter in the historical matching set. The fusion computing processing unit is used to perform information fusion computing using the ER evidence reasoning algorithm to obtain the confidence information of each category corresponding to each matching target within the identification framework; The result acquisition unit is used to obtain the target category fusion result corresponding to each matching target based on the confidence information; The weight allocation unit updates the weights corresponding to each matching target in each sensor in the following way: It assigns initial weight parameters to each sensor based on the characteristic parameter matrix of each input sensor; it calculates the confidence variance and standard deviation of the matched targets in the current frame corresponding to each matching target, as well as the category information of the historical matched targets; it selects a predetermined number of frames within the window period of the historical matching set, finds the frame with the smallest variance within the window period, calculates the difference between the confidence of the state estimate of this frame and the confidence of the original state output of the sensor at this moment, and uses the ratio of this difference to the confidence of the state estimate of this frame as the deviation coefficient; and it calculates the weights for each matching target for each sensor.
8. A computer-readable medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1-6.
9. A vehicle equipped with an intelligent driving system, characterized in that, The intelligent driving system has multiple types of sensors and integrates the weighted evidence fusion processing system as described in claim 7.
