Collaborative confidence fusion perception method applied to automatic driving

By performing time alignment and confidence correction on the intelligent sense body of the autonomous driving perception system, combining the BEV spatial confidence map and offline map, the results are fusion using the Softmax method of intersection ratio and temperature adjustment, and the limited perception range and unified feature are solved. The detection effect problems caused by this improve the robustness and accuracy of the autonomous driving perception system.

CN119027934AActive Publication Date: 2024-11-26SHENZHEN AUTOMOTIVE RES INST BEIJING INST OF TECH (SHENZHEN RES INST OF NAT ENG LAB FOR ELECTRIC VEHICLES)

Patent Information

Application Number
CN202411109995.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-26
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

On the basis of bicycle perception, the existing autonomous driving perception system has problems such as limited perception range, occlusion of field of view and sensor-specific environment failure. The collaborative perception algorithm results in the loss of semantic features of visual images through the unification of features into the BEV form, making it difficult to achieve effective detection effects.

Method used

By timely aligning the three-dimensional object detection results of multiple intelligent perception bodies, evaluating and correcting their confidence, and constructing a BEV spatial confidence map with bird's eye view, combining offline maps for global confidence correction, using interleaving ratio and temperature adjustment The Softmax method performs the results fusion to generate the final detection box and confidence.

Benefits of technology

It achieves more reliable and comprehensive consistent detection results, avoids the weakening of differences in different models of agents and modal data characteristics, and improves the robustness and accuracy of the autonomous driving perception system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027934A_ABST
    Figure CN119027934A_ABST
Patent Text Reader

Abstract

The invention discloses a collaborative confidence fusion sensing method applied to automatic driving, which belongs to the field of automatic driving, eliminates the sensing model difference between different intelligent agents, fully utilizes the advantages of different modal sensor data, and improves the sensing accuracy. The operation of forcibly converting a visual picture which is not good at extracting BEV features into the BEV features in order to realize feature unification is avoided, and a better collaborative perception effect is realized on the basis of giving full play to the advantages of different modal data features, the advantages of learning the features which are good at a specific model, and the collaborative advantage of collaborative perception.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a collaborative confidence fusion perception method applied to autonomous driving. Background Art

[0002] As my country's domestic and international dual circulation creates new opportunities and challenges, China has become the world's largest automobile market. The continuous development of electrification and intelligence continues to drive the growth of supply and demand cycles. The rapid progress of networking and sharing has gradually laid out a grand blueprint for China's automobile industry, among which autonomous driving conforms to the trend of the times. Perception, as one of the extremely important upstream conditions for autonomous driving, has also developed rapidly, but there is still a certain gap with the "unsupervised" level of autonomous driving. The failure of perception's robustness, accuracy, and timeliness to meet the requirements of high-level autonomous driving is one of the key reasons for the gap.

[0003] At present, a lot of work has been done on the basis of single-vehicle perception. Various autonomous driving sensor arrangements and efficient technical routes are emerging in an endless stream. However, a fatal flaw of single-vehicle perception is that the perception range is limited, and there are unavoidable situations such as field of view obstruction and sensor failure in specific environments. In order to solve these problems and deal with such occasions, the vehicle side needs intelligent sensors with different perspectives and different areas, such as road-side intelligent sensors, other vehicle intelligent sensors and the vehicle itself for collaborative perception.

[0004] The essence of collaborative perception is the fusion perception of different intelligent perception bodies. The mainstream technical solution for fusion perception is feature-level fusion. This solution has the problem that different intelligent perception bodies have different feature types (visual features, BEV features, sparse queries), which will make it difficult for different types of features to work in untrained framework models, and even reduce the detection effect. In order to avoid this problem, the main solution in the field is to unify the feature form. Most of the existing vehicle-road collaborative perception solutions are based on the BEV feature form to fuse multimodal features, and BEV features require accurate coordinate positions. This makes it difficult for visual images that are good at semantic understanding but not good at deep understanding to fully demonstrate their effects within the framework, resulting in the loss of semantic features, that is, the weakening of the ability of visual images.

[0005] The existing mainstream collaborative perception algorithm unifies features into the BEV feature form. This is because the voxels of the LiDAR point cloud can detect the three-dimensional target box with high accuracy in the BEV feature form, but this will cause the loss of semantic features of the visual image. Considering this kind of problem, the black box model is skipped, and different intelligent agents and different data types use specially trained model network structures and feature forms to directly perform result-level fusion to achieve collaborative perception. However, a core issue of result-level fusion is how to reasonably evaluate the confidence of the perception results of different intelligent perception agents. Summary of the invention

[0006] In order to solve the technical problems existing in the background technology, the present invention aims to provide a collaborative confidence fusion perception method applied to autonomous driving.

[0007] In order to solve the technical problem, the technical solution of the present invention is:

[0008] S1: performing time alignment processing on the three-dimensional target detection results of multiple original intelligent perception bodies to obtain the detection results of the time-aligned intelligent perception bodies;

[0009] This step is to perform time synchronization among multiple sensors to ensure that the detection results from different sensors can be compared and fused at the same time point, laying the foundation for subsequent processing.

[0010] S2: Use the time-aligned three-dimensional target detection results to evaluate the environmental perception ability of each intelligent sensor, and locally correct the confidence of its detection results to obtain the corrected detection results and confidence; in this step, by analyzing the time-aligned results, the performance of each sensor can be evaluated, and the evaluation results can be used to locally correct the confidence of the detection results to ensure that the detection results of each intelligent sensor are more reliable.

[0011] S3: Based on the corrected detection results and confidences, the offline map and the coordinates of the intelligent sensor, a bird's-eye view BEV spatial confidence map is constructed to generate corresponding confidence information for each area; according to the spatial confidence map and the detection results of other time-aligned intelligent sensors, the detection results of a single intelligent sensor are globally corrected to obtain globally corrected detection results and confidences;

[0012] In this step, the spatial confidence map is constructed by combining the detection results with the static information of the environment (such as offline maps). Based on this confidence map and the results of other sensors, a global confidence correction is performed to obtain a more comprehensive and reliable detection result.

[0013] S4: For overlapping detection frames, the intersection over union (IoU) is used to evaluate the degree of overlap, and similar target frames are aggregated. The temperature-adjusted Softmax method is used to fuse the globally corrected detection result confidences to obtain the final detection frames and confidences.

[0014] In this step, overlapping boxes are evaluated, similar boxes are merged, and the fused confidence is processed to generate the final, optimized detection box and its confidence. The temperature-adjusted Softmax method helps to weigh the contribution of each box and obtain a smoother and more reasonable final confidence value.

[0015] After obtaining the final detection box and confidence level, the system can perform a series of subsequent steps such as target tracking, decision-making, data recording and display based on this information. These steps can continue to promote the system's target management and environmental interaction.

[0016] Further, the step S1 comprises:

[0017] The linear interpolation method is used to align the 3D detection results of different time stamps. The following is the process of time alignment using linear interpolation. This method can simplify complex motion patterns:

[0018]

[0019] is the perceptual agent at timestamp t 2 3D target detection results;

[0020] is the perceptual agent at timestamp t 1 3D target detection results;

[0021] R t is the 3D target detection result of the perception agent after time alignment to timestamp t, t 2 <t 1 <t;

[0022] Linear interpolation method uses the time weight coefficient Multiply by t 1 ,t 2 The detection results of the two timestamps are different The detection result of the t timestamp is time-compensated to obtain the detection result of the compensated t timestamp. This method assumes that the object is at t 2 ,t 1 The movement between time t and time t is linear. The faster the perception result is updated, the better the alignment effect of this method.

[0023] Further, the step S2 specifically includes:

[0024] Before making confidence corrections to the 3D detection results of each intelligent sensor, it is necessary to evaluate the environmental perception capability of each intelligent sensor:

[0025] Ab=f(E,M,S)

[0026] The environmental perception ability A b ∈[0,1), the environmental parameters are E, the perception model is M, and the sensor configuration is S;

[0027] The n groups of 3D target frames detected by n intelligent perception bodies generally contain multiple 3D target frames. In order to weaken the intelligent perception bodies with poor perception ability but overconfidence and enhance the intelligent perception bodies with strong perception ability but relatively conservative, it is necessary to perform local confidence correction on the n groups of 3D target frames respectively to achieve the effect of matching the perception of the entire collaborative system:

[0028]

[0029] The locally corrected single confidence C i ′∈[0,1), the single confidence C before local correction i ∈[0,1), the minimum confidence value C of the 3D target frame group of a single intelligent perception body before local correction min , maximum value C max .

[0030] Further, the step S3 specifically includes:

[0031] The collaborative perception system of the road test needs to build a BEV spatial confidence map based on the offline map and the coordinates of each intelligent sensor, adjust the regional confidence of each intelligent sensor, make an objective evaluation of the detection results of the intelligent sensor, and globally correct and clear the detection results of some intelligent sensors that are difficult to detect;

[0032] The detected 3D target frame is in the part of the area of ​​the intelligent sensor, and whether the intelligent sensor is blocking the area in the middle. Based on this, the confidence of the intelligent sensor's detection result in the area is corrected. In the collaborative relationship between the A, B and C intelligent sensors, That is, it is the parameter for global correction of the detection result of target C.

[0033] Further, the step S4 specifically includes:

[0034] In order to obtain more accurate and consistent detection results, the intersection over union (IoU) is used to measure the overlap of detection results of different sensors. When the detection results of multiple intelligent sensors overlap more in the three-dimensional target frame, that is, the IoU value is higher, it is considered that these detection areas represent the same target object. The specific calculation formula is as follows:

[0035]

[0036] Where A is a detection target frame of a smart sensor, and B is a detection target frame of another smart sensor. The aggregation of multiple detection target frames of a single smart sensor is calculated by the smart sensor itself. This method calculates the target frame aggregation between the three-dimensional target detection results of multiple smart sensors.

[0037] Then, these overlapping areas are aggregated, and the corrected confidence probability coefficient is used to fuse the three-dimensional target box to obtain a more accurate detection result. The method used is a variant of the Softmax function: temperature-adjusted Softmax. When the temperature parameter is 0.2, the influence of high confidence (>0.5) results is enhanced, while the influence of low confidence (<=0.5) results is suppressed. The specific formula is as follows:

[0038]

[0039] Where T is the temperature parameter, T∈(0,∞); by adjusting the temperature T, the sparsity of the weight can be controlled. By adjusting T=0.20, the effect of enhancing the high confidence results and weakening the effect of the low confidence results can be achieved.

[0040] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the above-mentioned collaborative confidence fusion perception methods for autonomous driving is implemented.

[0041] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any one of the above-mentioned collaborative confidence fusion perception methods applied to autonomous driving.

[0042] Compared with the prior art, the advantages of the present invention are:

[0043] The present invention eliminates the model differences between different intelligent agents and the advantages of different modal sensor data, avoids the operation of forcibly converting visual images that are not good at extracting BEV features into BEV features in order to achieve feature unification, and achieves better collaborative perception effects on the basis of giving full play to the advantages of different modal data features, the advantages of a specific model being good at learning features adapted to the model, and the advantages of systematic adjustment of collaborative perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 , a main flow chart of a collaborative confidence fusion perception method applied to autonomous driving according to the present invention;

[0045] Figure 2 , intelligent driving car field of view occlusion and confidence analysis diagram;

[0046] Figure 3 , BEV spatial confidence map. DETAILED DESCRIPTION

[0047] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0048] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0049] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0050] Embodiment 1:

[0051] S1: Time alignment of intelligent sensor detection results

[0052] In three-dimensional target detection, the detection results of different timestamps may have errors due to object motion, sensor delay, etc. In order to perform time alignment, the present invention adopts a linear interpolation method to align the three-dimensional detection results of different timestamps. The following is the process of time alignment using linear interpolation. This method can simplify complex motion patterns.

[0053]

[0054] is the perceptual agent at timestamp t 2 3D target detection results;

[0055] is the perceptual agent at timestamp t 1 3D target detection results;

[0056] R t is the 3D target detection result of the perception agent after time alignment to timestamp t, t 2 <t 1 <t。

[0057] Linear interpolation method uses the time weight coefficient Multiply by t 1 ,t 2 The detection results of the two timestamps are different The detection result of timestamp t is time-compensated to obtain the detection result of timestamp t. This method assumes that the object is at t 2 ,t 1The movement between time t and time t is linear. The faster the perception result is updated, the better the alignment effect of this method.

[0058] S2: Local confidence correction of single intelligent sensor detection results

[0059] Since there is no unified confidence calculation standard in the field of autonomous driving target detection, the confidence of the 3D box detected by each intelligent perception does not match the confidence of the 3D box detected by other intelligent agents. For example, in the same scene, there may be a low-accuracy but aggressive perception agent and a high-accuracy but conservative intelligent perception agent. In this case, there may be a large difference between the 80% confidence of the 3D target box obtained by the two. There may even be a situation where the low-accuracy intelligent perception agent dominates the aggregation process due to overconfidence, reducing the accuracy of the final result. Therefore, in order to achieve accurate and effective collaborative perception, it is necessary to have its own confidence evaluation system to adapt to different environments, different perception models, and different sensor configurations.

[0060] Before making confidence corrections to the 3D detection results of each intelligent sensor, it is necessary to evaluate the environmental perception capability of each intelligent sensor:

[0061] A b =f(E,M,S)

[0062] The environmental perception ability A b ∈[0,1), the environmental parameters are E, the perception model is M, and the sensor configuration is S.

[0063] The n groups of 3D target frames detected by n intelligent perception bodies generally contain multiple 3D target frames. In order to weaken the intelligent perception bodies with poor perception ability but overconfidence and enhance the intelligent perception bodies with strong perception ability but relatively conservative, it is necessary to perform local confidence correction on the n groups of 3D target frames respectively to achieve the effect of matching the perception of the entire collaborative system:

[0064]

[0065] The locally corrected single confidence C i ′∈[0,1), the single confidence C before local correction i ∈[0,1), the minimum confidence value C of the 3D target frame group of a single intelligent perception body before local correction min , maximum value C min .

[0066] S3: BEV spatial confidence map construction and global confidence correction

[0067] After the above corrections, the detection results of each intelligent sensor meet the requirements of participating in the collaborative perception system in this scenario. However, due to the limitations of relatively independent intelligent sensors, such as Figure 2 The number ③ smart driving car has a serious field of vision obstruction due to the large number of vehicles around it at a close distance. Therefore, it is very easy for the car to miss or misdetect pedestrians during road tests. Figure 2 For the intelligent driving car No. ② in the figure, there must be a problem of low confidence in the detection of the area to the right of the driving direction of the intelligent driving car No. ③.

[0068] Therefore, if Figure 3 As shown in the figure, the collaborative perception system of the road test needs to build a BEV spatial confidence map based on the offline map and the coordinates of each intelligent perception body, adjust the regional confidence of each intelligent perception body, make an objective evaluation of the detection results of the intelligent perception body, and make global corrections or even clear the detection results of some intelligent perception bodies that may be difficult to detect. The principle of correction is mainly: the part of the area where the detected 3D target frame is located by the intelligent perception body, whether the intelligent perception body has mid-way occlusion in this area, and the confidence of the detection result of the area. As shown in the left figure, in the collaborative relationship between the A, B and C intelligent perception bodies, It is the global correction parameter for the detection result of target C.

[0069] S4: Target box aggregation and confidence fusion

[0070] For the same detection object, the detection model usually outputs a set of overlapping bounding box candidates. For example, as shown in the figure above, A and B may both detect smart car C and correspond to a 3D target box. In order to obtain more accurate and consistent detection results, the intersection over union (IoU) can be used to measure the overlap of detection results of different perception bodies. When the detection results of multiple smart perception bodies overlap more (that is, the IoU value is high), it can be considered that these detection areas represent the same target object. The specific calculation formula is as follows:

[0071]

[0072] Among them, A is a detection target frame of an intelligent sensor, B is a detection target frame of another intelligent sensor, and the aggregation of multiple detection target frames of a single intelligent sensor is calculated by the intelligent sensor itself.

[0073] These overlapping areas are then aggregated, and the corrected confidence probability coefficient is used to fuse the three-dimensional target frame to obtain a more accurate detection result. The method adopted in the present invention is a variant of the Softmax function: temperature-scaled Softmax. When the temperature parameter is 0.2, this method can enhance the influence of high confidence (>0.5) results and suppress the influence of low confidence (<=0.5) results. The specific formula is as follows:

[0074]

[0075] Wherein, T is a temperature parameter, T∈(0,∞). By adjusting the temperature T, the sparsity of the weight can be controlled. The present invention adjusts T=0.20 to achieve an effect of enhancing high confidence results and weakening the effect of low confidence results.

[0076] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented 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.

[0077] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0078] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0079] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0080] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0081] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A collaborative confidence fusion perception method for autonomous driving, characterized in that: The method comprises: S1: performing time alignment processing on the three-dimensional target detection results of multiple original intelligent perception bodies to obtain the detection results of the time-aligned intelligent perception bodies; S2: using the time-aligned three-dimensional target detection results to evaluate the environmental perception capability of each intelligent sensor, and locally correcting the confidence of its detection results to obtain corrected detection results and confidence; S3: Based on the corrected detection results and confidences, the offline map and the coordinates of the intelligent sensor, a bird's-eye view BEV spatial confidence map is constructed to generate corresponding confidence information for each area; according to the spatial confidence map and the detection results of other time-aligned intelligent sensors, the detection results of a single intelligent sensor are globally corrected to obtain globally corrected detection results and confidences; S4: For overlapping detection frames, the intersection over union (IoU) is used to evaluate the degree of overlap, and similar target frames are aggregated. The temperature-adjusted Softmax method is used to fuse the globally corrected detection result confidences to obtain the final detection frames and confidences.

2. The collaborative confidence fusion perception method for autonomous driving according to claim 1, characterized in that: The step S1 comprises: The linear interpolation method is used to align the 3D detection results of different time stamps. The following is the process of time alignment using linear interpolation. This method can simplify complex motion patterns: is the three-dimensional target detection result of the perception agent at timestamp t2; is the three-dimensional target detection result of the perception agent at timestamp t1; R t is the 3D target detection result of the perception agent after time alignment to timestamp t, t2 <t1<t; Linear interpolation method uses the time weight coefficient Multiply the difference between the detection results of the two timestamps t1 and t2 The detection result of timestamp t is time compensated to obtain the detection result of timestamp t after compensation. This method assumes that the movement of the object between times t2, t1, and t is linear. The faster the perception result is updated, the better the alignment effect of this method.

3. The collaborative confidence fusion perception method for autonomous driving according to claim 1, characterized in that: The step S2 specifically includes: Before making confidence corrections to the 3D detection results of each intelligent sensor, it is necessary to evaluate the environmental perception capability of each intelligent sensor: A b =f(E,M,S) The environmental perception ability A b ∈[0,1), the environmental parameters are E, the perception model is M, and the sensor configuration is S; The n groups of 3D target frames detected by n intelligent perception bodies generally contain multiple 3D target frames. In order to weaken the intelligent perception bodies with poor perception ability but overconfidence and enhance the intelligent perception bodies with strong perception ability but relatively conservative, it is necessary to perform local confidence correction on the n groups of 3D target frames respectively to achieve the effect of matching the perception of the entire collaborative system: The locally corrected single confidence C′ i ∈[0,1), the single confidence C before local correction i ∈[0,1), the minimum confidence value C of the 3D target frame group of a single intelligent perception body before local correction min , maximum value C max .

4. The collaborative confidence fusion perception method for autonomous driving according to claim 1, characterized in that: The step S3 specifically includes: The collaborative perception system of the road test needs to build a BEV spatial confidence map based on the offline map and the coordinates of each intelligent sensor, adjust the regional confidence of each intelligent sensor, make an objective evaluation of the detection results of the intelligent sensor, and globally correct and clear the detection results of some intelligent sensors that are difficult to detect; The detected three-dimensional target frame is located in the part of the area of ​​the intelligent sensor. Whether the intelligent sensor is blocking this area in the middle, the confidence of the detection result of the intelligent sensor in this area is corrected based on this.

5. The collaborative confidence fusion perception method for autonomous driving according to claim 1, characterized in that: The step S4 specifically includes: In order to obtain more accurate and consistent detection results, the intersection over union (IoU) is used to measure the overlap of detection results of different sensors. When the detection results of multiple intelligent sensors overlap more in the three-dimensional target frame, that is, the IoU value is higher, it is considered that these detection areas represent the same target object. The specific calculation formula is as follows: Where A is a detection target frame of a smart sensor, and B is a detection target frame of another smart sensor. The aggregation of multiple detection target frames of a single smart sensor is calculated by the smart sensor itself. This method calculates the target frame aggregation between the three-dimensional target detection results of multiple smart sensors. Then, these overlapping areas are aggregated, and the corrected confidence probability coefficient is used to fuse the three-dimensional target box to obtain a more accurate detection result. The method used is a variant of the Softmax function: temperature-adjusted Softmax. When the temperature parameter is 0.2, the influence of high confidence (>0.5) results is enhanced, while the influence of low confidence (<=0.5) results is suppressed. The specific formula is as follows: Where T is the temperature parameter, T∈(0,∞); by adjusting the temperature T, the sparsity of the weight can be controlled. By adjusting T=0.20, the effect of enhancing the high confidence results and weakening the effect of the low confidence results can be achieved.

6. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a collaborative confidence fusion perception method for autonomous driving as described in any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements a collaborative confidence fusion perception method for autonomous driving as described in any one of claims 1 to 5.

Citation Information

Patent Citations

  • Millimeter-wave radar and vision fusion fog-penetrating target recognition algorithm processing method

    CN113189581A

  • Multi-agent fault diagnosis method and system

    CN114578792A

  • Emotion recognition method and device based on artificial intelligence, computer equipment and medium

    CN114974310A

  • Three-dimensional environment target detection method based on multi-sensor fusion

    CN115049821A

  • Multi-source traffic data processing method based on confidence evaluation

    CN115909742A

Cited By

  • Intelligent collaborative awareness-oriented instance-level semantic simplification interaction method

    CN120047914A

  • An instance-level semantic interaction method for intelligent collaborative perception

    CN120047914B

  • A unified neural feature space construction method for open heterogeneous multi-agent collaboration

    CN122819298A