Target enhanced fusion perception method, system, device and medium

Abstract

The invention relates to the field of vehicle infrastructure cooperation, and discloses a target enhanced fusion perception method and system, computer equipment and a medium, and the method comprises the steps: receiving a plurality of sub-detection models adaptive to different environments; in response to receiving the environmental data, determining an environmental condition; matching a first sub-detection model for target detection based on the current environmental condition; and performing target detection on sensing data based on the first sub-detection model to obtain a target sensing result. According to the scheme provided by the invention, the optimal model can be automatically selected for target sensing according to different scenes, the performance reduction of a single model in a complex scene is avoided, the scene adaptability of the model is improved, and a personalized roadside sensing system with higher performance is obtained.

Description

Technical Field

[0001] The present invention relates to the field of vehicle-road collaboration, and in particular to a target enhanced fusion perception method, system, device and medium. Background Art

[0002] To better capture information about road users, multiple sensors of various types are typically installed and deployed along the roadside, such as lidar, millimeter-wave radar, and cameras. Target detection is performed using the data collected by these sensors, enabling effective collaboration among drivers, vehicles, and roads.

[0003] Current object detection methods simply fuse data from multiple sensors and then use perception models to identify the fused features. However, these methods fail to account for the limitations of sensors in varying weather conditions. For example, inclement weather like rain, fog, and snow can affect lidar performance, resulting in noisy data. Video camera performance degrades significantly in poor lighting conditions. Each sensor technology has its own unique advantages and limitations. Sensor fusion can address certain deficiencies or shortcomings, but the range and combination of factors, such as road conditions, traffic flow, and weather, vary significantly. For example, highways are significantly affected by rain or fog, non-arterial roads may have poor lighting conditions, and scenes with a high number of pedestrians require a greater reliance on cameras.

[0004] Therefore, simply fusing multiple sensor data and using a unified model or fusion strategy cannot take into account the individual characteristics of each road and it is difficult to maximize the perception effects of different roads. Summary of the Invention

[0005] In view of this, in order to overcome at least one aspect of the above-mentioned problems, an embodiment of the present invention proposes a method for enhanced target fusion perception, comprising performing the following steps on the roadside: Receive multiple sub-detection models adapted to different environments; determining an environmental condition in response to receiving the environmental data; matching a first sub-detection model for target detection based on current environmental conditions; Perform target detection on the perception data based on the first sub-detection model to obtain a target perception result.

[0006] In some embodiments, the following steps are further performed on the cloud: Obtain sensor data collected from the roadside; Training basic detection models based on sensor data; Performing data enhancement based on the sensor data and multiple environmental conditions to obtain multiple sub-training sets; A plurality of sub-detection models are obtained based on the plurality of sub-training sets, the sensor data, and the basic detection model.

[0007] In some embodiments, obtaining a plurality of sub-detection models based on a plurality of the sub-training sets, the sensor data, and the basic detection model further includes: Obtaining target features based on the basic detection model and the training set; Obtaining sub-target features based on the sub-detection model to be trained and the sub-training set; Calculating a loss function based on the target feature and the sub-target feature; The basic detection model is trained based on the loss function to obtain the sub-detection model.

[0008] In some embodiments, calculating the loss function based on the target feature and the sub-target feature further includes: Obtaining a first loss based on a similarity between the target feature and the sub-target feature; Obtaining a second loss based on the distance between the target feature and the sub-target feature of the same category and the distance between the target feature and the sub-target feature of different categories; The loss function is calculated based on the first loss and the second loss.

[0009] In some embodiments, calculating the loss function based on the first loss and the second loss further includes: Get the preset weight; The first loss is weighted based on the weight, and a loss function is obtained based on the second loss and the weighted first loss.

[0010] In some embodiments, matching a first sub-detection model for target detection based on current environmental conditions further includes: Detecting the environmental data based on an environmental monitoring model to obtain environmental conditions, where the environmental conditions include any one or more of current weather, road conditions, and lighting conditions; The first sub-detection model is matched based on the environmental condition.

[0011] In some embodiments, performing data augmentation based on the sensor data and multiple environmental conditions to obtain multiple training subsets further includes: Determine noise categories based on environmental conditions; Data enhancement is performed on the training set based on the noise category to obtain the sub-training set.

[0012] Based on the same inventive concept, according to another aspect of the present invention, an embodiment of the present invention further provides a target enhanced fusion perception system, including: A receiving module configured to receive multiple sub-detection models adapted to different environments; a determination module configured to determine an environmental condition in response to receiving the environmental data; a selection module configured to match a first sub-detection model for target detection based on current environmental conditions; The detection module is configured to perform target detection on the perception data based on the first sub-detection model to obtain a target perception result.

[0013] Based on the same inventive concept, according to another aspect of the present invention, an embodiment of the present invention further provides a computer device, including: at least one processor; and A memory storing a computer program that can be run on the processor, wherein the processor executes the steps of any one of the target enhanced fusion perception methods described above when executing the program.

[0014] Based on the same inventive concept, according to another aspect of the present invention, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the target enhanced fusion perception methods described above are performed.

[0015] The present invention has one of the following beneficial technical effects: the solution proposed in the present invention obtains a basic model by training basic data, and then performs targeted data enhancement on the basis of the basic data to obtain a sub-training set, and uses the sub-training set to train a sub-detection model. In this way, the optimal model can be automatically selected for target perception in different scenarios, avoiding the performance degradation of a single model in complex scenarios, improving the scenario adaptability of the model, and obtaining a more personalized roadside perception system with higher performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A schematic diagram of a flow chart of a target enhanced fusion perception method provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a target perception and recognition system provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention; Figure 4 A schematic structural diagram of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0019] It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are for distinguishing two non-identical entities with the same name or non-identical parameters. It can be seen that "first" and "second" are only for the convenience of expression and should not be understood as limitations on the embodiments of the present invention. Subsequent embodiments will not explain this one by one.

[0020] According to one aspect of the present invention, an embodiment of the present invention proposes a target enhanced fusion perception method, such as Figure 1 As shown, it may include performing the following steps on the roadside: S1, receives multiple sub-detection models adapted to different environments; S2, in response to receiving the environmental data, determining an environmental condition; S3, matching the first sub-detection model for target detection based on the current environmental conditions; S4: Perform target detection on the perception data based on the first sub-detection model to obtain a target perception result.

[0021] The solution proposed in the present invention obtains a basic model by training basic data, and then performs targeted data enhancement on the basic data to obtain a sub-training set. The sub-training set is used to train a sub-detection model. In this way, the optimal model can be automatically selected for target perception in different scenarios, avoiding the performance degradation of a single model in complex scenarios, improving the scenario adaptability of the model, and obtaining a more personalized roadside perception system with higher performance.

[0022] In some embodiments, the following steps are further performed on the cloud: Obtain sensor data collected from the roadside; Training basic detection models based on sensor data; Performing data enhancement based on the sensor data and multiple environmental conditions to obtain multiple sub-training sets; A plurality of sub-detection models are obtained based on the plurality of sub-training sets, the sensor data, and the basic detection model.

[0023] Specifically, a base model can be trained in the cloud based on data collected by multiple roadside sensors. This base model can achieve relatively good perception performance in most scenarios or on most roads. Targeted data augmentation is then performed on this base data to generate sub-training sets. Each sub-training set corresponds to a specific environment. For example, for scenes with poor lighting conditions, a sub-training set containing image data from various lighting conditions can be generated. In this way, the sub-training model generated from each sub-training set can achieve even better detection performance for specific scenarios, thereby improving the model's adaptability to different scenarios.

[0024] In some embodiments, obtaining a plurality of sub-detection models based on a plurality of the sub-training sets, the sensor data, and the basic detection model further includes: Obtaining target features based on the basic detection model and the training set; Obtaining sub-target features based on the sub-detection model to be trained and the sub-training set; Calculating a loss function based on the target feature and the sub-target feature; The basic detection model is trained based on the loss function to obtain the sub-detection model.

[0025] Specifically, after obtaining the basic detection model, the target features of the samples in the training set can be obtained by inputting them into the basic detection model. Similarly, the sub-target features of the corresponding samples can be obtained by inputting them into the sub-detection model. When training the sub-detection model, the loss function can be calculated based on the target features and sub-target features, thereby training the sub-detection model based on the loss function.

[0026] For example, for the training set D i The sample k in the sub-training set D is a i The corresponding sample k', sample k passes through the basic detection model M base Then we get the feature f i (k)=M base (D i (k)), sample k' passes through the sub-detection model M i Then we get the feature f a i (k')=M i (D a i (k')), using the feature f a i (k') and feature f i (k) Calculate the loss function and train the sub-model based on the loss function to ultimately obtain a sub-detection model that is more suitable for the corresponding scene.

[0027] In some embodiments, calculating the loss function based on the target feature and the sub-target feature further includes: Obtaining a first loss based on a similarity between the target feature and the sub-target feature; Obtaining a second loss based on the distance between the target feature and the sub-target feature of the same category and the distance between the target feature and the sub-target feature of different categories; The loss function is calculated based on the first loss and the second loss.

[0028] Specifically, the first loss can be obtained by the similarity between the target feature and the sub-target feature, that is, the target feature obtained after the samples in the training set pass through the basic detection model, and the sub-target feature obtained after the corresponding samples in the sub-training set pass through the sub-detection model, the two should be as close as possible, that is, after data enhancement, the data distribution will not change. In this way, even if the data in the training set is enhanced, the detection performance of the model will not change when the data in the sub-training set is tested. At the same time, the second loss is obtained based on the distance between the target feature and the sub-target feature of the same category, that is, after data enhancement, the distance between samples of the same category is still smaller than the distance between samples of different categories, that is, after data enhancement, the sample category will not be changed, that is, when different data enhancements are performed on the samples, while the adaptability of the model to different environments is improved, the performance will not decrease.

[0029] Based on this objective, the final loss function is obtained based on the first loss and the second loss.

[0030] Among them, when the first loss is obtained by the similarity between the target feature and the sub-target feature, it can be quantified based on the L2 distance: L distill (k)=|M base (D i (k)), M i (D a i (k'))| 2 2 When calculating the first loss, you can calculate the L2 distance of each sample, or you can perform sampling calculations. For example, if there are 50 samples in the training set and 50 samples in the sub-training set, you can calculate 50 L2 distances, or you can sample and calculate from the 50 samples, for example, sampling 25 samples.

[0031] Among them, the second loss is obtained based on the distance between the target feature and the sub-target feature of the same category, which can also be quantified based on the L2 distance: L constrative (f i (k), fi a (k'), f i a (j'))= max(0,|f a i (k')- f i (k)| 2 2-|f a i (j')- f a i (k')| 2 2+ε) k and j represent the training set D i For any two samples of different categories, k' and j' represent the corresponding samples in the sub-training set, f i (k), f a i (k') and f a i (j') is the use of the basic model M base and sub-detection model M i The three features obtained, among which f i (k) Using the basic model M base The target feature obtained; f a i (k') and f a i (j') is the sample after enhancement, using the sub-detection model M i The obtained sub-goal features. ε is a preset parameter.

[0032] Similarly, when calculating the second loss, you can calculate the L2 distance of each sample, or you can perform sampling calculations. For example, if there are 50 samples in the training set and 50 samples in the sub-training set, you can calculate 50 L2 distances, or you can sample and calculate from the 50 samples, for example, sampling 25 samples.

[0033] In some embodiments, calculating the loss function based on the first loss and the second loss further includes: Get the preset weight; The first loss is weighted based on the weight, and a loss function is obtained based on the second loss and the weighted first loss.

[0034] Specifically, the weights can be set by empirical values. After obtaining the preset weights, weighted calculations are performed. For example, the weights are used to perform weighted calculations on the first loss, that is: Loss Total =w d *L distill +L constrative where w d It is a weight parameter that can balance the contribution of the first loss to the total loss, making the two more balanced.

[0035] In some embodiments, matching a first sub-detection model for target detection based on current environmental conditions further includes: Detecting the environmental data based on an environmental monitoring model to obtain environmental conditions, where the environmental conditions include any one or more of current weather, road conditions, and lighting conditions; The first sub-detection model is matched based on the environmental condition.

[0036] Specifically, the distribution of image features varies significantly in different environments. Sensors can be used to detect environmental data such as the current weather, lighting conditions, or the current road complexity. The environmental data can then be detected using an environmental monitoring model to obtain the current environmental conditions. Finally, a suitable model can be selected for target detection based on the current environmental conditions. For example, if the current lighting conditions are poor, a sub-detection model trained with a sub-training set for the corresponding scene can be used for target recognition. In some embodiments, if there is no corresponding sub-model, the basic detection model can be quickly switched to for detection. This dynamic binding of environmental perception and detection models can significantly improve the robustness and efficiency of the system in complex scenarios.

[0037] In some embodiments, constructing a training set based on sensor data further includes: Data collection based on multiple sensors; The data of various sensors collected at the same time are taken as a sample, and the training set is constructed using multiple samples.

[0038] Specifically, sensors can include image sensors, lidar, millimeter-wave radar, and other sensors. Each type of sensor has its own unique advantages. For example, lidar offers very high spatial resolution, enabling highly accurate detection of an object's shape, size, and distance. Image sensors provide rich color information, aiding in object and scene recognition. Millimeter-wave radar is particularly sensitive to metal objects, so using multiple sensors to capture the target helps extract a more comprehensive feature representation. After using multiple sensors to collect data, the multiple data points collected simultaneously are combined as a single sample, resulting in a training set consisting of multiple data samples.

[0039] In some embodiments, a basic detection model is obtained based on the training set. Specifically, the raw data in the training set can be first converted into a matrix, and then the matrix can be converted into a vector through an explicit projection. The corresponding vectors are then processed in parallel by multiple encoders, thereby simultaneously extracting rich target features from different sensor data. For example, image features, such as image BEV features, can be obtained by performing feature extraction on image data collected by an image sensor through an image encoder. LiDAR point cloud data collected by a LiDAR can be subjected to feature extraction through a first point cloud encoder to obtain LiDAR point cloud features, such as LiDAR point cloud BEV features. Millimeter-wave radar point cloud data collected by a millimeter-wave radar can be subjected to feature extraction through a second point cloud encoder to obtain millimeter-wave radar point cloud features, such as millimeter-wave radar point cloud BEV features. In some embodiments, the image encoder, the first point cloud encoder, and the second point cloud encoder can be convolutional neural networks or Transformers.

[0040] Then use the following formula to perform feature fusion: F Fused =[ BEV c , BEV l , BEV r ]= [BEV(Enc(c)) , BEV(Enc(l)) , BEV(Enc(r))] Where BEV represents the BEV feature extraction network, Enc represents the encoder, c, l, and r represent image, lidar point cloud, and millimeter-wave radar point cloud data, respectively.

[0041] After obtaining the fusion features, the detection model is trained using target detection loss, such as cross entropy and regression loss. For example, the loss function can be: Loss (GT, Det( F Fused )) Among them, GT represents ground truth, that is, true value; Det represents the detection model, F Fused It is a multi-sensor fusion feature.

[0042] In some embodiments, performing multiple data enhancements corresponding to different environments on the sensor data in the training set to obtain multiple sub-training sets further includes: Determine noise categories based on environmental conditions; Data enhancement is performed on the training set based on the noise category to obtain the sub-training set.

[0043] Specifically, the training set can be enhanced with different data corresponding to different environments according to different scenarios. For example, for scenes with poor lighting conditions, data under multiple lighting conditions can be generated, and sub-training sets can be constructed using the enhanced data. Or for rainy and foggy scenes, data enhancement can be performed by adding rain and fog noise. In this way, targeted data enhancement is performed for each scene, and then the targeted enhanced training set is used for training. This can improve the generalization ability of the model, make it more adaptable to the scene, and have a higher upper limit on target detection performance.

[0044] The solution proposed in the present invention obtains a basic model by training basic data, and then performs targeted data enhancement on the basic data to obtain a sub-training set. The sub-training set is used to train a sub-detection model. In this way, the optimal model can be automatically selected for target perception in different scenarios, avoiding the performance degradation of a single model in complex scenarios, improving the scenario adaptability of the model, and obtaining a more personalized roadside perception system with higher performance.

[0045] Based on the same inventive concept, according to another aspect of the present invention, an embodiment of the present invention further provides a target enhanced fusion perception system 400, such as Figure 2 Shown, including: A receiving module 401 is configured to receive multiple sub-detection models adapted to different environments; a determination module 402 configured to determine an environmental condition in response to receiving environmental data; A selection module 403 is configured to match a first sub-detection model for target detection based on current environmental conditions; The detection module 404 is configured to perform target detection on the perception data based on the first sub-detection model to obtain a target perception result.

[0046] In some embodiments, the training module is further configured to: Obtain sensor data collected from the roadside; Training basic detection models based on sensor data; Performing data enhancement based on the sensor data and multiple environmental conditions to obtain multiple sub-training sets; A plurality of sub-detection models are obtained based on the plurality of sub-training sets, the sensor data, and the basic detection model.

[0047] In some embodiments, obtaining a plurality of sub-detection models based on a plurality of the sub-training sets, the sensor data, and the basic detection model further includes: Obtaining target features based on the basic detection model and the training set; Obtaining sub-target features based on the sub-detection model to be trained and the sub-training set; Calculating a loss function based on the target feature and the sub-target feature; The basic detection model is trained based on the loss function to obtain the sub-detection model.

[0048] In some embodiments, calculating the loss function based on the target feature and the sub-target feature further includes: Obtaining a first loss based on a similarity between the target feature and the sub-target feature; Obtaining a second loss based on the distance between the target feature and the sub-target feature of the same category and the distance between the target feature and the sub-target feature of different categories; The loss function is calculated based on the first loss and the second loss.

[0049] In some embodiments, calculating the loss function based on the first loss and the second loss further includes: Get the preset weight; The first loss is weighted based on the weight, and a loss function is obtained based on the second loss and the weighted first loss.

[0050] In some embodiments, matching a first sub-detection model for target detection based on current environmental conditions further includes: Detecting the environmental data based on an environmental monitoring model to obtain environmental conditions, where the environmental conditions include any one or more of current weather, road conditions, and lighting conditions; The first sub-detection model is matched based on the environmental condition.

[0051] In some embodiments, performing data augmentation based on the sensor data and multiple environmental conditions to obtain multiple training subsets further includes: Determine noise categories based on environmental conditions; Data enhancement is performed on the training set based on the noise category to obtain the sub-training set.

[0052] The solution proposed in the present invention obtains a basic model by training basic data, and then performs targeted data enhancement on the basic data to obtain a sub-training set. The sub-training set is used to train a sub-detection model. In this way, the optimal model can be automatically selected for target perception according to different scenarios or sensor conditions, avoiding the performance degradation of a single model in complex scenarios, improving the scenario adaptability of the model, and obtaining a more personalized roadside perception system with higher performance.

[0053] Based on the same inventive concept, according to another aspect of the present invention, Figure 3 As shown, an embodiment of the present invention further provides a computer device 501, including: at least one processor 520; and The memory 510 stores a computer program 511 that can be run on the processor. When the processor 520 executes the program, the processor 520 performs the steps of any of the above target enhanced fusion perception methods.

[0054] Based on the same inventive concept, according to another aspect of the present invention, Figure 4 As shown, an embodiment of the present invention further provides a computer-readable storage medium 601, which stores a computer program 610. When the computer program 610 is executed by a processor, it performs the steps of any of the above target enhanced fusion perception methods.

[0055] Finally, it should be noted that a person skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0056] Furthermore, it should be appreciated that the computer-readable storage media (eg, memory) herein can be either volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory.

[0057] It will also be appreciated by those skilled in the art that the various exemplary logic blocks, modules, circuits and algorithmic steps described in conjunction with the disclosure herein can be implemented as electronic hardware, computer software or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, a general description has been given of the functions of various schematic components, blocks, modules, circuits and steps. Whether this function is implemented as software or hardware depends on specific applications and the design constraints imposed on the entire system. Those skilled in the art can implement the function in various ways for each specific application, but this implementation decision should not be interpreted as causing a departure from the disclosed scope of the embodiments of the present invention.

[0058] The above are exemplary embodiments disclosed in the present invention, but it should be noted that various changes and modifications may be made without departing from the scope of the embodiments disclosed in the claims. The functions, steps and / or actions of the method claims according to the disclosed embodiments described herein do not need to be performed in any particular order. In addition, although the elements disclosed in the embodiments of the present invention may be described or required in individual form, they may also be understood as multiple unless expressly limited to the singular.

[0059] It should be understood that, as used herein, the singular forms "a" and "an" are intended to include the plural forms as well, unless the context clearly supports an exception. It should also be understood that, as used herein, "and / or" is intended to include any and all possible combinations of one or more of the associated listed items.

[0060] The serial numbers of the embodiments disclosed in the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0061] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by a program instructing the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk, or an optical disk, etc.

[0062] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to limit the scope of the disclosure of the present invention (including the claims) to these examples. Within the spirit of the present invention, the technical features of the above embodiments or different embodiments may be combined, and many other variations exist in different aspects of the above embodiments, which are not provided in detail for the sake of clarity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A target enhanced fusion perception method, characterized in that: It involves performing the following steps on the roadside: Receive multiple sub-detection models adapted to different environments; determining an environmental condition in response to receiving the environmental data; matching a first sub-detection model for target detection based on current environmental conditions; Perform target detection on the perception data based on the first sub-detection model to obtain a target perception result.

2. The method according to claim 1, wherein It also includes the following steps on the cloud: Obtain sensor data collected from the roadside; Training basic detection models based on sensor data; Performing data enhancement based on the sensor data and multiple environmental conditions to obtain multiple sub-training sets; A plurality of sub-detection models are obtained based on the plurality of sub-training sets, the sensor data, and the basic detection model.

3. The method according to claim 2, wherein The method further comprises: obtaining a plurality of sub-detection models based on the plurality of sub-training sets, the sensor data, and the basic detection model; Obtaining target features based on the basic detection model and the training set; Obtaining sub-target features based on the sub-detection model to be trained and the sub-training set; Calculating a loss function based on the target feature and the sub-target feature; The basic detection model is trained based on the loss function to obtain the sub-detection model.

4. The method according to claim 3, wherein Calculating a loss function based on the target feature and the sub-target feature further includes: Obtaining a first loss based on a similarity between the target feature and the sub-target feature; Obtaining a second loss based on the distance between the target feature and the sub-target feature of the same category and the distance between the target feature and the sub-target feature of different categories; The loss function is calculated based on the first loss and the second loss.

5. The method according to claim 4, wherein Calculating the loss function based on the first loss and the second loss further includes: Get the preset weight; The first loss is weighted based on the weight, and a loss function is obtained based on the second loss and the weighted first loss.

6. The method according to claim 1, wherein Matching a first sub-detection model for target detection based on current environmental conditions further includes: Detecting the environmental data based on an environmental monitoring model to obtain environmental conditions, where the environmental conditions include any one or more of current weather, road conditions, and lighting conditions; The first sub-detection model is matched based on the environmental condition.

7. The method according to claim 2, wherein Performing data enhancement based on the sensor data and multiple environmental conditions to obtain multiple sub-training sets further includes: Determine noise categories based on environmental conditions; Data enhancement is performed on the training set based on the noise category to obtain the sub-training set.

8. A target enhanced fusion perception system, characterized in that: include: A receiving module configured to receive multiple sub-detection models adapted to different environments; a determination module configured to determine an environmental condition in response to receiving the environmental data; a selection module configured to match a first sub-detection model for target detection based on current environmental conditions; The detection module is configured to perform target detection on the perception data based on the first sub-detection model to obtain a target perception result.

9. A computer device comprising: at least one processor; as well as A memory storing a computer program executable on the processor, wherein the processor executes the steps of the method according to any one of claims 1 to 7 when executing the program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are performed.

Images