A dynamic adaptive BEV perception multi-scale feature fusion method

Abstract

The application relates to the field of BEV-based intelligent driving, in particular to a dynamic self-adaptive BEV perception multi-scale feature fusion method.The scheme comprises the following steps: data acquisition, multiple wide-angle cameras arranged around a vehicle body are used to synchronously collect multi-view image data of 360 degrees around the vehicle; feature extraction, ResNet or Swin Transformer is used to extract multi-scale feature maps F from the collected image data; dynamic self-adaptive BEV query generation; pyramid multi-scale feature fusion; space-time fusion and BEV representation generation; and finally, the final BEV feature map after the space-time fusion is decoded into specific perception results for output.The application is suitable for BEV-based intelligent driving.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving based on BEV (Bird's-Eye-View), specifically to a multi-scale feature fusion method for dynamic adaptive BEV perception. Background Technology

[0002] Intelligent driving is a core development direction for the automotive industry. Its key lies in accurately perceiving the surrounding environment through sensors to provide a basis for vehicle decision-making, planning, and control. Environmental perception technology integrates data from multiple sensors, such as LiDAR and cameras, to build a three-dimensional understanding of the driving scenario.

[0003] In recent years, BEV perception has become a key technological paradigm in the field of environmental perception. The core of this technology lies in converting independent perspective images captured by multiple onboard cameras into a unified bird's-eye view centered on the vehicle. However, existing BEV-based static perception technologies still have the following shortcomings and deficiencies in practical applications, especially when dealing with complex and changing road scenarios:

[0004] Insufficient scene adaptability and unreasonable allocation of computing resources: Most existing mainstream BEV perception models (such as BEVFormer) use fixed-resolution BEV query grids to represent physical space (e.g., using a 200x200 grid to cover a specific area). This static and uniform grid division method lacks adaptability to scene complexity. In scenarios with simple road structures (such as straight urban roads), overly dense query grids can cause a lot of computational redundancy; while in scenarios with complex road topology (such as interchange ramp weaving areas and construction zones), fixed query granularity may not be fine enough to accurately capture key geometric details such as lane curvature and guide line boundaries, leading to a decrease in perception accuracy.

[0005] In addition, there are problems such as low coupling efficiency of multi-granularity features, easy loss of scene details, and poor perception robustness under complex and extreme conditions. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a multi-scale feature fusion method for dynamic adaptive BEV perception, which achieves high-precision and high-robust output of BEV static perception results.

[0007] The present invention achieves the above objectives by adopting the following technical solution: the present invention provides a multi-scale feature fusion method for dynamic adaptive BEV perception, comprising:

[0008] S1, Data Acquisition;

[0009] Multiple wide-angle cameras deployed around the vehicle body simultaneously collect 360° multi-view image data of the vehicle's surroundings;

[0010] S2, Feature Extraction;

[0011] Extract multi-scale feature maps F from the acquired image data using ResNet or Swing Transformer;

[0012] S3. Generate dynamic adaptive BEV query;

[0013] S4, Multi-scale feature fusion of pyramids;

[0014] S5, Spatiotemporal Fusion and BEV Representation Generation;

[0015] S6. Decode the final BEV feature map after spatiotemporal fusion into specific perception results and output them.

[0016] Furthermore, step S3 specifically includes:

[0017] Taking a multi-scale feature map F and a set of initial queries Q as input, the output is a query coordinate that is dynamically adjusted in resolution according to the current scene complexity. The specific process is as follows:

[0018] S301, Scene Complexity Analysis;

[0019] First, a lightweight semantic segmentation network is used to analyze the multi-scale feature map F, extract the lane line curvature variance and obstacle distribution entropy features, and then linearly combine the lane line curvature variance and obstacle distribution entropy features to generate a scene complexity heatmap H corresponding to the BEV plane.

[0020] S302, Dynamic resolution adjustment;

[0021] Design a lightweight gated network based on an MLP, taking a heatmap H and an initial query Q as input, and calculate a resolution scaling factor for each grid location as follows:

[0022] ;

[0023] In the formula, This represents the resolution scaling factor. This indicates a feature concatenation operation. This represents the eigenvalue or eigenvector corresponding to the heatmap H at grid position (i, j). Let Q represent the initial query at grid position (i, j). Indicates the bias term. This represents the weight matrix of the last linear transformation layer in the gated network.

[0024] S303, Dynamic coordinate mapping;

[0025] Based on the calculated scaling factor, the original uniform grid coordinates are nonlinearly transformed using bilinear interpolation to generate adaptive query coordinates, as follows:

[0026] ;

[0027] In the formula, , These represent the preset maximum offsets in the x and y directions, respectively, defining the maximum distance each grid point can move. , These represent the x and y coordinates of the original uniform grid, respectively. , These represent the x and y coordinates of the query coordinates, respectively.

[0028] Furthermore, step S3 specifically includes:

[0029] S304, Semantic Constraints of K-means++ Clustering;

[0030] Multi-view image features Divided into K semantic clusters The optimized objective function considers not only geometric distance but also KL divergence. Semantic consistency is introduced as a constraint, as follows:

[0031] ;

[0032] In the formula, Denotes the set of K cluster centers. Indicates the first The center vector of each cluster This indicates the allocation of N data points. This represents the index of the cluster to which the nth data point is assigned. This represents the feature vector of the nth multi-view image. Indicates assignment to feature The center point of that cluster, This represents a weighting coefficient. This represents a function used to obtain the semantic probability distribution of a given feature vector.

[0033] S305, Relevance Weights and Local Attention;

[0034] Each BEV query First, calculations are performed with each cluster center. Association weight This allows us to locate highly correlated clusters, as follows:

[0035] ;

[0036] In the formula, For temperature coefficient, Let the vector represent the center vector of the j-th cluster. Let represent the i-th BEV query vector, and T represent the transpose.

[0037] Furthermore, step S4 specifically includes:

[0038] Taking a multi-scale feature map F as input, the output is a feature map F_fused after fusing multi-granularity information. The specific process is as follows:

[0039] S401, Parallel extraction of multi-scale features;

[0040] The input multi-scale feature map F is divided into three groups along the channel dimension and processed in parallel through heterogeneous convolutional branches with different receptive fields.

[0041] High-resolution branches, using 3 × 3 small convolution kernels, capture subpixel-level lane line edge geometric details;

[0042] In the mesoscopic branch, a 5 × 5 dilated convolution is used to model the spatial continuity of roadside curbs and medians;

[0043] The low-resolution branch uses a 7 × 7 large convolution kernel to analyze the macroscopic topological relationship of the traffic flow divider at the intersection.

[0044] S402, Dynamic Channel Attention Weighting;

[0045] The multi-scale features processed in parallel are concatenated to obtain concatenated features, which are then fed into the SE module. The SE module learns a dynamic weight for each feature channel through global average pooling and a two-layer fully connected network, as follows:

[0046] ;

[0047] In the formula, The dynamic weights are represented by σ and δ, which represent the Sigmoid and ReLU activation functions, respectively. Indicates splicing characteristics, , These represent the weight matrices of the two fully connected layers in the SE module.

[0048] Furthermore, step S4 specifically includes introducing temporal continuity constraints based on hidden Markov models, specifically including:

[0049] BEV feature sequence The temporal evolution model is a state transition process, which takes into account the vehicle's own speed. The method is as follows:

[0050] ;

[0051] In the formula, This represents the bird's-eye view features or state at the current time point t. This represents the bird's-eye view features or state at the previous time point t-1. Indicates a Gaussian distribution. This represents speed multiplied by time, used to estimate the spatial displacement of the BEV feature due to the vehicle's own motion during this time period. This represents the covariance matrix of the normal distribution.

[0052] By using Bayesian theory, the observation information of the current frame is fused. and time-series priors based on historical information The query results for the current frame The optimization can be performed as follows:

[0053] ;

[0054] In the formula, This represents the observation information at the current time point t, that is, the multi-view image data collected by the vehicle-mounted camera at time t. It is a conditional probability, representing the probability under given observation conditions. The next state is The probability of.

[0055] Furthermore, step S5 specifically includes:

[0056] Through a spatial cross-attention mechanism, the dynamic query grid is used as the Query and interacts with the fused feature map F_fused as the Key and Value to aggregate spatial information and generate the BEV feature BEV_t at the current time. Then, the BEV feature BEV_t-1 at the previous time is read, the coordinates are aligned according to the vehicle's own motion, and then fused with BEV_t through a temporal self-attention mechanism.

[0057] Furthermore, step S1 specifically includes:

[0058] Multiple wide-angle cameras deployed around the vehicle body simultaneously collect 360° multi-view image data of the vehicle's surroundings.

[0059] Furthermore, step S2 specifically includes:

[0060] Multiscale feature maps F are extracted from the acquired image data using ResNet or Swing Transformer.

[0061] The beneficial effects of this invention are:

[0062] This invention innovatively integrates a semantic density-driven BEV query (Queries) elastic resolution mechanism and a heterogeneous feature fusion architecture based on pyramid multi-scale attention. It acquires data through an onboard multi-view camera, generates a non-uniform BEV query grid that matches the scene complexity via a dynamic query module, and constructs a feature representation that takes into account both global and detailed aspects using a multi-scale fusion module. Finally, it achieves efficient interaction and fusion of the two within a spatiotemporal attention framework, outputting high-precision and highly robust BEV static perception results. Attached Figure Description

[0063] Figure 1 This is a flowchart of a multi-scale feature fusion method for dynamic adaptive BEV perception provided in an embodiment of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0065] This invention provides a multi-scale feature fusion method for dynamic adaptive BEV perception, such as... Figure 1 As shown, it specifically includes:

[0066] S1, Data Acquisition;

[0067] Multiple wide-angle cameras deployed around the vehicle body simultaneously collect 360° multi-view image data of the vehicle's surroundings.

[0068] S2, Feature Extraction;

[0069] Multiscale feature maps F are extracted from the acquired image data using ResNet or Swing Transformer.

[0070] S3. Generate dynamic adaptive BEV query;

[0071] Taking a multi-scale feature map F and a set of initial queries Q as input, the output is a query coordinate that is dynamically adjusted in resolution according to the current scene complexity. The specific process is as follows:

[0072] S301, Scene Complexity Analysis;

[0073] First, a lightweight semantic segmentation network is used to analyze the multi-scale feature map F, extract the lane line curvature variance and obstacle distribution entropy features, and then linearly combine the lane line curvature variance and obstacle distribution entropy features to generate a scene complexity heatmap H corresponding to the BEV plane.

[0074] S302, Dynamic resolution adjustment;

[0075] Design a lightweight gated network based on an MLP, taking a heatmap H and an initial query Q as input, and calculate a resolution scaling factor for each grid location as follows:

[0076] ;

[0077] In the formula, This represents the resolution scaling factor. This indicates a feature concatenation operation. This represents the eigenvalue or eigenvector corresponding to the heatmap H at grid position (i, j). Let Q represent the initial query at grid position (i, j). This represents the bias term, which provides additional learnable parameters to the model, increasing its fitting ability and making the output less dependent on the weighted sum of the inputs. It allows for overall shifting. This represents the weight matrix of the last linear transformation layer in the gated network. Its function is to perform a weighted summation of the high-dimensional features extracted by the MLP, mapping (or transforming) them into a single value. This value is then fed into the Sigmoid function, ultimately generating a resolution scaling factor between 0 and 1.

[0078] S303, Dynamic coordinate mapping;

[0079] Based on the calculated scaling factor, the original uniform grid coordinates are nonlinearly transformed using bilinear interpolation to generate adaptive query coordinates, achieving the effect of densification in complex regions and sparseness in simple regions. The method is as follows:

[0080] ;

[0081] In the formula, , These represent the preset maximum offsets in the x and y directions, respectively, defining the maximum distance each grid point can move. , These represent the x and y coordinates of the original uniform grid, respectively. , These represent the x and y coordinates of the query coordinates, respectively.

[0082] To further improve query efficiency and generalization ability for long-tail scenarios, this invention introduces a sparse attention mechanism guided by feature clustering.

[0083] Semantically Constrained K-means++ Clustering: Integrating Multi-View Image Features Divided into K semantic clusters Its optimization objective function considers not only geometric distance but also KL divergence. Semantic consistency is introduced as a constraint:

[0084] ;

[0085] In the formula, Denotes the set of K cluster centers. Indicates the first The center vector of each cluster This indicates the allocation of N data points. This represents the index of the cluster to which the nth data point is assigned. This represents the feature vector of the nth multi-view image. Indicates assignment to feature The center point of that cluster, This represents a weighting coefficient. This represents a function used to obtain the semantic probability distribution of a given feature vector. This represents the KL divergence.

[0086] Association weights and local attention: per BEV query First, calculations are performed with each cluster center. Association weight This allows us to locate a few (e.g., M=3) highly correlated clusters.

[0087] ;

[0088] In the formula, For temperature coefficient, Let represent the center vector of the j-th cluster. Subsequent Top-K feature selection and attention calculation are performed only within these M high-weight clusters, significantly reducing computational complexity and improving the model's detection efficiency and accuracy in unconventional scenarios such as construction areas. This represents the i-th BEV query vector, where each... Each corresponds to a specific query point in the bird's-eye view space. Its task is to actively aggregate information related to this spatial location from image features. T represents transpose.

[0089] S4, Multi-scale feature fusion of pyramids;

[0090] Taking a multi-scale feature map F as input, the output is a feature map F_fused after fusing multi-granularity information. The specific process is as follows:

[0091] S401, Parallel extraction of multi-scale features;

[0092] The input multi-scale feature map F is divided into three groups along the channel dimension and processed in parallel through heterogeneous convolutional branches with different receptive fields.

[0093] High-resolution branches, using 3 × 3 small convolution kernels, capture subpixel-level lane line edge geometric details;

[0094] In the mesoscopic branch, a 5 × 5 dilated convolution is used to model the spatial continuity of roadside curbs and medians;

[0095] The low-resolution branch uses a 7 × 7 large convolution kernel to analyze the macroscopic topological relationship of the traffic flow divider at the intersection.

[0096] S402, Dynamic Channel Attention Weighting;

[0097] The multi-scale features processed in parallel are concatenated to obtain concatenated features, which are then fed into the SE module. The SE module learns a dynamic weight for each feature channel through global average pooling and a two-layer fully connected network, as follows:

[0098] ;

[0099] In the formula, The dynamic weights are represented by σ and δ, which represent the Sigmoid and ReLU activation functions, respectively. Indicates splicing characteristics, , These represent the weight matrices of the two fully connected layers in the SE (Squeeze-and-Excitation) module.

[0100] To enhance perception robustness under dynamic changes and occlusion conditions, this invention introduces temporal continuity constraints based on hidden Markov models.

[0101] State transition model: BEV feature sequence The temporal evolution is modeled as a state transition process, which takes into account the vehicle's own speed. .

[0102] ;

[0103] In the formula, This represents the bird's-eye view features or state at the current time point t. This represents the bird's-eye view features or state at the previous time point t-1. Indicates a Gaussian distribution. This represents speed multiplied by time, used to estimate the spatial displacement of the BEV feature due to the vehicle's own motion during this time period. This represents the covariance matrix of the normal distribution.

[0104] Maximum a posteriori probability estimation: By fusing observation information from the current frame using Bayesian theory. and time-series priors based on historical information The query results for the current frame Optimize.

[0105] ;

[0106] In the formula, This represents the observation information at the current time point t, that is, the multi-view image data collected by the vehicle-mounted camera at time t. It is a conditional probability, representing the probability under given observation conditions. The next state is The probability of.

[0107] S5, Spatiotemporal Fusion and BEV Representation Generation;

[0108] By using a spatial cross-attention mechanism, the dynamic query grid is used as the Query and interacts with the fused feature map F_fused as the Key and Value to aggregate spatial information and generate the BEV feature BEV_t at the current time. Then, the BEV feature BEV_t-1 at the previous time is read, and the coordinates are aligned according to the vehicle's own motion. Finally, it is fused with BEV_t through a temporal self-attention mechanism to enhance temporal continuity and eliminate ambiguity.

[0109] S6. Decode the final BEV feature map after spatiotemporal fusion into specific perception results and output them.

[0110] Compared with existing technologies, this invention achieves the following significant benefits through two core innovations: a dynamic adaptive BEV query mechanism and a pyramid multi-scale attention (PSA) fusion architecture.

[0111] 1. This invention solves the problems of insufficient scene adaptability and unreasonable allocation of computing resources, and achieves a synergistic improvement in accuracy and efficiency.

[0112] Specific improvements: This invention abandons the fixed, uniform BEV query grid used in existing technologies. By introducing an elastic resolution mechanism driven by real-time semantic density, the density of the query grid can be autonomously adjusted according to dynamic changes in road topology (such as interchange weaving areas or straight sections). In structurally complex areas, the system automatically increases the resolution to sub-meter level (e.g., 0.25m) to accurately fit curved lane boundaries (curvature error <5cm); while in structurally simple areas, the resolution is reduced, thereby significantly reducing redundant calculations.

[0113] The beneficial effects include: This design allows computing resources to be allocated "on demand," intelligently focusing on areas requiring the most precise perception. This not only enables centimeter-level full-element analysis of static elements in complex scenarios, significantly improving perception accuracy, but also reduces the computational load in simple scenarios. Compared to fixed-grid methods, it significantly improves inference speed while maintaining high accuracy, providing strong support for the dual requirements of real-time performance and accuracy for Level 4 autonomous driving.

[0114] 2. This invention solves the problem of low coupling efficiency of multi-granularity features, and significantly improves the integrity of scene details and the accuracy of perception.

[0115] Specific improvements: To address the shortcomings of existing technologies that perform feature interactions at a single scale, leading to a break in the connection between macroscopic structure and microscopic geometry, this invention designs a Pyramid Multi-Scale Attention (PSA) fusion architecture. This architecture extracts fine-grained features (such as zebra crossing texture), mesoscopic continuity features (such as curb edges), and macroscopic topological features (such as the direction of traffic dividers at intersections) at different scales through grouped parallel heterogeneous convolutions. Subsequently, the SE module dynamically learns and assigns weights to each channel, and finally performs deep fusion of the weighted multi-scale features.

[0116] Beneficial effects: This multi-scale feature symbiosis system effectively solves the detail-structure imbalance problem. It ensures that the model can simultaneously "see" the sub-pixel edges of lane lines and the overall layout of intersections, avoiding the loss of key information. For example, in the scenario of a spiral ramp on an overpass, this architecture can accurately fuse the gradient details of solid and dashed lines with the global curvature features of the ramp, significantly improving the accuracy of lane line interruption area completion.

[0117] 3. This invention solves the problem of poor perception robustness under complex and extreme working conditions, and improves the all-weather and all-scenario adaptability of autonomous driving systems.

[0118] Specific improvements: This invention not only enhances performance in conventional scenarios through the two innovations mentioned above, but also strengthens the system's generalization ability through an open-domain adaptive mechanism guided by feature clustering and temporal continuity constraints. The feature clustering mechanism enables the model to improve the feature recognition rate in long-tail scenarios such as temporary construction areas and irregular intersections without the need for additional labeled data. Simultaneously, the introduced HMM temporal model utilizes historical information to correct and complete the perception results of the current frame.

[0119] The beneficial effects are as follows: The integrated application of these designs gives the perception system of this invention extremely high robustness. Even in low-visibility scenarios such as heavy rain and backlighting, the recall rate for detecting static elements remains at a high level. The introduction of the temporal model effectively ensures the continuity and stability of vehicle perception in scenarios such as tunnels, avoiding perception interruptions caused by signal loss or temporary obstruction. Ultimately, this invention provides autonomous driving systems with a core perception capability that can operate stably and reliably under various complex and even extreme conditions.

[0120] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A multi-scale feature fusion method for dynamic adaptive BEV perception, characterized in that, include: S1, Data Acquisition; S2, Feature Extraction; S3. Generate dynamic adaptive BEV query; S4, Multi-scale feature fusion of pyramids; S5, Spatiotemporal Fusion and BEV Representation Generation; S6. Decode the final BEV feature map after spatiotemporal fusion into specific perception results and output them. Step S3 specifically includes: Taking a multi-scale feature map F and a set of initial queries Q as input, the output is a query coordinate that is dynamically adjusted in resolution according to the current scene complexity. The specific process is as follows: S301, Scene Complexity Analysis; First, a lightweight semantic segmentation network is used to analyze the multi-scale feature map F, extract the lane line curvature variance and obstacle distribution entropy features, and then linearly combine the lane line curvature variance and obstacle distribution entropy features to generate a scene complexity heatmap H corresponding to the BEV plane. S302, Dynamic resolution adjustment; Design a lightweight gated network based on an MLP, taking a heatmap H and an initial query Q as input, and calculate a resolution scaling factor for each grid location as follows: ； In the formula, This represents the resolution scaling factor. This indicates a feature concatenation operation. This represents the eigenvalue or eigenvector corresponding to the heatmap H at grid position (i, j). Let Q represent the initial query at grid position (i, j). Indicates the bias term. This represents the weight matrix of the last linear transformation layer in the gated network. S303, Dynamic coordinate mapping; Based on the calculated scaling factor, the original uniform grid coordinates are nonlinearly transformed using bilinear interpolation to generate adaptive query coordinates, as follows: ； In the formula, , These represent the preset maximum offsets in the x and y directions, respectively, defining the maximum distance each grid point can move. , These represent the x and y coordinates of the original uniform grid, respectively. , These represent the x and y coordinates of the query coordinates, respectively. S304, Semantic Constraints of K-means++ Clustering; Multi-view image features Divided into K semantic clusters The optimized objective function considers not only geometric distance but also KL divergence. Semantic consistency is introduced as a constraint, as follows: ； In the formula, Denotes the set of K cluster centers. Indicates the first The center vector of each cluster This indicates the allocation of N data points. This represents the index of the cluster to which the nth data point is assigned. This represents the feature vector of the nth multi-view image. Indicates assignment to feature The center point of that cluster, This represents a weighting coefficient. This represents a function used to obtain the semantic probability distribution of a given feature vector; S305, Relevance Weights and Local Attention; Each BEV query First, calculations are performed with each cluster center. Association weight This allows us to locate highly correlated clusters, as follows: ； In the formula, For temperature coefficient, Let the vector represent the center vector of the j-th cluster. Let represent the i-th BEV query vector, and T represent the transpose.

2. The multi-scale feature fusion method for dynamic adaptive BEV perception according to claim 1, characterized in that, Step S4 specifically includes: Taking a multi-scale feature map F as input, the output is a feature map F_fused after fusing multi-granularity information. The specific process is as follows: S401, Parallel extraction of multi-scale features; The input multi-scale feature map F is divided into three groups along the channel dimension and processed in parallel through heterogeneous convolutional branches with different receptive fields. High-resolution branches, using 3 × 3 small convolution kernels, capture subpixel-level lane line edge geometric details; In the mesoscopic branch, a 5 × 5 dilated convolution is used to model the spatial continuity of roadside curbs and medians; The low-resolution branch uses a 7 × 7 large convolution kernel to analyze the macroscopic topological relationship of the traffic flow divider at the intersection; S402, Dynamic Channel Attention Weighting; The multi-scale features processed in parallel are concatenated to obtain concatenated features, which are then fed into the SE module. The SE module learns a dynamic weight for each feature channel through global average pooling and a two-layer fully connected network, as follows: ； In the formula, The dynamic weights are represented by σ and δ, which represent the Sigmoid and ReLU activation functions, respectively. Indicates splicing characteristics, , These represent the weight matrices of the two fully connected layers in the SE module.

3. The multi-scale feature fusion method for dynamic adaptive BEV perception according to claim 2, characterized in that, Step S4 further includes introducing temporal continuity constraints based on hidden Markov models, specifically including: BEV feature sequence The temporal evolution model is a state transition process, taking into account the vehicle's own speed. The method is as follows: ； In the formula, This represents the bird's-eye view features or state at the current time point t. This represents the bird's-eye view features or state at the previous time point t-1. Indicates a Gaussian distribution. This represents speed multiplied by time, used to estimate the spatial displacement of the BEV feature due to the vehicle's own motion during this time period. Represents the covariance matrix of a normal distribution; By using Bayesian theory, the observation information of the current frame is fused. and time-series priors based on historical information The query results for the current frame The optimization can be performed as follows: ； In the formula, This represents the observation information at the current time point t, that is, the multi-view image data collected by the vehicle-mounted camera at time t. It is a conditional probability, representing the probability under given observation conditions. The next state is The probability of.

4. The multi-scale feature fusion method for dynamic adaptive BEV perception according to claim 3, characterized in that, Step S5 specifically includes: Through a spatial cross-attention mechanism, the dynamic query grid is used as the Query and interacts with the fused feature map F_fused as the Key and Value to aggregate spatial information and generate the BEV feature BEV_t at the current time. Then, the BEV feature BEV_t-1 at the previous time is read, the coordinates are aligned according to the vehicle's own motion, and then fused with BEV_t through a temporal self-attention mechanism.

5. The multi-scale feature fusion method for dynamic adaptive BEV perception according to claim 1, characterized in that, Step S1 specifically includes: Multiple wide-angle cameras deployed around the vehicle body simultaneously collect 360° multi-view image data of the vehicle's surroundings.

6. The multi-scale feature fusion method for dynamic adaptive BEV perception according to claim 1, characterized in that, Step S2 specifically includes: Multiscale feature maps F are extracted from the acquired image data using ResNet or Swing Transformer.

Images