Artificial intelligence-based autonomous driving environment perception method and system

Abstract

The application discloses an automatic driving environment perception method and system based on artificial intelligence, and relates to the technical field of automatic driving environment perception, in particular to an automatic driving environment perception method and system based on artificial intelligence.The application obtains original data through multi-source data collection; adopts a raw data optimization method of sensor calibration, dynamic and static data separation, data normalization, data enhancement and data set segmentation; adopts a double-branch transformer model as an environment perception model, and respectively captures irregular behavior patterns of moving targets and geometric topological constraints of road structures through a double-branch architecture; and adopts an improved ant colony optimization algorithm as a decision optimization algorithm, converts a multi-dimensional risk field into a calculable optimization target, and realizes optimal decision through global exploration of elite ants and local optimization of non-elite ants.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving environment perception, in particular to an automatic driving environment perception method and system based on artificial intelligence. BACKGROUND

[0002] Automatic driving environment perception refers to a technical process of collecting real-time data of the dynamic and static environment around the vehicle through multiple sensors such as laser radar, camera and millimeter wave radar, and structuring and understanding the environmental information by means of deep learning and multi-source data fusion algorithms, so as to construct a three-dimensional dynamic environment model that can be recognized by the vehicle. Through accurate capture of multi-dimensional information in complex traffic scenes, it provides key basis for vehicle decision planning, can significantly reduce the rate of traffic accidents caused by human errors, and improves the safety of the traffic system.

[0003] However, the traditional automatic driving environment perception method has the technical problems of insufficient feature extraction capability for irregular motion patterns, lack of topological relevance in static environment modeling, difficulty in adapting to roads with varying curvatures, and lack of explicit modeling of the interaction between dynamic and static elements, resulting in blind spots in risk field assessment; the traditional automatic driving environment perception method has the technical problems of not establishing a coupled optimization model of risk field and vehicle kinematics when making automatic driving decisions, leading to the possibility of obstacle avoidance strategies violating the dynamics constraints or causing passenger discomfort, and lack of global situation field guidance and adaptive exploration mechanism, which seriously limits the stability of the decision. SUMMARY

[0004] In view of the above, in order to overcome the defects of the prior art, the present application provides an automatic driving environment perception method and system based on artificial intelligence, which is aimed at the technical problems of the traditional automatic driving environment perception method, such as insufficient feature extraction capability for irregular motion patterns, lack of topological relevance in static environment modeling, difficulty in adapting to roads with varying curvature, and the interactive relationship between dynamic and static elements not being explicitly modeled, resulting in blind spots in risk field evaluation. The present scheme creatively uses a double-branch transformer model as an environment perception model, which captures the irregular behavior patterns of moving targets and the geometric topological constraints of road structures through a double-branch architecture, achieving feature complementarity between dynamic risk and static constraints. At the same time, superpixel graph convolution is introduced to analyze the semantic continuity of the road, providing a precise basis for environmental awareness for subsequent decision-making. For the technical problems of the traditional automatic driving environment perception method, such as not establishing a coupled optimization model of risk field and vehicle kinematics when making automatic driving decisions, leading to the possibility of violating the dynamics constraints or causing passenger discomfort in obstacle avoidance strategies, and lacking global situation field guidance and adaptive exploration mechanism, which severely limits the stability of the decision, the present scheme creatively uses an improved ant colony optimization algorithm as a decision optimization algorithm, which converts the multi-dimensional risk field output by the environment perception into a computable optimization target. Through the dual mechanism of global exploration by elite ant colony and local optimization by non-elite, the optimal decision is realized under the condition of satisfying the vehicle dynamics constraints.

[0005] The technical solutions adopted by the present application are as follows: The automatic driving environment perception method based on artificial intelligence provided by the present application comprises the following steps:

[0006] Step S1: Multi-source data acquisition;

[0007] Step S2: Raw data optimization;

[0008] Step S3: Environment perception model construction;

[0009] Step S4: Decision optimization algorithm design;

[0010] Step S5: Automatic driving decision.

[0011] Further, in step S1, the multi-source data acquisition is used to collect the raw data required for automatic driving environment perception, specifically through multi-type sensors to collect data and obtain an environment perception raw data set. The environment perception raw data set specifically includes a past perception raw data set and a current perception raw data set. Both the past perception raw data set and the current perception raw data set include ego data, dynamic environment data, and static environment data. The past perception raw data set further includes past dynamic obstacle threat scores and past static region passage scores.

[0012] Further, in step S2, the raw data optimization is used to optimize the collected environment perception raw data, specifically including the following steps:

[0013] Step S21: sensor calibration, for unifying multi-sensor coordinate system and timestamp, specifically converting different sensor data to vehicle coordinate system and aligning timestamp through interpolation method to obtain spatio-temporal aligned data;

[0014] Step S22: dynamic and static data separation, for distinguishing moving targets and static backgrounds in the environment, specifically separating dynamic and static environment data through optical flow method;

[0015] Step S23: data normalization, for eliminating dimension differences of different sensor data, specifically using minimum-maximum normalization method to convert data to a unified range;

[0016] Step S24: data augmentation, for increasing the diversity of training data, specifically performing random transformation on the original data to obtain expanded data, the random transformation including rotation, translation, scaling and adding noise conforming to physical laws;

[0017] Step S25: data set segmentation, specifically performing data set segmentation on the past perception raw data set to obtain perception training set and perception test set;

[0018] Through the sensor calibration, the dynamic and static data separation and the data normalization, the current perception raw data set is optimized to obtain a to-be-processed data set, and through the sensor calibration, the dynamic and static data separation, the data normalization, the data augmentation and the data set segmentation, the past perception raw data set is optimized to obtain the perception training set and the perception test set.

[0019] Further, in step S3, the environment perception model construction is used to construct the model required for automatic driving environment perception, specifically constructing a double-branch transformer model as an environment perception model, the double-branch transformer model specifically including a dynamic environment branch, a static environment branch, a feature fusion module and a stacked transformer module;

[0020] The environment perception model construction specifically includes the following steps:

[0021] Step S31: dynamic environment branch design, for capturing spatio-temporal features of moving targets around the vehicle, including:

[0022] Step S311: three-dimensional deformable convolution, for extracting irregular motion features of dynamic targets, specifically extracting features in time and space dimensions through a three-dimensional convolution kernel with a learnable offset to obtain preliminary dynamic spatio-temporal features;

[0023] Step S312: Temporal coherence unit construction is configured to model the motion consistency of dynamic targets at continuous time, specifically to capture the long-time dependence of dynamic targets through a gated recurrent residual connection and a historical state decay factor, to obtain time-enhanced dynamic spatio-temporal features;

[0024] Step S313: Spatial coherence mask design is configured to enhance the spatial structure features of dynamic targets, specifically to focus on dynamic features through a spatial decay mask based on target center, to obtain final dynamic spatio-temporal features;

[0025] Step S32: Static environment branch design is configured to analyze the road structure and spatial topological relationship, and the step includes:

[0026] Step S321: Direction response convolution is configured to extract direction-sensitive features in the static environment, specifically to perform convolution on the input static environment data through a plurality of preset direction convolution kernels, to obtain preliminary static spatial features;

[0027] Step S322: Superpixel graph convolution unit construction is configured to abstract from pixel-level features to regional topological structure, specifically to perform superpixel segmentation on the input static environment data through a simple linear iterative clustering algorithm, to process each superpixel region through the direction response convolution, to obtain preliminary static spatial features of each superpixel region, to take the superpixel region as a node, the preliminary static spatial features of the superpixel region as node features, and the edges between nodes representing the adjacent relationship between superpixel regions, to construct graph structure data and perform graph convolution operation, to obtain superpixel static spatial features;

[0028] Step S323: Road curvature encoding is configured to embed road geometric shape information into features, specifically to fuse curvature information with superpixel static spatial features by calculating road curvature and combining B-spline interpolation, to obtain final static spatial features;

[0029] Step S33: Feature fusion module design is configured to perform multi-scale adaptive fusion of features output by the dynamic environment branch and the static environment branch, specifically to dynamically weight and fuse the final dynamic spatio-temporal features and the final static spatial features at different scales through a learnable gating mechanism and a channel attention weight, to obtain dynamic-static fusion features;

[0030] Step S34: Stack transformer module design is configured to model long-distance spatio-temporal dependence in the environment, specifically to process dynamic-static fusion features through a plurality of stacked transformer layers, and the stack transformer module includes three output layers, respectively outputting a predicted trajectory, a dynamic obstacle threat score, and a static region passing score;

[0031] Step S35: constructing and training the model, specifically, constructing the double-branch transformer model through the dynamic environment branch design, the static environment branch design, the feature fusion module design and the stacked transformer module design, training the model based on the perception training set, verifying the model performance based on the perception test set, obtaining the double-branch transformer model as the environment perception model.

[0032] Further, in step S4, the decision optimization algorithm design is used to design the optimization algorithm required for optimizing automatic driving decisions, specifically, an improved ant colony optimization algorithm is used as the decision optimization algorithm.

[0033] The decision optimization algorithm design specifically includes the following steps:

[0034] Step S41: state space construction, for quantifying environmental risks and constraints, the steps include:

[0035] Step S411: threat potential field construction, for quantifying the safety threats of dynamic and static environments to the vehicle, specifically, by converting the predicted trajectory, dynamic obstacle threat score and static region passing score output by the environment perception model into a dynamic potential field function, obtaining the threat potential field.

[0036] Step S412: lane constraint construction, for expressing the driving constraints of road structure to the vehicle, specifically, by calculating the static potential field through the position of the ego vehicle from the lane center line and the lane boundary, obtaining the constraint potential field.

[0037] Step S413: state space definition, specifically, constructing the state vector that constitutes the state space, the elements of which include ego vehicle position data, ego vehicle speed data, ego vehicle acceleration data, ego vehicle heading angle data, predicted driving time, threat potential field and constraint potential field, constructing an ant unit set, the search path set of the ant unit set in the state space, as the automatic driving decision set, and the optimal search path as the optimal automatic driving decision.

[0038] Step S42: design the objective function, for designing the path objective function required for optimizing automatic driving decisions.

[0039] Step S43: design the ant colony state transition mechanism, for realizing global search path optimization, the steps include:

[0040] Step S431: pheromone field initialization, for initializing the pheromone distribution of ant colony search, specifically, setting the initial pheromone concentration according to the risk cost, obtaining the initial pheromone field guiding the search of ant units.

[0041] Step S432: elite ant global search, used for performing global optimal search path exploration, specifically guiding the movement of elite ant units by combining pheromone concentration, potential field and transfer probability formula of kinematic energy constraint, the elite ant units being specifically the ant units ranked in ascending order of search path path target function value in the top 30%;

[0042] Step S44: dual population collaborative optimization, used for collaborative optimization of elite ant units and non-elite ant units, the steps including:

[0043] Step S441: dynamic update of pheromone field, used for updating the pheromone field according to the search situation to feed back to the next search, specifically updating the pheromone field based on exponential decay of the path target function value of the elite ant units to obtain an updated pheromone field;

[0044] Step S442: local optimization of non-elite ant units, used for fine search near the search path of the elite ant units, specifically generating a new candidate path by applying a Gaussian disturbance decaying over time to the search path of the elite ant units;

[0045] Step S45: obtaining an optimal search path, specifically designing an improved ant colony optimization algorithm through the state space construction, the design target function, the design ant colony state transition mechanism and the dual population collaborative optimization, and continuously searching until a search stop condition is reached to obtain an optimal search path as an optimal automatic driving decision, the search stop condition specifically including reaching a maximum search round and a path target function value being less than a preset threshold.

[0046] Further, in step S5, the automatic driving decision is specifically taking the to-be-processed data set as an input of the environment perception model to obtain a predicted trajectory, a dynamic obstacle threat score and a static area passing score, and using the decision optimization algorithm to optimize the automatic driving decision based on the predicted trajectory, the dynamic obstacle threat score and the static area passing score to obtain and execute an optimal automatic driving decision.

[0047] The automatic driving environment perception system based on artificial intelligence provided by the application includes a multi-source data acquisition module, an original data optimization module, an environment perception model construction module, a decision optimization algorithm design module and an automatic driving decision module.

[0048] The multi-source data acquisition module is used for acquiring original data, obtaining an environment perception original data set through the acquisition of original data, and sending the environment perception original data set to the original data optimization module.

[0049] The original data optimization module is used for original data optimization, and through original data optimization, a to-be-processed data set, a perception training set and a perception test set are obtained, and the to-be-processed data set is sent to the automatic driving decision module, and the perception training set and the perception test set are sent to the environment perception model construction module.

[0050] The environment perception model construction module is used for environment perception model construction, and through constructing a double-branch transformer model as an environment perception model, the environment perception model is sent to the automatic driving decision module.

[0051] The decision optimization algorithm design module is used for decision optimization algorithm design, and through adopting an improved ant colony optimization algorithm as a decision optimization algorithm, the decision optimization algorithm is sent to the automatic driving decision module.

[0052] The automatic driving decision module is used for automatic driving decision, and through real-time processing data by using the environment perception model, a predicted trajectory, a dynamic obstacle threat score and a static area passing score are obtained, and automatic driving decision is optimized in real time by combining a decision optimization algorithm, so that an optimal automatic driving decision is obtained and executed.

[0053] The above scheme has the following beneficial effects:

[0054] (1) In view of the technical problems that the traditional automatic driving environment perception method has insufficient feature extraction capability for irregular motion patterns, lacks topological correlation in static environment modeling, is difficult to adapt to roads with varying curvatures, and the interaction relationship between dynamic and static elements is not explicitly modeled, resulting in a blind area in risk field evaluation, the scheme creatively adopts a double-branch transformer model as an environment perception model, which captures irregular behavior patterns of moving targets and geometric topological constraints of road structures through a double-branch architecture, realizes feature complementarity of dynamic risk and static constraints, and introduces superpixel graph convolution to analyze road semantic continuity, thereby providing accurate environment cognition basis for subsequent decision-making.

[0055] (2) In view of the technical problems that in the traditional automatic driving environment perception method, a coupling optimization model of risk field and vehicle kinematics is not established when automatic driving decision is made, resulting in that the obstacle avoidance strategy may violate the dynamics constraint or cause passenger discomfort, and lacking global situation field guidance and adaptive exploration mechanism, the decision stability is seriously limited, the scheme creatively adopts an improved ant colony optimization algorithm as a decision optimization algorithm, which converts the multi-dimensional risk field output by environment perception into a calculable optimization target, and through the dual mechanism of global exploration of elite ant colony and local optimization of non-elite, realizes optimal decision under the condition of satisfying the vehicle dynamics constraint. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1A flowchart of an automatic driving environment perception method based on artificial intelligence provided by the present application is shown in the figure.

[0057] Figure 2 A module diagram of an automatic driving environment perception system based on artificial intelligence provided by the present application is shown in the figure.

[0058] Figure 3 A flowchart of raw data optimization in step S2 is shown in the figure.

[0059] Figure 4 A flowchart of environment perception model construction in step S3 is shown in the figure.

[0060] Figure 5 A flowchart of decision optimization algorithm design in step S4 is shown in the figure.

[0061] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, which is used to explain the present application together with embodiments of the present application, and do not constitute a limitation of the present application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0063] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the present application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0064] Embodiment one, refer to Figure 1 The technical solutions adopted by the present application are as follows: The automatic driving environment perception method based on artificial intelligence provided by the present application comprises the following steps:

[0065] Step S1: Multi-source data acquisition;

[0066] Step S2: Raw data optimization;

[0067] Step S3: Environment perception model construction;

[0068] Step S4: Decision optimization algorithm design;

[0069] Step S5: autonomous driving decision.

[0070] Embodiment two, see Figure 1 and Figure 2 In step S1, the multi-source data collection is used to collect the original data required for autonomous driving environment perception, specifically through multi-type sensor data collection to obtain an environment perception original data set, the environment perception original data set specifically includes a past perception original data set and a current perception original data set, both the past perception original data set and the current perception original data set include ego vehicle data, dynamic environment data and static environment data, the past perception original data set further includes a past dynamic obstacle threat score and a past static area passing score, the ego vehicle data specifically includes ego vehicle mass data, ego vehicle position data, ego vehicle speed data, ego vehicle acceleration data and ego vehicle heading angle data, the dynamic environment data is specifically moving target detection data, including image data of surrounding vehicles, pedestrians and animals, and retaining their position data, speed data and acceleration data, the static environment data specifically includes curbstone geometric parameters, lane curvature, static obstacle data, road structure data and traffic sign image data.

[0071] Embodiment three, see Figure 1 、 Figure 2 and Figure 3 This embodiment is based on the above embodiment, in step S2, the original data optimization is used to optimize the collected environment perception original data, specifically including the following steps:

[0072] Step S21: sensor calibration, for unifying multi-sensor coordinate system and time stamp, specifically converting different sensor data to vehicle coordinate system and aligning time stamp through interpolation method to obtain spatio-temporal aligned data;

[0073] Step S22: dynamic and static data separation, for distinguishing moving targets and static backgrounds in the environment, specifically separating dynamic environment data and static environment data through optical flow method;

[0074] Step S23: data normalization, for eliminating dimension difference of different sensor data, specifically using minimum-maximum normalization method to convert data to a unified range;

[0075] Step S24: data augmentation, for increasing diversity of training data, specifically randomly transforming original data to obtain expanded data, the random transformation includes rotation, translation, scaling and adding noise conforming to physical laws;

[0076] Step S25: data set segmentation, specifically data set segmentation of the past perception original data set to obtain a perception training set and a perception test set;

[0077] The current perception raw data set is optimized through the sensor calibration, the dynamic and static data separation, and the data normalization to obtain a to-be-processed data set, and the past perception raw data set is optimized through the sensor calibration, the dynamic and static data separation, the data normalization, the data enhancement, and the data set segmentation to obtain a perception training set and a perception test set.

[0078] Embodiment four, refer to Figure 1 , Figure 2 and Figure 4 , based on the above embodiment, in step S3, the environment perception model is constructed, which is used to construct the model required for automatic driving environment perception, specifically to construct a double-branch transformer model as an environment perception model, the double-branch transformer model specifically includes a dynamic environment branch, a static environment branch, a feature fusion module, and a stacked transformer module.

[0079] The environment perception model construction specifically includes the following steps:

[0080] Step S31: dynamic environment branch design, used to capture the spatiotemporal features of moving targets around the vehicle, the step includes:

[0081] Step S311: three-dimensional deformable convolution, used to extract irregular motion features of dynamic targets, specifically through a three-dimensional convolution kernel with a learnable offset to adaptively extract features in time and space dimensions to obtain preliminary dynamic spatiotemporal features, the formula used is as follows:

[0082] ;

[0083] In the formula, represents the preliminary dynamic spatiotemporal features, represents a three-dimensional deformable convolution function, represents input dynamic environment data, represents a learnable offset, represents a learnable modulation scalar;

[0084] Step S312: time coherence unit construction, used to model the motion consistency of dynamic targets in consecutive time, specifically through a gated recurrent residual connection and a history state decay factor to capture the long-time dependence relationship of dynamic targets to obtain time-enhanced dynamic spatiotemporal features, the formula used is as follows:

[0085] ;

[0086] In the formula, represents the initial time coherence unit hidden state, represents the preliminary dynamic spatiotemporal features at time step t, denotes a history state decay factor, denotes a sigmoid function, denotes a decay factor transformation matrix, denotes an initial time-coherent cell hidden state at time step t-1, denotes an initial time-coherent cell hidden state at time step t, a gating transformation matrix, denotes a hyperbolic tangent function, denotes a cell input transformation matrix, denotes a cell input bias term, denotes a tensor product operation;

[0087] Step S313: Spatial coherence mask design is used to enhance the spatial structure features of dynamic targets. Specifically, a target center-based spatial decay mask is used to weight and focus on dynamic features to obtain final dynamic spatiotemporal features. The formula used is as follows:

[0088] ;

[0089] In the formula, Cm denotes a target center-based spatial decay mask, le denotes a learnable distance weight parameter, denotes the Euclidean distance from the ego vehicle pixel point to the target center pixel point, denotes the final dynamic spatiotemporal feature, denotes a time-enhanced dynamic spatiotemporal feature, denotes a bilinear interpolation alignment operation function, denotes an element-wise multiplication operation;

[0090] Step S32: Static environment branch design is used to analyze road structures and spatial topological relationships. The steps include:

[0091] Step S321: Direction response convolution is used to extract direction-sensitive features in the static environment. Specifically, a plurality of preset direction convolution kernels are used to respectively convolve the input static environment data to obtain preliminary static spatial features. The formula used is as follows:

[0092] ;

[0093] In the formula, denotes a preliminary static spatial feature with a direction angle of ; denotes a direction convolution function with a convolution kernel direction angle of ; denotes input static environment data, denotes a convolution kernel direction angle, denotes a preliminary static spatial feature;

[0094] Step S322: superpixel graph convolution unit construction, for abstracting from pixel-level features to regional-level topology, specifically, superpixel segmentation is performed on the input static environment data by a simple linear iterative clustering algorithm, the direction response convolution is used to process each superpixel region, the preliminary static spatial features of each superpixel region are obtained, the superpixel regions are taken as nodes, the preliminary static spatial features of the superpixel regions are taken as node features, the edges between the nodes represent the adjacent relationship between the superpixel regions, the graph structure data is constructed, and the graph convolution operation is performed to obtain superpixel static spatial features;

[0095] Step S323: road curvature encoding, for embedding road geometric shape information into features, specifically, the curvature information is fused with the superpixel static spatial features by calculating the road curvature and combining B-spline interpolation to obtain the final static spatial features, and the formula is as follows:

[0096] ;

[0097] In the formula, represents the final static spatial features, represents a multi-layer perception operation function, represents the superpixel static spatial features, represents a B-spline interpolation function, represents the road curvature;

[0098] Step S33: feature fusion module design, for multi-scale adaptive fusion of the features output by the dynamic environment branch and the static environment branch, specifically, the final dynamic spatio-temporal features and the final static spatial features are dynamically weighted and fused at different scales by a learnable gating mechanism and a channel attention weight to obtain dynamic-static fusion features, and the formula is as follows:

[0099] ;

[0100] In the formula, represents the fusion gating weight of the kth scale, represents a fully connected layer function of the kth scale, represents the final dynamic spatio-temporal features of the kth scale, represents the final static spatial features of the kth scale, and Ch represents a channel attention weight, represents a softmax function, represents a channel attention transformation matrix, represents a global average pooling function, represents the dynamic-static fusion features of the kth scale, represents the dynamic-static fusion features, and K represents the total number of scales;

[0101] Step S34: a stacked transformer module design is used to model long-range spatio-temporal dependencies in the modeling environment, specifically by processing dynamic-static fusion features through stacking multiple transformer layers, the stacked transformer module includes three output layers, which respectively output predicted trajectories, dynamic obstacle threat scores and static region passage scores;

[0102] Step S35: the model is constructed and trained, specifically by using the dynamic environment branch design, the static environment branch design, the feature fusion module design and the stacked transformer module design to construct the double-branch transformer model, training the model based on the perception training set, verifying the model performance based on the perception test set, obtaining the double-branch transformer model as the environment perception model.

[0103] By performing the above operations, the technical problem of the conventional automatic driving environment perception method that the feature extraction capability for irregular motion patterns is insufficient, the static environment modeling lacks topological correlation, it is difficult to adapt to roads with changing curvature, and the interaction between dynamic and static elements is not explicitly modeled, resulting in blind spots in risk field evaluation is solved. The present scheme creatively uses a double-branch transformer model as an environment perception model, which captures irregular behavior patterns of moving targets and geometric topological constraints of road structures through a double-branch architecture, realizes feature complementarity of dynamic risk and static constraints, and introduces superpixel graph convolution to analyze road semantic continuity, providing accurate environmental awareness basis for subsequent decision-making.

[0104] Embodiment five, refer to Figure 1 、 Figure 2 and Figure 5 This embodiment is based on the above-mentioned embodiments. In step S4, the decision optimization algorithm design is used to design an optimization algorithm required for optimizing automatic driving decisions, specifically an improved ant colony optimization algorithm is used as the decision optimization algorithm.

[0105] The decision optimization algorithm design specifically includes the following steps:

[0106] Step S41: state space construction is used to quantify environmental risks and constraints, which includes the following steps:

[0107] Step S411: threat potential field construction is used to quantify the safety threats of dynamic and static environments to vehicles, specifically by converting the predicted trajectories, dynamic obstacle threat scores and static region passage scores output by the environment perception model into a dynamic potential field function to obtain a threat potential field, the formula used is as follows:

[0108] ;

[0109] In the formula, Tf represents the threat potential field, represents the dynamic threat weight, Static threat weight, Sa represents the number of samples, Dynamic obstacle threat score of the sa-th sample, Static area pass score of the sa-th sample, Euclidean distance from the ego vehicle to the predicted trajectory of the sa-th sample, Prevent zero from the minimum value;

[0110] Step S412: Lane constraint construction, for expressing the driving constraints of the road structure on the vehicle, specifically calculating the static potential field by the position of the ego vehicle from the lane center line and the lane boundary, obtaining the constraint potential field, and the formula is as follows:

[0111] ;

[0112] In the formula, Lf represents the constraint potential field, Center constraint weight, Boundary constraint weight, Euclidean distance from the ego vehicle to the lane center line, Euclidean distance from the ego vehicle to the lane boundary;

[0113] Step S413: State space definition, specifically constructing the state vector that constitutes the state space, whose elements include ego vehicle position data, ego vehicle speed data, ego vehicle acceleration data, ego vehicle heading angle data, predicted driving time, threat potential field and constraint potential field, constructing an ant unit set, which is a set of search paths in the state space, as a set of automatic driving decisions, and the optimal search path as the optimal automatic driving decision;

[0114] Step S42: Design the objective function, which is used to design the path objective function required for optimizing the automatic driving decision, and the formula is as follows:

[0115] ;

[0116] In the formula, Fun represents the path objective function value, Time cost weight, Threat cost weight, Comfort cost weight, Time cost, which is the predicted driving time, Risk cost, which is the integral of the threat potential field along the path , Ego vehicle acceleration data, Comfort cost, which is the integral of the square of the acceleration change rate;

[0117] Step S43: Design the ant colony state transition mechanism, which is used to realize the global search path optimization, and the steps include:

[0118] Step S431: pheromone field initialization, used to initialize the pheromone distribution of the ant colony search, specifically to set the initial pheromone concentration according to the risk cost, to obtain the initial pheromone field guiding the search of the ant unit, and the formula used is as follows:

[0119] ;

[0120] In the formula, represents the initial pheromone concentration from node i to node j, represents the initial pheromone risk weight, represents the initial pheromone basic weight, represents the average of the risk cost of node i and node j, represents the basic pheromone concentration;

[0121] Step S432: elite ant global search, used to perform global optimal search path exploration, specifically to guide the movement of the elite ant unit by combining the transition probability formula of pheromone concentration, potential field and kinematic energy constraint, the elite ant unit, specifically the ant unit with the ascending order of the search path path target function value being the top 30%, and the transition probability formula is as follows:

[0122] ;

[0123] In the formula, represents the transition probability of the elite ant unit from node i to node j, represents the pheromone concentration from node i to node j, represents the average of the threat potential field of node i and node j, represents the average of the constraint potential field of node i and node j, represents the change amount of kinetic energy from node i to node j, represents the pheromone concentration from node i to node l, represents the average of the threat potential field of node i and node l, represents the average of the constraint potential field of node i and node l, represents the change amount of kinetic energy from node i to node l, represents the neighbor node set of node i, represents the pheromone index weight, represents the potential field index weight, represents the kinematic energy index weight, represents the self-vehicle mass data, represents the self-vehicle speed data of node j, represents the self-vehicle speed data of node i;

[0124] Step S44: dual population collaborative optimization, for collaborative optimization of elite ant units and non-elite ant units, the steps comprising:

[0125] Step S441: dynamic update of pheromone field, for updating the pheromone field according to the search situation to feed back to the next search, specifically, updating the pheromone field based on the exponential decay of the path objective function value of the elite ant unit, obtaining the updated pheromone field, the formula used is as follows:

[0126] ;

[0127] In the formula, denotes the strength of the search path of the elite ant unit, denotes the path objective function value of the historical optimal search path, denotes the path objective function value of the search path of the elite ant unit, denotes the change amount of pheromone concentration from node i to node j, denotes the decay coefficient, m denotes the current search round, denotes the pheromone concentration from node i to node j in the m+1th search, denotes the evaporation coefficient, denotes the pheromone concentration from node i to node j in the mth search, N denotes the number of elite ant units, denotes the change amount of pheromone concentration from node i to node j of the search path of the nth elite ant unit;

[0128] Step S442: local optimization of non-elite ant units, for fine search near the search path of the elite ant unit, specifically, generating a new candidate path by applying a Gaussian disturbance with time decay to the search path of the elite ant unit;

[0129] Step S45: obtaining the optimal search path, specifically, through the state space construction, the design objective function, the design ant colony state transition mechanism and the dual population collaborative optimization, the improved ant colony optimization algorithm is designed, and the search is continuously searched until the search stopping condition is reached, the optimal search path is obtained as the optimal automatic driving decision, the search stopping condition, specifically, including reaching the maximum search round and the path objective function value being less than the preset threshold.

[0130] By performing the above operation, for the technical problem that the conventional automatic driving environment perception method does not establish a coupling optimization model of risk field and vehicle kinematics when making an automatic driving decision, resulting in that the obstacle avoidance strategy may violate the dynamics constraint or cause passenger discomfort, and lacking a global situation field guiding and adaptive exploration mechanism, the decision stability is seriously limited, the present application creatively adopts an improved ant colony optimization algorithm as a decision optimization algorithm, converts the multi-dimensional risk field output by the environment perception into a calculable optimization target, and realizes the optimal decision under the condition of meeting the vehicle dynamics constraint through the cooperation of the global exploration of elite ant colony and the local optimization of non-elite.

[0131] Embodiment six, refer to Figure 1 and Figure 2 This embodiment is based on the above-mentioned embodiment, in step S5, the automatic driving decision, specifically, the to-be-processed data set is taken as the input of the environment perception model to obtain the predicted trajectory, the dynamic obstacle threat score and the static area passing score, and based on the predicted trajectory, the dynamic obstacle threat score and the static area passing score, the decision optimization algorithm is used to optimize the automatic driving decision to obtain the optimal automatic driving decision and execute.

[0132] Embodiment seven, refer to Figure 1 and Figure 2 This embodiment is based on the above-mentioned embodiment, and the automatic driving environment perception system based on artificial intelligence provided by the present application comprises a multi-source data acquisition module, an original data optimization module, an environment perception model construction module, a decision optimization algorithm design module and an automatic driving decision module.

[0133] The multi-source data acquisition module is used for acquiring original data, obtaining an environment perception original data set through the acquisition of original data, and sending the environment perception original data set to the original data optimization module.

[0134] The original data optimization module is used for original data optimization, obtaining a to-be-processed data set, a perception training set and a perception test set through the original data optimization, and sending the to-be-processed data set to the automatic driving decision module and sending the perception training set and the perception test set to the environment perception model construction module.

[0135] The environment perception model construction module is used for environment perception model construction, and a double-branch transformer model is constructed as an environment perception model, and the environment perception model is sent to the automatic driving decision module.

[0136] The decision optimization algorithm design module is used for decision optimization algorithm design, and an improved ant colony optimization algorithm is adopted as a decision optimization algorithm, and the decision optimization algorithm is sent to the automatic driving decision module.

[0137] The automatic driving decision module is configured to automatically make a driving decision by using the environment perception model to process data in real time to obtain a predicted trajectory, a dynamic obstacle threat score, and a static area passing score, and combining a decision optimization algorithm to optimize the automatic driving decision in real time to obtain an optimal automatic driving decision and execute the same.

[0138] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any actual relationship or order between or among such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0139] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and alterations can be made hereto without departing from the principles and spirit of the application.

[0140] The above description of the present application and its embodiments is not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which should belong to the protection scope of the present application.

Claims

1. An automatic driving environment perception method based on artificial intelligence, characterized in that: The method comprises the following steps: S1: multi-source data acquisition, data acquisition is carried out through multiple sensors, and an environment perception original data set is obtained, wherein the environment perception original data set specifically comprises a past perception original data set and a current perception original data set; S2: original data optimization, the collected original data is optimized to obtain a to-be-processed data set, a perception training set and a perception test set; S3: environment perception model construction, which is used for constructing a model required for automatic driving environment perception, specifically a double-branch transformer model as an environment perception model, wherein the double-branch transformer model specifically comprises a dynamic environment branch, a static environment branch, a feature fusion module and a stacked transformer module; S4: decision optimization algorithm design, which is used for designing an optimization algorithm required for optimizing automatic driving decision, specifically an improved ant colony optimization algorithm as a decision optimization algorithm; S5: automatic driving decision, specifically taking the to-be-processed data set as the input of the environment perception model to obtain a predicted trajectory, a dynamic obstacle threat score and a static region passing score, and using the decision optimization algorithm to optimize automatic driving decision based on the predicted trajectory, the dynamic obstacle threat score and the static region passing score to obtain an optimal automatic driving decision and execute the optimal automatic driving decision; The environment perception model construction specifically comprises the following steps: Step S31: dynamic environment branch design, which is used for capturing the space-time features of the moving targets around the vehicle, and the step comprises: Step S311: three-dimensional deformable convolution, which is used for extracting irregular motion features of dynamic targets, specifically through a three-dimensional convolution kernel with a learnable offset to adaptively extract features in the time and space dimensions to obtain preliminary dynamic space-time features; Step S312: time coherence unit construction, which is used for modeling the motion consistency of dynamic targets in continuous time, specifically through a gated recurrent residual connection and a historical state decay factor to capture the long-time dependence relationship of dynamic targets to obtain time-enhanced dynamic space-time features; Step S313: spatial coherence mask design, which is used for enhancing the spatial structure features of dynamic targets, specifically through a spatial decay mask based on the target center to weight and focus the dynamic features to obtain final dynamic space-time features; Step S32: static environment branch design, which is used for analyzing the road structure and spatial topological relationship, and the step comprises: Step S321: direction response convolution, which is used for extracting direction-sensitive features in the static environment, specifically through a plurality of preset direction convolution kernels to respectively convolve the input static environment data to obtain preliminary static spatial features; Step S322: Constructing superpixel graph convolutional units to abstract from pixel-level features to region-level topology. Specifically, the input static environment data is segmented into superpixels using a simple linear iterative clustering algorithm. Each superpixel region is processed by the directional response convolution to obtain the preliminary static spatial features of each superpixel region. The superpixel region is used as a node, and the preliminary static spatial features of the superpixel region are used as node features. The edges between nodes represent the adjacency relationship between superpixel regions. Graph structure data is constructed, and graph convolution operation is performed to obtain the superpixel static spatial features. Step S323: Road curvature encoding, used to embed road geometry information into features, specifically by calculating road curvature and combining it with B-spline interpolation, fusing curvature information with superpixel static spatial features to obtain the final static spatial features; Step S33: Feature fusion module design, used to perform multi-scale adaptive fusion of features output from dynamic environment branch and static environment branch. Specifically, through learnable gating mechanism and channel attention weight, the final dynamic spatiotemporal features and the final static spatial features are dynamically weighted and fused at different scales to obtain dynamic-static fused features. Step S34: Stacked transformer module design, used to model long-distance spatiotemporal dependencies in the environment, specifically by stacking multiple transformer layers to process dynamic-static fusion features. The stacked transformer module contains three output layers, which output the predicted trajectory, dynamic obstacle threat score and static area access score, respectively. Step S35: Construct and train the model. Specifically, the dual-branch transformer model is constructed through the dynamic environment branch design, the static environment branch design, the feature fusion module design, and the stacked transformer module design. The model is trained based on the perception training set, and the model performance is verified based on the perception test set to obtain the dual-branch transformer model as the environment perception model. 2.The AI-based automatic driving environment perception method of claim 1, wherein: The decision optimization algorithm design specifically includes the following steps: Step S41: State space construction, used to quantify environmental risks and constraints, includes the following steps: Step S411: Threat potential field construction, used to quantify the safety threats to the vehicle from dynamic and static environments. Specifically, the threat potential field is obtained by converting the predicted trajectory, dynamic obstacle threat score and static area passage score output by the environmental perception model into a dynamic potential field function. Step S412: Lane constraint construction, used to express the driving constraints of the road structure on the vehicle, specifically by calculating the static potential field based on the vehicle's distance from the lane centerline and lane boundary to obtain the constraint potential field; Step S413: State space definition, specifically, constructing a state vector that makes up the state space, whose elements include vehicle position data, vehicle speed data, vehicle acceleration data, vehicle heading angle data, estimated travel time, threat potential field and constraint potential field, constructing an ant cell set, the search path set of the ant cell set in the state space, as the autonomous driving decision set, and the optimal search path as the optimal autonomous driving decision. Step S42: Design the objective function, which is used to design the path objective function required to optimize autonomous driving decisions; Step S43: design an ant colony state transition mechanism for global search path optimization, including the following steps: Step S431: pheromone field initialization, for initializing the pheromone distribution of ant colony search, specifically setting the initial pheromone concentration according to the risk cost to obtain the initial pheromone field guiding the search of the ant unit; Step S432: elite ant global search, for performing global optimal search path exploration, specifically guiding the movement of the elite ant unit through the transition probability formula combining pheromone concentration, potential field and kinematic energy constraints, the elite ant unit being specifically the ant unit with the path target function value sorted in ascending order in the top 30%; Step S44: double population collaborative optimization, for collaborative optimization of the elite ant unit and the non-elite ant unit, including the following steps: Step S441: dynamic update of the pheromone field, for updating the pheromone field according to the search situation to feed back to the next search, specifically updating the pheromone field based on the exponential decay of the path target function value of the elite ant unit to obtain the updated pheromone field; Step S442: local optimization of the non-elite ant unit, for fine search near the search path of the elite ant unit, specifically generating a new candidate path by applying a Gaussian disturbance with time decay to the search path of the elite ant unit; Step S45: obtaining the optimal search path, specifically designing the improved ant colony optimization algorithm through the state space construction, the design target function, the design ant colony state transition mechanism and the double population collaborative optimization, and continuously searching until the search stopping condition is reached to obtain the optimal search path as the optimal automatic driving decision, the search stopping condition specifically including reaching the maximum search round and the path target function value being less than a preset threshold. 3.The AI-based automatic driving environment perception method of claim 1, wherein: The past perception raw data set and the current perception raw data set both include ego data, dynamic environment data and static environment data, and the past perception raw data set further includes a past dynamic obstacle threat score and a past static region passage score. 4.The AI-based automatic driving environment perception method of claim 1, wherein: The raw data optimization specifically includes the following steps: Step S21: sensor calibration, for unifying the multi-sensor coordinate system and time stamp, specifically converting different sensor data to the vehicle coordinate system and aligning the time stamp through the interpolation method to obtain spatio-temporally aligned data; Step S22: dynamic and static data separation, for distinguishing moving targets and stationary backgrounds in the environment, specifically separating dynamic environment data and static environment data through the optical flow method; Step S23: data normalization, for eliminating the dimensional differences of different sensor data, specifically converting the data to a unified range using the minimum-maximum normalization method; Step S24: data augmentation, for increasing the diversity of training data, specifically performing random transformation on the original data to obtain expanded data, the random transformation including rotation, translation, scaling and adding noise conforming to physical laws; Step S25: data set segmentation, specifically data set segmentation of the past perception raw data set to obtain a perception training set and a perception test set; The current perception raw data set is optimized by the sensor calibration, the dynamic and static data separation and the data normalization to obtain a to-be-processed data set, and the past perception raw data set is optimized by the sensor calibration, the dynamic and static data separation, the data normalization, the data enhancement and the data set segmentation to obtain a perception training set and a perception test set.

5. The artificial intelligence-based automatic driving environment perception system for implementing the artificial intelligence-based automatic driving environment perception method according to any one of claims 1-4, characterized in that: The system comprises a multi-source data acquisition module, an original data optimization module, an environment perception model construction module, a decision optimization algorithm design module and an automatic driving decision module. 6.The artificial intelligence-based automatic driving environment perception system according to claim 5, characterized in that: The multi-source data acquisition module is configured to acquire original data, obtain an environment perception raw data set through the acquisition of the original data, and send the environment perception raw data set to the original data optimization module. The original data optimization module is configured to optimize the original data, obtain a to-be-processed data set, a perception training set and a perception test set through the optimization of the original data, and send the to-be-processed data set to the automatic driving decision module, and send the perception training set and the perception test set to the environment perception model construction module. The environment perception model construction module is configured to construct an environment perception model through the construction of a double-branch transformer model, and send the environment perception model to the automatic driving decision module. The decision optimization algorithm design module is configured to design a decision optimization algorithm through the use of an improved ant colony optimization algorithm, and send the decision optimization algorithm to the automatic driving decision module. The automatic driving decision module is configured to make an automatic driving decision through the real-time processing of data by the environment perception model to obtain a predicted trajectory, a dynamic obstacle threat score and a static region passing score, and real-time optimization of the automatic driving decision in combination with the decision optimization algorithm to obtain an optimal automatic driving decision and execute the optimal automatic driving decision.

Images