An Unmanned Driving Control Method and System Combining Dynamic Obstacle Trajectory Prediction

Abstract

The present invention relates to a driverless control method and system combining dynamic obstacle trajectory prediction, belonging to the technical field of driverless driving, including: after the vehicle starts, detecting the radar point cloud time series information within a preset radius through a millimeter wave radar deployed on the vehicle; performing moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information; performing moving state analysis on the moving target positioning time series information to obtain a moving rule dichotomy parameter, if it is equal to 0, classifying it as one type of moving target and retrieving a terrain grid model; based on the terrain grid model, performing terrain-guided trajectory analysis on the moving target to obtain moving target predicted position time series information for driverless control. It solves the technical problem in the prior art that due to the lack of training standard data, the avoidance rate of unpredictable obstacles is not high.

Description

Technical Field

[0001] The present invention relates to the technical field of driverless technology, and in particular, to a driverless control method and system combining dynamic obstacle trajectory prediction. Background Art

[0002] In the field of driverless technology, a vehicle needs to process a large amount of environmental information in real time during driving to ensure safe driving. Traditionally, driverless vehicles rely on a variety of sensors, such as cameras, lidar (LiDAR), and millimeter-wave radars, etc., to detect and identify surrounding obstacles. However, for dynamic obstacles, especially those that do not follow traffic rules or are unpredictable (such as stones), traditional prediction methods may not provide a sufficient avoidance rate.

[0003] Existing driverless systems often rely on machine learning and pattern recognition technologies when dealing with dynamic obstacles. These technologies require a large amount of labeled data to train the model. However, for unpredictable obstacles such as stones on mountain roads, it is very difficult to obtain sufficient training data, resulting in a low avoidance rate for unpredictable obstacles. Summary of the Invention

[0004] In view of the technical problem in the prior art that due to the lack of standard training data, the avoidance rate for unpredictable obstacles is low, the present invention provides a driverless control method and system combining dynamic obstacle trajectory prediction to solve this problem.

[0005] The technical solution of the present invention to solve the above technical problem is as follows:

[0006] In a first aspect, the present invention provides a driverless control method combining dynamic obstacle trajectory prediction, including: after the vehicle starts, detecting the radar point cloud time series information within a preset radius through a millimeter-wave radar deployed on the vehicle; performing moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information; performing moving state analysis on the moving target positioning time series information to obtain a moving rule dichotomy parameter; when the moving rule dichotomy parameter is equal to 0, classifying the moving target into a type of moving target, and at the same time, retrieving a terrain grid model within a preset radius through a road network map; performing terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain moving target predicted position time series information; and performing driverless control according to the moving target predicted position time series information.

[0007] In a second aspect, the present invention provides a combined dynamic obstacle trajectory prediction unmanned driving control system, including: a radar detection unit, configured to detect the time series information of radar point clouds within a preset radius through a millimeter-wave radar deployed on the vehicle after the vehicle starts; a moving target recognition unit, configured to perform moving target recognition on the time series information of the radar point clouds to obtain moving target positioning time series information; a moving state analysis unit, configured to perform moving state analysis on the moving target positioning time series information to obtain a moving law dichotomy parameter; a terrain model retrieval unit, configured to, when the moving law dichotomy parameter is equal to 0, classify the moving target into a type of moving target, and at the same time retrieve a terrain grid model within a preset radius through a road network map; a target moving prediction unit, configured to perform terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain moving target predicted position time series information; and an unmanned driving control unit, configured to perform unmanned driving control according to the moving target predicted position time series information.

[0008] In a third aspect, the present invention provides an electronic device, including: a memory, configured to store a computer software program; and a processor, configured to read and execute the computer software program, thereby implementing a combined dynamic obstacle trajectory prediction unmanned driving control method described in the first aspect.

[0009] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium, in which a computer software program is stored, and when the computer software program is executed by a processor, a combined dynamic obstacle trajectory prediction unmanned driving control method described in the first aspect is implemented.

[0010] The beneficial effects of the present invention are as follows: By using a millimeter-wave radar to detect the surrounding environment, real-time detection and recognition of dynamic obstacles are carried out. For irregular moving targets, such as stones, a terrain-guided trajectory analysis is used to predict their possible position time series information. This method does not rely on a large amount of training data, but uses a terrain grid model to assist in prediction, thereby improving the avoidance rate of unpredictable obstacles. In this way, it can respond more effectively to emergencies, achieving the technical effects of improving the safety and reliability of unmanned vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a schematic flowchart of a combined dynamic obstacle trajectory prediction unmanned driving control method provided by the present invention;

[0012] Figure 2 is a schematic structural diagram of a combined dynamic obstacle trajectory prediction unmanned driving control system provided by the present invention;

[0013] Figure 3 is a schematic structural diagram of the electronic device provided by the present invention;

[0014] Figure 4 Schematic diagram of a computer-readable storage medium provided by the present invention.

[0015] In the drawings, the components represented by the reference numerals are described as follows:

[0016] Electronic device 500, memory 510, processor 520, first computer program 511, computer-readable storage medium 600, second computer program 611. Specific embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0018] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes, and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0019] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or having more advantages than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present invention.

[0020] Embodiment 1:

[0021] As Figure 1 shown, the embodiment of the present invention provides a combined dynamic obstacle trajectory prediction unmanned driving control method, including the steps of:

[0022] S10: After the vehicle starts, detect the radar point cloud time series information within a preset radius range through the millimeter-wave radar deployed on the vehicle;

[0023] Specifically, the preset radius range refers to the distance threshold for screening point clouds. Point clouds with a distance greater than the preset radius range are considered to pose no immediate threat to the vehicle and do not need to be considered for the time being. Specifically, the radar point cloud timing information refers to the point cloud distribution data detected by the millimeter-wave radar at consecutive moments within the preset radius range.

[0024] The detection steps are as follows: After the vehicle starts, the millimeter-wave radar system begins to operate and prepares to collect data. The radar emits millimeter waves and receives the reflected signals. By processing the received signals, the radar can calculate information such as the distance, speed, and angle of the object and convert it into point cloud data. According to the preset radius range, point clouds with a distance greater than the preset radius range from the radar are deleted. The remaining point clouds are stored according to the timing sequence to obtain the radar point cloud timing information.

[0025] The millimeter-wave radar provides precise environmental perception capabilities for driverless vehicles, enabling them to understand and predict dynamic changes in the surrounding environment. Through the point cloud timing information collected by the millimeter-wave radar, the vehicle can detect and track dynamic obstacles and obtain the motion information of the obstacles, which is crucial for subsequent prediction of the motion trajectories of the obstacles and generation of obstacle avoidance paths.

[0026] S20: Perform moving target recognition on the radar point cloud timing information to obtain moving target positioning timing information;

[0027] Specifically, moving target recognition refers to the process of distinguishing stationary objects and moving objects from these continuous point cloud data, with the aim of identifying which point clouds represent moving targets. Moving target positioning timing information refers to the data obtained by continuously recording the positions and motion states of these targets over time after the moving targets are identified.

[0028] The specific process of moving target recognition includes: extracting features from the point cloud data that are helpful for identifying moving targets, such as speed, acceleration, shape, and size; using the extracted features to analyze moving targets and the stationary background; tracking the detected moving targets and recording the changes in their positions and motion states over time to generate moving target positioning timing information.

[0029] Providing precise information about surrounding moving targets for driverless vehicles is crucial for the safe driving of the vehicle. Through moving target recognition and positioning, the vehicle can identify and distinguish dynamic and static obstacles around the vehicle, providing a basis for obstacle avoidance and path planning.

[0030] S30: Perform moving state analysis on the moving target positioning timing information to obtain moving pattern binary parameters;

[0031] Further, perform a moving state analysis on the moving target positioning timing information to obtain a moving pattern dichotomy parameter, including:

[0032] Analyze the first velocity direction and the first velocity magnitude of the positioning information at the first moment and the positioning information at the second moment of the moving target positioning timing information;

[0033] Until analyzing the (N - 1)th velocity direction and the (N - 1)th velocity magnitude of the positioning information at the (N - 1)th moment and the positioning information at the Nth moment of the moving target positioning timing information;

[0034] Calculate the first deviation angle between the first velocity direction and the second velocity direction, until calculating the (N - 2)th deviation angle between the (N - 2)th velocity direction and the (N - 1)th velocity direction, and statistically calculate the first variance parameter from the first deviation angle to the (N - 2)th deviation angle;

[0035] Calculate the first deviation magnitude between the first velocity magnitude and the second velocity magnitude, until calculating the (N - 2)th deviation magnitude between the (N - 2)th velocity magnitude and the (N - 1)th velocity magnitude, and statistically calculate the second variance parameter from the first deviation magnitude to the (N - 2)th deviation magnitude;

[0036] When the first variance parameter is greater than the first variance parameter threshold, or / and the second variance parameter is greater than the second variance parameter threshold, the moving pattern dichotomy parameter is equal to 0;

[0037] Otherwise, the moving pattern dichotomy parameter is equal to 1.

[0038] Specifically, the moving pattern dichotomy parameter is a parameter obtained from the moving state analysis, which classifies the movement patterns of moving targets into two categories. For example, one category is targets that follow specific traffic rules, and the other category is targets that do not follow rules or whose behavior patterns are difficult to predict (such as obstacles like stones). Moving state analysis refers to analyzing the moving regularity of a moving target based on the moving target positioning timing information to determine whether it is a predictable dynamic obstacle. Preferably, when it belongs to a target that does not follow rules or whose behavior pattern is difficult to predict, the moving pattern dichotomy parameter is equal to 0; otherwise, the moving pattern dichotomy parameter is equal to 1.

[0039] When the moving pattern dichotomy parameter is equal to 1, existing algorithms for target trajectory prediction can be used to predict and avoid the target trajectory. When the moving pattern dichotomy parameter is equal to 0, the algorithm steps set in the embodiments of the present application need to be executed to predict and avoid the target trajectory.

[0040] Specifically, the steps of the moving state analysis specifically include:

[0041] By comparing the positioning information at two consecutive moments, the velocity direction and velocity magnitude of the moving target are analyzed. The velocity direction refers to the direction of the target's movement, while the velocity magnitude refers to the distance the target moves within a unit of time. The velocity direction of the positioning information at two consecutive moments refers to the azimuth vector from the target position at the previous consecutive moment to the target position at the next consecutive moment, and the velocity magnitude of the positioning information at two consecutive moments refers to the absolute value of the distance between the target position at the previous consecutive moment and the target position at the next consecutive moment.

[0042] Based on the above principle, the positioning information at the first moment and the positioning information at the second moment of the moving target positioning time series information are extracted, and then the first velocity direction and the first velocity magnitude from the positioning information at the first moment to the positioning information at the second moment are analyzed; further, the positioning information at the second moment and the positioning information at the third moment of the moving target positioning time series information are extracted, and then the second velocity direction and the second velocity magnitude from the positioning information at the second moment to the positioning information at the third moment are analyzed... until the (N - 1)th velocity direction and the (N - 1)th velocity magnitude from the positioning information at the (N - 1)th moment and the positioning information at the Nth moment of the moving target positioning time series information are extracted, where N represents the total number of moments.

[0043] Then, the first deviation angle between the first velocity direction and the second velocity direction is calculated, until the (N - 2)th deviation angle between the (N - 2)th velocity direction and the (N - 1)th velocity direction is calculated, and the variance of the first deviation angle to the (N - 2)th deviation angle is calculated to obtain the first variance parameter.

[0044] In addition, the first deviation magnitude between the first velocity magnitude and the second velocity magnitude is calculated, until the (N - 2)th deviation magnitude between the (N - 2)th velocity magnitude and the (N - 1)th velocity magnitude is calculated, and the variance of the first deviation magnitude to the (N - 2)th deviation magnitude is calculated to obtain the second variance parameter.

[0045] By statistically analyzing the variance parameters of a series of deviation angles and deviation magnitudes, the stability and regularity of the target's motion state can be quantified. The variance parameter is a statistic that measures the degree of data dispersion. The larger the variance, the greater the data fluctuation, and the greater the fluctuation, the weaker the movement regularity. When either the first variance parameter is greater than the first variance parameter threshold or the second variance parameter is greater than the second variance parameter threshold occurs, the movement regularity binary parameter is equal to 0. The first variance parameter threshold and the second variance parameter threshold are both parameters that can be custom-set by users in specific applications. The larger the first variance parameter threshold and the second variance parameter threshold, the greater the user's tolerance for targets with irregular movement, and vice versa. Preferably, in urban and plain areas, the first variance parameter threshold and the second variance parameter threshold can be set larger because the probability of uncontrollable dynamic obstacles appearing in these areas is relatively low. While in rural and mountainous areas, the first variance parameter threshold and the second variance parameter threshold can be set smaller to ensure driving safety. In the above screening scheme, some non-threatening moving targets will also be classified as uncontrollable dynamic obstacles, which is the accuracy that can be sacrificed to ensure the safety of driverless driving. In other cases, that is, when the first variance parameter is less than or equal to the first variance parameter threshold and the second variance parameter is less than or equal to the second variance parameter threshold, the movement regularity binary parameter is equal to 1, indicating that the moving target is a conventional movement obstacle, and then the existing trajectory prediction algorithm can be executed.

[0046] By providing a quantitative analysis of the motion state of surrounding moving targets for the driverless system, the system can perform reasonable path planning and obstacle avoidance decisions according to the motion laws of the targets.

[0047] S40: When the movement regularity binary parameter is equal to 0, classify the moving target into a category of moving targets, and at the same time, through the road network map, retrieve the terrain grid model within the preset radius.

[0048] S50: Based on the terrain grid model, perform a terrain-guided trajectory analysis on the moving target to obtain the predicted position time series information of the moving target.

[0049] Further, before performing a terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain the predicted position time series information of the moving target, it includes:

[0050] Based on the terrain grid model, perform a terrain-guided weight analysis on the positioning time series information of the moving target to obtain the terrain-guided weight coefficient.

[0051] When the terrain-guided weight coefficient is greater than or equal to the terrain-guided weight coefficient threshold, perform a terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain the predicted position time series information of the moving target.

[0052] When the terrain guidance weight coefficient is less than the terrain guidance weight coefficient threshold, a safety warning signal is generated based on the real-time position of the moving target for warning the central control platform.

[0053] Further, a terrain guidance weight analysis is performed on the positioning time series information of the moving target based on the terrain grid model to obtain a terrain guidance weight coefficient, including:

[0054] Perform a terrain guidance trajectory analysis on the moving target positioning information at the first moment of the moving target positioning time series information based on the terrain grid model to obtain the moving target terrain guidance positioning information at the second moment;

[0055] Until a terrain guidance trajectory analysis is performed on the moving target positioning information at the (N - 1)-th moment of the moving target positioning time series information based on the terrain grid model to obtain the moving target terrain guidance positioning information at the N-th moment;

[0056] Calculate the first positioning distance between the moving target terrain guidance positioning information at the second moment and the moving target positioning information at the second moment;

[0057] Until the N-th positioning distance between the moving target terrain guidance positioning information at the N-th moment and the moving target positioning information at the N-th moment is calculated;

[0058] Statistically analyze the proportion of the first positioning distance to the N-th positioning distance that is less than or equal to the positioning distance threshold, and set it as the terrain guidance weight coefficient.

[0059] Further, a terrain guidance trajectory analysis is performed on the moving target based on the terrain grid model to obtain the moving target predicted position time series information, including:

[0060] Obtain the real-time position of the moving target and the real-time speed of the moving target;

[0061] Input the real-time position of the moving target and the terrain grid model into a moving trajectory matching model for analysis to obtain a moving prediction trajectory;

[0062] Obtain the terrain height sequence of the moving prediction trajectory;

[0063] Input the real-time speed of the moving target and the terrain height sequence into a moving speed matching model for analysis to obtain a moving speed sequence;

[0064] Fuse the moving speed sequence and the moving prediction trajectory to obtain the moving target predicted position time series information.

[0065] Specifically, when the movement rule binary parameter is equal to 0, it indicates that the moving target is classified into a special type of moving target, which may not follow conventional traffic rules or behavior patterns, such as obstacles like stones. In response to such a situation, the embodiments of the present application analyze a type of moving target through a custom trajectory prediction algorithm, as detailed below:

[0066] The road network map represents the big data map, which stores the terrain map of the vehicle driving area. A terrain grid model within a preset radius is selected. The terrain grid model refers to the terrain map model obtained by dividing the terrain map according to a preset grid. Since most of the obstacles that do not follow the convention are items and cannot dominate their own moving directions, the only trajectory influence comes from gravity and terrain. Therefore, the moving situation of the moving target can be analyzed based on the terrain and gravity, and then the predicted position time series information of the moving target within a future time interval can be specified.

[0067] Specifically, the terrain-guided trajectory analysis process is as follows:

[0068] The real-time position of the moving target and the real-time position of the moving target refer to the instant dynamic data obtained by continuously tracking and positioning the moving target through a millimeter-wave radar. The moving trajectory matching model is used to combine the real-time position and historical trajectory of the moving target with the terrain grid model to predict the future movement trajectory of the target; the moving predicted trajectory is the future movement trajectory of the moving target.

[0069] The terrain height sequence refers to the terrain height data sequence corresponding one-to-one to the positions of the moving predicted trajectory; the moving speed matching model is used to analyze the relationship between the real-time speed of the moving target and the terrain height sequence to predict the speed change of the target on different terrains; the moving speed sequence is the speed change sequence of the moving target on the predicted trajectory obtained by analyzing with the moving speed matching model.

[0070] The predicted position time series information of the moving target refers to the position prediction information of the moving target at each future time point obtained by fusing the moving speed sequence and the moving predicted trajectory. Given the speed information of the moving target at any two points, the position prediction information of the moving target on the trajectory can be analyzed.

[0071] Specifically, both the moving trajectory matching model and the moving speed matching model are long short-term memory neural network topologies. Preferably, the training steps of the moving trajectory matching model include: collecting moving target position record data, terrain grid record models, and moving trajectory identification data; constructing a moving trajectory matching loss function: , a > 1, represents the matching trajectory, represents the identification trajectory, represents the area enclosed by the matching trajectory and the identification trajectory, Characterize the moving trajectory matching loss value. Using the moving trajectory identification data as supervision, and taking the moving target position recording data and the terrain grid recording model as inputs, train the long short-term memory neural network topology. When the loss is less than or equal to the loss threshold for a continuous preset number of times, generate a moving trajectory matching model. The training steps of the moving speed matching model include: collecting moving speed sequence recording data, moving predicted trajectory recording data, and moving target predicted position time series identification data; constructing a moving speed matching loss function: , Characterize the moving speed matching loss value, characterize the distance between the predicted position at the i-th moment and the identified position at the i-th moment, characterize the predicted position at the i-th moment, characterize the identified position at the i-th moment, characterize the total number of preset moments, b > 1. Using the moving speed sequence recording data and the moving predicted trajectory recording data as inputs to the long short-term memory neural network, and using the moving target predicted position time series identification data as supervision for training. When the loss is less than or equal to the loss threshold for a continuous preset number of times, generate a moving speed matching model. Through the moving speed matching model and the moving trajectory matching model, realize the fusion analysis of the motion state of the moving target and the terrain, obtain the motion trajectory time series information of obstacles such as stones, and provide a basis for effective obstacle avoidance in the subsequent steps.

[0072] In a preferred embodiment, the terrain guiding weight coefficient refers to the degree of measuring the influence of terrain factors on the prediction of the moving target trajectory. If the terrain guiding weight coefficient is less than the terrain guiding weight coefficient threshold preset by the user, it means that the correlation between the moving target and the terrain is not significant. At this time, using the terrain guiding predicted trajectory is meaningless. In this case, if the nature of the moving target is not clear, a safety warning signal is generated according to the real-time position of the moving target for warning on the central control platform to remind the user to make a manual decision. When the terrain guiding weight coefficient is greater than or equal to the terrain guiding weight coefficient threshold, based on the terrain grid model, perform terrain guiding trajectory analysis on the moving target to obtain the moving target predicted position time series information.

[0073] In a preferred embodiment, the detailed configuration process of the terrain guiding weight coefficient is as follows:

[0074] Perform the above terrain-guided trajectory analysis on the moving target positioning information at the first moment of the moving target positioning timing information based on the terrain grid model to obtain the moving target terrain-guided positioning information at the second moment; until the terrain-guided trajectory analysis is performed on the moving target positioning information at the (N-1)th moment of the moving target positioning timing information based on the terrain grid model to obtain the moving target terrain-guided positioning information at the Nth moment;

[0075] Then calculate the Euclidean distance between the moving target terrain-guided positioning information at the second moment and the moving target positioning information at the second moment, and store it as the first positioning distance; until the Euclidean distance between the moving target terrain-guided positioning information at the Nth moment and the moving target positioning information at the Nth moment is calculated and stored as the Nth positioning distance; count the proportion of the positioning distances from the first positioning distance to the Nth positioning distance that are less than or equal to the positioning distance threshold, and set it as the terrain-guided weight coefficient.

[0076] S60: Perform driverless control according to the moving target predicted position timing information.

[0077] Specifically, after determining the moving target predicted position timing information, when there is an intersection with the preset path of the driverless vehicle, deceleration, acceleration or detour avoidance path planning is performed for control.

[0078] Further, perform moving target recognition on the radar point cloud timing information to obtain the moving target positioning timing information, including:

[0079] Traverse the radar point cloud timing information for outlier removal analysis to obtain a dense radar point cloud timing diagram;

[0080] Segment the dense radar point cloud timing diagram according to the first segmentation step length to obtain the first dense radar point cloud timing diagram;

[0081] Segment the dense radar point cloud timing diagram according to one-eighth of the first segmentation step length to obtain the second dense radar point cloud timing diagram;

[0082] Input the first dense radar point cloud timing diagram into the moving target positioning channel of the moving target recognition network, and input the second dense radar point cloud timing diagram into the moving target segmentation channel of the moving target recognition network for processing to obtain the moving target positioning timing information.

[0083] Specifically, in order to achieve the rapid sorting of lightning point cloud data in the embodiments of the present application, the radar point cloud is processed through the SLOWFAST neural network model to identify the moving target positioning timing information. The preferred process is as follows:

[0084] For each moment of the radar point cloud time series information, outliers are removed from the points according to their position distribution to obtain the concentrated point cloud at each moment, which is stored as a dense radar point cloud time series graph. The outlier analysis algorithm can be implemented using the Local Outlier Factor (LOF) algorithm. Calculate the outlier factor of each point at a certain moment, and delete the points whose outlier factor is greater than the outlier factor threshold. The remaining points are regarded as the dense radar point cloud at that moment.

[0085] Then, the dense radar point cloud time series graph is segmented according to one-eighth of the first segmentation step length to obtain a second dense radar point cloud time series graph. Specifically, the unit of the first segmentation step length is milliseconds. The dense radar point cloud time series graph is segmented according to one-eighth of the first segmentation step length to obtain a second dense radar point cloud time series graph. The first dense radar point cloud time series graph is input into the moving target localization channel of the moving target recognition network, and the second dense radar point cloud time series graph is input into the moving target segmentation channel of the moving target recognition network for processing to obtain the moving target localization time series information. The moving target localization channel is the SLOW channel, which can achieve the localization of the target, and the moving target segmentation channel is the FAST channel, which can achieve the fast tracking of the target.

[0086] Furthermore, the construction steps of the moving target recognition network include:

[0087] Collect radar point cloud time series graph record data and target position identification time series information;

[0088] Segment the radar point cloud time series graph record data according to the first segmentation step length to obtain the first radar point cloud time series graph record data;

[0089] Segment the radar point cloud time series graph record data according to one-eighth of the first segmentation step length to obtain the second radar point cloud time series graph record data;

[0090] Input the first radar point cloud time series graph record data into the SLOW channel, and input the second radar point cloud time series graph record data into the FAST channel to obtain the target position prediction time series information;

[0091] Take the proportion of the moments in the position deviation distance time series information between the target position identification time series information and the target position prediction time series information where the position deviation distance is greater than or equal to the position deviation distance threshold as the moving target recognition loss value;

[0092] When the moving target recognition loss value is less than or equal to the moving target recognition loss value threshold in a continuous preset number of training sessions, generate the moving target recognition network.

[0093] Specifically, both the radar point cloud time series diagram recording data and the target position identification time series information are training data pre-identified by the user. The radar point cloud time series diagram recording data is segmented according to the first segmentation step length to obtain the first radar point cloud time series diagram recording data; the radar point cloud time series diagram recording data is segmented according to one-eighth of the first segmentation step length to obtain the second radar point cloud time series diagram recording data. The first radar point cloud time series diagram recording data is input into the SLOW channel, and the second radar point cloud time series diagram recording data is input into the FAST channel to obtain the target position prediction time series information; the proportion of the moments when the position deviation distance in the position deviation distance time series information between the target position identification time series information and the target position prediction time series information is greater than or equal to the position deviation distance threshold is used as the moving target recognition loss value; when the moving target recognition loss value is less than or equal to the moving target recognition loss value threshold for at least a preset number of times in a continuous preset number of training sessions, it is considered convergent, and then a moving target recognition network is generated. Through the moving target recognition network, the recognition and sorting of moving targets can be quickly realized, improving timeliness.

[0094] A combined dynamic obstacle trajectory prediction unmanned driving control method provided by an embodiment of the present invention has at least the following technical effects:

[0095] A technical solution that uses a millimeter-wave radar to detect the surrounding environment, performs real-time detection and recognition of dynamic obstacles, and for irregular moving targets, such as stones, predicts their possible position time series information through terrain-guided trajectory analysis. This method does not rely on a large amount of training data, but uses a terrain grid model to assist in prediction, thereby increasing the avoidance rate of unpredictable obstacles. In this way, it can respond more effectively to emergencies, achieving the technical effects of improving the safety and reliability of unmanned vehicles.

[0096] Embodiment 2:

[0097] As Figure 2 shown, based on the same inventive concept as the combined dynamic obstacle trajectory prediction unmanned driving control method provided in Embodiment 1, an embodiment of the present invention also provides a combined dynamic obstacle trajectory prediction unmanned driving control system, including:

[0098] A radar detection unit, configured to detect radar point cloud time series information within a preset radius through a millimeter-wave radar deployed on the vehicle after the vehicle starts;

[0099] A moving target recognition unit, configured to perform moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information;

[0100] A moving state analysis unit, configured to perform moving state analysis on the moving target positioning time series information to obtain a moving pattern dichotomy parameter;

[0101] A terrain model retrieval unit, configured to, when the movement law dichotomy parameter is equal to 0, classify a moving target into one type of moving target, and meanwhile, retrieve a terrain grid model within a preset radius through a road network map;

[0102] A target movement prediction unit, configured to analyze a terrain-guided trajectory of the moving target based on the terrain grid model to obtain time-sequence information of the predicted position of the moving target;

[0103] An unmanned driving control unit, configured to perform unmanned driving control according to the time-sequence information of the predicted position of the moving target.

[0104] Further, before analyzing the terrain-guided trajectory of the moving target based on the terrain grid model to obtain the time-sequence information of the predicted position of the moving target, it includes:

[0105] Analyzing the terrain-guided weight of the time-sequence information of the positioning of the moving target based on the terrain grid model to obtain a terrain-guided weight coefficient;

[0106] When the terrain-guided weight coefficient is greater than or equal to a terrain-guided weight coefficient threshold, analyzing the terrain-guided trajectory of the moving target based on the terrain grid model to obtain the time-sequence information of the predicted position of the moving target;

[0107] When the terrain-guided weight coefficient is less than the terrain-guided weight coefficient threshold, generating a safety warning signal according to the real-time position of the moving target for warning the central control platform.

[0108] Further, analyzing the terrain-guided weight of the time-sequence information of the positioning of the moving target based on the terrain grid model to obtain a terrain-guided weight coefficient includes:

[0109] Analyzing the terrain-guided trajectory of the positioning information of the moving target at the first moment of the time-sequence information of the positioning of the moving target based on the terrain grid model to obtain the terrain-guided positioning information of the moving target at the second moment;

[0110] Until analyzing the terrain-guided trajectory of the positioning information of the moving target at the (N - 1)-th moment of the time-sequence information of the positioning of the moving target based on the terrain grid model to obtain the terrain-guided positioning information of the moving target at the N-th moment;

[0111] Calculating a first positioning distance between the terrain-guided positioning information of the moving target at the second moment and the positioning information of the moving target at the second moment;

[0112] Until calculating an N-th positioning distance between the terrain-guided positioning information of the moving target at the N-th moment and the positioning information of the moving target at the N-th moment;

[0113] Statistically calculate the proportion of the first positioning distance to the Nth positioning distance that is less than or equal to the positioning distance threshold, and set it as the terrain guidance weight coefficient.

[0114] Further, perform moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information, including:

[0115] Traverse the radar point cloud time series information for outlier analysis to obtain a dense radar point cloud time series diagram;

[0116] Segment the dense radar point cloud time series diagram according to the first segmentation step length to obtain the first dense radar point cloud time series diagram;

[0117] Segment the dense radar point cloud time series diagram according to one-eighth of the first segmentation step length to obtain the second dense radar point cloud time series diagram;

[0118] Input the first dense radar point cloud time series diagram into the moving target positioning channel of the moving target recognition network, and input the second dense radar point cloud time series diagram into the moving target segmentation channel of the moving target recognition network for processing to obtain the moving target positioning time series information.

[0119] Further, the construction steps of the moving target recognition network include:

[0120] Collect radar point cloud time series diagram record data and target position identification time series information;

[0121] Segment the radar point cloud time series diagram record data according to the first segmentation step length to obtain the first radar point cloud time series diagram record data;

[0122] Segment the radar point cloud time series diagram record data according to one-eighth of the first segmentation step length to obtain the second radar point cloud time series diagram record data;

[0123] Input the first radar point cloud time series diagram record data into the SLOW channel, and input the second radar point cloud time series diagram record data into the FAST channel to obtain target position prediction time series information;

[0124] Use the proportion of the moments in the position deviation distance time series information of the target position identification time series information and the target position prediction time series information where the position deviation distance is greater than or equal to the position deviation distance threshold as the moving target recognition loss value;

[0125] When the moving target recognition loss value is less than or equal to the moving target recognition loss value threshold for a preset number of consecutive training times, generate the moving target recognition network.

[0126] Further, perform a moving state analysis on the moving target positioning timing information to obtain a moving pattern dichotomy parameter, including:

[0127] Analyze the first moment positioning information, the first velocity direction, and the first velocity magnitude of the second moment positioning information of the moving target positioning timing information;

[0128] Until analyzing the (N - 1)th moment positioning information, the (N - 1)th velocity direction, and the (N - 1)th velocity magnitude of the Nth moment positioning information of the moving target positioning timing information;

[0129] Calculate the first deviation angle between the first velocity direction and the second velocity direction, until calculating the (N - 2)th deviation angle between the (N - 2)th velocity direction and the (N - 1)th velocity direction, and statistically calculate the first variance parameter from the first deviation angle to the (N - 2)th deviation angle;

[0130] Calculate the first deviation magnitude between the first velocity magnitude and the second velocity magnitude, until calculating the (N - 2)th deviation magnitude between the (N - 2)th velocity magnitude and the (N - 1)th velocity magnitude, and statistically calculate the second variance parameter from the first deviation magnitude to the (N - 2)th deviation magnitude;

[0131] When the first variance parameter is greater than the first variance parameter threshold, or / and the second variance parameter is greater than the second variance parameter threshold, the moving pattern dichotomy parameter is equal to 0;

[0132] Otherwise, the moving pattern dichotomy parameter is equal to 1.

[0133] Further, based on the terrain grid model, perform a terrain guiding trajectory analysis on the moving target to obtain the moving target predicted position timing information, including:

[0134] Obtain the real - time position of the moving target and the real - time speed of the moving target;

[0135] Input the real - time position of the moving target and the terrain grid model into the moving trajectory matching model for analysis to obtain the moving prediction trajectory;

[0136] Obtain the terrain height sequence of the moving prediction trajectory;

[0137] Input the real - time speed of the moving target and the terrain height sequence into the moving speed matching model for analysis to obtain the moving speed sequence;

[0138] Fuse the moving speed sequence and the moving prediction trajectory to obtain the moving target predicted position timing information.

[0139] Embodiment III:

[0140] Please refer to Figure 3 , Figure 3Schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. As Figure 3 shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a first computer program 511 stored on the memory 510 and executable on the processor 520. When the processor 520 executes the first computer program 511, the following steps are implemented:

[0141] After the vehicle starts, detect the radar point cloud time series information within a preset radius through a millimeter-wave radar deployed on the vehicle;

[0142] Perform moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information;

[0143] Perform moving state analysis on the moving target positioning time series information to obtain a moving rule dichotomy parameter;

[0144] When the moving rule dichotomy parameter is equal to 0, classify the moving target into a type of moving target, and at the same time, through the road network map, retrieve the terrain grid model within a preset radius;

[0145] Based on the terrain grid model, perform terrain guidance trajectory analysis on the moving target to obtain moving target predicted position time series information;

[0146] Perform driverless control according to the moving target predicted position time series information.

[0147] Embodiment 4:

[0148] Please refer to Figure 4 , Figure 4 Schematic diagram of an embodiment of a computer-readable storage medium provided by an embodiment of the present invention. As Figure 4 shown, this embodiment provides a computer-readable storage medium 600, on which a second computer program 611 is stored. When the second computer program 611 is executed by a processor, the following steps are implemented:

[0149] After the vehicle starts, detect the radar point cloud time series information within a preset radius through a millimeter-wave radar deployed on the vehicle;

[0150] Perform moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information;

[0151] Perform moving state analysis on the moving target positioning time series information to obtain a moving rule dichotomy parameter;

[0152] When the moving rule dichotomy parameter is equal to 0, classify the moving target into a type of moving target, and at the same time, through the road network map, retrieve the terrain grid model within a preset radius;

[0153] Perform terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain the time-series information of the predicted positions of the moving target;

[0154] Perform driverless control according to the time-series information of the predicted positions of the moving target.

[0155] It should be noted that in the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not described in detail in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0156] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0157] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0158] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0159] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0160] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn of the basic inventive concept.

[0161] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A driverless control method combined with dynamic obstacle trajectory prediction, characterized in that, Including: After the vehicle starts, detect the radar point cloud time series information within a preset radius through a millimeter-wave radar deployed on the vehicle; Perform moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information; Perform moving state analysis on the moving target positioning time series information to obtain a moving pattern dichotomy parameter; When the moving pattern dichotomy parameter is equal to 0, classify the moving target into one type of moving target, and at the same time, through the road network map, retrieve the terrain grid model within the preset radius; Perform terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain moving target predicted position time series information; Perform driverless control according to the moving target predicted position time series information; Among them, performing moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information includes: Traverse the radar point cloud time series information for outlier analysis to obtain a dense radar point cloud time series map; Segment the dense radar point cloud time series map according to the first segmentation step size to obtain a first dense radar point cloud time series map; Segment the dense radar point cloud time series map according to one-eighth of the first segmentation step size to obtain a second dense radar point cloud time series map; Input the first dense radar point cloud time series map into the moving target positioning channel of the moving target recognition network, and input the second dense radar point cloud time series map into the moving target segmentation channel of the moving target recognition network for processing to obtain the moving target positioning time series information; Among them, performing moving state analysis on the moving target positioning time series information to obtain a moving pattern dichotomy parameter includes: Analyze the first velocity direction and the first velocity magnitude of the positioning information at the first moment and the positioning information at the second moment of the moving target positioning time series information; Until analyzing the (N - 1)th velocity direction and the (N - 1)th velocity magnitude of the positioning information at the (N - 1)th moment and the positioning information at the Nth moment of the moving target positioning time series information; Calculate the first deviation angle between the first velocity direction and the second velocity direction, until calculating the (N - 2)th deviation angle between the (N - 2)th velocity direction and the (N - 1)th velocity direction, and statistically calculate the first variance parameter of the first deviation angle until the (N - 2)th deviation angle; Calculate the first deviation magnitude between the first velocity magnitude and the second velocity magnitude, until calculating the (N - 2)th deviation magnitude between the (N - 2)th velocity magnitude and the (N - 1)th velocity magnitude, and statistically calculate the second variance parameter of the first deviation magnitude until the (N - 2)th deviation magnitude; When the first variance parameter is greater than the first variance parameter threshold, or / and the second variance parameter is greater than the second variance parameter threshold, the moving pattern dichotomy parameter is equal to 0; Otherwise, the moving pattern dichotomy parameter is equal to 1; Among them, performing terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain moving target predicted position time series information includes: Obtain the real-time position of the moving target and the real-time velocity of the moving target; Input the real-time position of the moving target and the terrain grid model into the moving trajectory matching model for analysis to obtain a moving prediction trajectory; Obtain the terrain height sequence of the moving prediction trajectory; Input the real-time speed of the moving target and the terrain height sequence into the moving speed matching model for analysis to obtain a moving speed sequence; Fuse the moving speed sequence and the moving prediction trajectory to obtain the timing information of the predicted position of the moving target.

2. The method according to claim 1, wherein Performing terrain guidance trajectory analysis on the moving target based on the terrain grid model to obtain the timing information of the predicted position of the moving target previously included: Performing terrain guidance weight analysis on the timing information of the moving target positioning based on the terrain grid model to obtain a terrain guidance weight coefficient; When the terrain guidance weight coefficient is greater than or equal to the terrain guidance weight coefficient threshold, perform terrain guidance trajectory analysis on the moving target based on the terrain grid model to obtain the timing information of the predicted position of the moving target; When the terrain guidance weight coefficient is less than the terrain guidance weight coefficient threshold, generate a safety warning signal based on the real-time position of the moving target for warning on the central control platform.

3. The method according to claim 2, wherein Performing terrain guidance weight analysis on the timing information of the moving target positioning based on the terrain grid model to obtain a terrain guidance weight coefficient, including: Performing terrain guidance trajectory analysis on the moving target positioning information at the first moment of the timing information of the moving target positioning based on the terrain grid model to obtain the terrain guidance positioning information of the moving target at the second moment; Until performing terrain guidance trajectory analysis on the moving target positioning information at the (N - 1)th moment of the timing information of the moving target positioning based on the terrain grid model to obtain the terrain guidance positioning information of the moving target at the Nth moment; Calculate the first positioning distance between the terrain guidance positioning information of the moving target at the second moment and the moving target positioning information at the second moment; Until calculating the Nth positioning distance between the terrain guidance positioning information of the moving target at the Nth moment and the moving target positioning information at the Nth moment; Statistically calculate the proportion of those less than or equal to the positioning distance threshold from the first positioning distance to the Nth positioning distance, and set it as the terrain guidance weight coefficient.

4. The method according to claim 1, wherein The construction steps of the moving target recognition network include: Collect radar point cloud time series map recording data and target position identification time series information; Segment the radar point cloud time series map recording data according to the first segmentation step length to obtain the first radar point cloud time series map recording data; Segment the radar point cloud time series map recording data according to one-eighth of the first segmentation step length to obtain the second radar point cloud time series map recording data; Input the first radar point cloud time series map recording data into the SLOW channel of the SLOWFAST neural network model, and input the second radar point cloud time series map recording data into the FAST channel of the SLOWFAST neural network model to obtain the target position prediction time series information; Use the proportion of the moments in the position deviation distance time series information of the position deviation distance between the target position identification time series information and the target position prediction time series information that is greater than or equal to the position deviation distance threshold as the moving target recognition loss value; When the moving target recognition loss value is less than or equal to the moving target recognition loss value threshold in the preset number of continuous training sessions, generate the moving target recognition network.

5. A driverless control system combined with dynamic obstacle trajectory prediction, characterized in that, For implementing the method according to any one of claims 1 to 4, including: A radar detection unit, configured to detect the radar point cloud time series information within a preset radius through a millimeter wave radar deployed on the vehicle after the vehicle starts; A moving target recognition unit, configured to perform moving target recognition on the radar point cloud time series information to obtain moving target positioning time series information; A moving state analysis unit, configured to perform moving state analysis on the moving target positioning time series information to obtain a moving rule dichotomy parameter; A terrain model retrieval unit, configured to classify the moving target into a type of moving target when the moving rule dichotomy parameter is equal to 0, and at the same time retrieve the terrain grid model within a preset radius through the road network map; A target movement prediction unit, configured to perform terrain-guided trajectory analysis on the moving target based on the terrain grid model to obtain moving target predicted position time series information; An unmanned driving control unit, configured to perform unmanned driving control according to the moving target predicted position time series information.

6. An electronic device, characterized in that, Comprising: A memory, configured to store a computer software program; A processor, configured to read and execute the computer software program, thereby implementing a combined dynamic obstacle trajectory prediction unmanned driving control method according to any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium, characterized in that, A computer software program is stored in the storage medium, and when the computer software program is executed by the processor, a combined dynamic obstacle trajectory prediction unmanned driving control method according to any one of claims 1 to 4 is implemented.

Images