Method for making training data set and method for judging height attribute of obstacle

Abstract

The invention provides a training data set making method and an obstacle height attribute judgment method, and the training data set making method comprises the steps: constructing a training data set used for training a height classification model, and the training data set comprises a plurality of pieces of sample data; wherein the step of constructing any piece of sample data comprises: for any type of sample obstacle with known height, acquiring sample echo data obtained by detecting the sample obstacle by an ultrasonic radar, and analyzing sample echo parameters used for describing the characteristics of the sample obstacle from the sample echo data. And any piece of sample data is constructed based on the sample echo parameter, feature data in the piece of sample data comprises the sample echo parameter, and a label in the piece of sample data is used for representing a height category of the sample obstacle relative to a preset height.

Description

Technical Field

[0001] The present specification relates to the field of vehicle technology, and in particular to a method for preparing a training data set and a method for determining the height attribute of an obstacle. Background Art

[0002] With the development of autonomous driving and intelligent assisted driving technology, the collision detection system of vehicles is becoming more and more perfect. It is mainly used for close-range obstacle detection and anti-collision warning. Generally, the vehicle-mounted ultrasonic radar is used to detect obstacles in the surrounding environment in real time. The direction and distance between the obstacle and the vehicle are judged according to the echo data detected by the vehicle-mounted ultrasonic radar, and avoidance or warning operations are made accordingly. However, due to the propagation characteristics of ultrasonic radar, it lacks the perception ability to identify obstacle height information. For example, for obstacles that are lower than the height of the vehicle chassis and do not pose a collision risk, when the distance to these obstacles is close, they may still be misjudged as obstacles that need to be avoided, resulting in unnecessary alarms.

[0003] In response to the above technical problems, relevant technologies propose to calculate the height information of obstacles in the surrounding environment through three-dimensional modeling based on the same environmental data collected by lidar and cameras, and determine the obstacles that will not cause collision risks based on the height information, so as to avoid the collision detection system from misjudging these obstacles as obstacles that need to be avoided.

[0004] However, LiDAR is expensive and not all vehicles are equipped with it. Therefore, some vehicles that are not equipped with LiDAR cannot use the above technology, resulting in the lack of universality of the above technology. Summary of the invention

[0005] In order to overcome the problems existing in the related art, this specification provides a method for preparing a training data set and a method for determining the height attribute of an obstacle.

[0006] According to a first aspect of an embodiment of this specification, a method for preparing a training data set is provided, the method comprising: Constructing a training data set for training a high-level classification model, wherein the training data set includes a plurality of sample data; Among them, constructing any sample data includes: For any type of sample obstacle of known height, sample echo data obtained by ultrasonic radar detecting the sample obstacle is obtained, and sample echo parameters used to describe characteristics of the sample obstacle are parsed from the sample echo data; The any one piece of sample data is constructed based on the sample echo parameters, the characteristic data in the piece of sample data includes the sample echo parameters, and the label in the piece of sample data is used to characterize the height category of the sample obstacle relative to a preset height.

[0007] According to a second aspect of an embodiment of this specification, a method for determining an obstacle height attribute is provided. The method is applied to a vehicle equipped with a vehicle-mounted ultrasonic radar, and the method includes: Acquire echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and parse echo parameters used to describe the characteristics of the obstacle from the echo data; The echo parameters are input into a height classification model for prediction to determine a target height category of the obstacle relative to a preset height, wherein the feature data in any sample data in a training data set includes sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to the preset height, and the training data set is used to train the height classification model.

[0008] According to a third aspect of an embodiment of this specification, a device for preparing a training data set is provided, the device comprising: The training data set preparation module is used to construct a training data set for training a high-level classification model, wherein the training data set includes a plurality of sample data; wherein constructing any sample data includes: A sample echo parameter determination submodule is used to obtain, for any type of sample obstacle of known height, sample echo data obtained by ultrasonic radar detecting the sample obstacle, and parse the sample echo parameters used to describe the characteristics of the sample obstacle from the sample echo data; The sample data construction submodule is used to construct any one of the sample data based on the sample echo parameters, the characteristic data in the sample data includes the sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to a preset height.

[0009] According to a fourth aspect of an embodiment of this specification, a device for determining an obstacle height attribute is provided, the device being applied to a vehicle equipped with a vehicle-mounted ultrasonic radar, the device comprising: An echo parameter determination module is used to obtain echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and parse the echo parameters used to describe the characteristics of the obstacle from the echo data; The height category judgment module is used to input the echo parameters into the height classification model for prediction to determine the target height category of the obstacle relative to the preset height, wherein the feature data in any sample data in the training data set includes sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to the preset height, and the training data set is used to train the height classification model.

[0010] According to the fifth aspect of the embodiments of this specification, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method described in the first aspect and / or the second aspect are implemented.

[0011] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in the first aspect and / or the second aspect are implemented.

[0012] The technical solutions provided by the embodiments of this specification may have the following beneficial effects: In an embodiment of the present specification, a training data set is provided, wherein the feature data in each sample data in the training data set includes sample echo parameters parsed from sample echo data, and the label in the sample data is used to characterize the height category of the sample obstacle relative to a preset height. On the one hand, the training data set can be used to specifically train a height classification model, so that a vehicle deployed with a height classification model can determine whether the obstacle is an obstacle with a collision risk based on the echo data obtained by ultrasonic radar detecting surrounding obstacles. It can be seen that by applying the height classification model trained by this training data set to all vehicles equipped with vehicle-mounted ultrasonic radars, the judgment of the collision risk of obstacles can be completed without the support of other expensive hardware (such as laser radar), so that this solution has universality and can be applied to more ordinary vehicles. On the other hand, the feature data used in this solution is the echo parameters parsed from the echo data. Since the echo parameters are decoupled from the configuration of the ultrasonic radar, the training data set produced by this solution can be applied to vehicles with different configurations from the ultrasonic radar in the production environment, thereby improving the universality of the training data set.

[0013] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and, together with the description, serve to explain the principles of the specification.

[0015] Figure 1 It is a flowchart of a method for preparing a training data set according to an exemplary embodiment of this specification.

[0016] Figure 2 It is a flow chart of a method for determining the height attribute of an obstacle according to an exemplary embodiment of the present specification.

[0017] Figure 3It is a schematic diagram of a process of determining the height attribute of an obstacle according to an exemplary embodiment of the present specification.

[0018] Figure 4 It is a structural schematic diagram of an electronic device according to an exemplary embodiment of the present specification.

[0019] Figure 5 It is a block diagram of a device for producing a training data set according to an exemplary embodiment of the present specification.

[0020] Figure 6 It is a block diagram of a device for determining the height attribute of an obstacle according to an exemplary embodiment of the present specification. DETAILED DESCRIPTION

[0021] With the development of autonomous driving and intelligent assisted driving technology, the collision detection system of vehicles is becoming more and more perfect. It is mainly used for close-range obstacle detection and anti-collision warning. Generally, the vehicle-mounted ultrasonic radar is used to detect obstacles in the surrounding environment in real time. The direction and distance between the obstacle and the vehicle are judged according to the echo data detected by the vehicle-mounted ultrasonic radar, and avoidance or warning operations are made accordingly. However, due to the propagation characteristics of ultrasonic radar, it lacks the perception ability to identify obstacle height information. For example, for some obstacles that are lower than the height of the vehicle chassis and do not pose a collision risk, when the distance to these obstacles is close, they may still be misjudged as obstacles that need to be avoided, resulting in unnecessary alarms.

[0022] In response to the above technical problems, relevant technologies propose to calculate the height information of obstacles in the surrounding environment through three-dimensional modeling based on the same environmental data collected by lidar and cameras, and determine the obstacles that will not cause collision risks based on the height information, so as to avoid the collision detection system from misjudging these obstacles as obstacles that need to be avoided.

[0023] However, LiDAR is expensive and not all vehicles are equipped with it. Therefore, some vehicles that are not equipped with LiDAR cannot use the above technology, resulting in the lack of universality of the above technology.

[0024] In response to the above technical problems, this solution designs a training data set specifically for training a height classification model. By training the height classification model, it is enabled to have the ability to judge the height category of obstacles based on the echo data obtained by ultrasonic radar detecting obstacles. This allows vehicles deployed with the height classification model to complete the judgment of the collision risk of obstacles without the support of other expensive hardware (such as lidar).

[0025] Next, the embodiments of this specification are described in detail: like Figure 1 As shown, Figure 1 This is a flowchart of a method for preparing a training data set according to an exemplary embodiment of the present specification. Specifically, a training data set for training a high-level classification model is constructed, and the training data set includes multiple sample data, wherein constructing any sample data includes steps 101-102: Step 101: For any type of sample obstacle of known height, sample echo data obtained by detecting the sample obstacle with ultrasonic radar is acquired, and sample echo parameters for describing characteristics of the sample obstacle are parsed from the sample echo data.

[0026] Step 102: construct any one of the sample data based on the sample echo parameters, wherein the characteristic data in the sample data includes the sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to a preset height.

[0027] Different obstacle types have different shapes, materials, heights, sizes and other characteristics, and the characteristics of the echo data obtained by ultrasonic radar detecting different types of obstacles are also different. Therefore, the wider the coverage of the types of sample obstacles, the higher the feature diversity of the sample echo data in the data set, and the more it can help the height classification model to improve its generalization ability. In this regard, sample obstacles can be actual objects in a variety of different practical application scenarios, such as PVC pipes, no-stop signs, ground locks, etc. in static scenes, and shopping carts, cones, and column corners in dynamic scenes. Of course, in addition to directly using objects in actual scenes as sample obstacles, you can also artificially design some objects of different shapes, textures or sizes to act as sample obstacles to further enhance the generalization ability of the height classification model.

[0028] In order to generate sample echo data, an experimental scenario can be built, which is used to simulate the process of the vehicle-mounted ultrasonic radar detecting surrounding obstacles in the moving state and the stopped state of the vehicle to obtain echo data. Next, this manual exemplarily introduces these two experimental scenarios: In the experimental scene, there should be no other interference sources except the sample obstacles to ensure the purity and accuracy of the echo data collected by the ultrasonic radar during the experiment. For example, there should be no non-test objects or pedestrians in the experimental site.

[0029] Application scenarios for a stopped vehicle using ultrasonic radar to detect surrounding obstacles: Exemplarily, an experimental vehicle can be parked in the experimental site, and a sample obstacle can be placed within the detection range of the vehicle-mounted ultrasonic radar installed on the experimental vehicle. For example, the detection range can be between 50 cm and 200 cm from the vehicle-mounted ultrasonic radar to ensure that the vehicle-mounted ultrasonic radar can obtain more reliable sample echo data. Turn on the vehicle-mounted ultrasonic radar and continuously collect sample echo data about the sample obstacle for a period of time. Finally, the recorded experimental data includes the sample echo data of the sample obstacle and the height information of the sample obstacle. The height information can be the height category of the sample obstacle or the actual height information of the sample obstacle. Among them, the height of any sample obstacle refers to the height of the highest point of the obstacle, or the maximum height of the part where the obstacle may collide with the vehicle, or the maximum height of the part of the obstacle located in the detectable area of ​​the ultrasonic radar, etc. The height category can be classified into two categories: high and short. If the height category is high, it means that the obstacle is higher than the preset height and there is a risk of collision; if the height category is short, it means that the obstacle is not higher than the preset height, there is no risk of collision, indicating that the vehicle can ignore the obstacle. Among them, the preset height can be the height of the vehicle chassis from the ground. Since the heights of the chassis of different vehicles from the ground are different, the actual height of the same sample obstacle will be classified as a category with collision risk relative to some models, and will be classified as a category without collision risk relative to other models. For example, the chassis of a truck is much higher than that of a sports car. Although some obstacles are higher than the chassis of a sports car, they are lower than the chassis of a truck. If the height information of the sample obstacle is recorded, the height category of the sample obstacle can be flexibly adjusted according to the chassis height of the model and the actual height information of the sample obstacle when the training data set is subsequently produced, depending on the model adapted by the training data set. Of course, if the training data set is applicable to all vehicles, the preset height can be set to the lowest height acceptable to all models, and the height category of the sample obstacle can be directly determined in the recorded experimental data.

[0030] Optionally, an obstacle can be placed within the detectable range of each ultrasonic radar installed on the vehicle to conduct a full range of obstacle detection experiments. In the same detection experiment, the type and height of each obstacle can be consistent, which helps to effectively distinguish the characteristics of obstacles at different heights in subsequent data analysis and model training. By simulating the workflow of vehicle-mounted ultrasonic radars at different locations on the vehicle, the actual driving scene can be more comprehensively reflected.

[0031] Optionally, in this experimental scenario, it is not necessary to use an experimental vehicle, and an independent ultrasonic radar device can be used to complete the data acquisition task of detecting sample obstacles to obtain sample echo data. Since the independent ultrasonic radar device can be moved and adjusted more conveniently, it can be more flexibly adapted to different experimental conditions.

[0032] Application scenarios for vehicles in motion to detect surrounding obstacles using ultrasonic radar: Exemplarily, an experimental vehicle can be parked in the experimental site, and sample obstacles can be placed at a preset distance in front of, to the side of, or behind the experimental vehicle. For example, a sample obstacle is placed 3 meters in front of or behind the experimental vehicle, and the vehicle moves forward or backward at a safe distance in front of the sample obstacle at speeds of 5 km / h, 10 km / h, and 15 km / h, respectively, such as driving to a safe distance in front of the sample obstacle and stopping smoothly 50 cm in front of the sample obstacle. During the driving of the experimental vehicle, the on-board ultrasonic radar of the experimental vehicle continuously collects sample echo data of the sample obstacle. In this embodiment, sample echo data detected by the vehicle at different driving speeds and distances can be obtained, thereby enriching the feature diversity of the training data set.

[0033] Optionally, sample obstacles may be placed in front of and on both sides of the experimental vehicle in the direction of travel. For example, three obstacles may be placed in front and two obstacles may be placed on both sides to fully simulate the situation where obstacles appear in front of the vehicle and on both sides of the road at the same time.

[0034] In addition, before collecting sample echo data, the ultrasonic radar can be calibrated to ensure that the ultrasonic radar is in the best condition. After the sample echo data is collected, the integrity and validity of the sample echo data can be checked. If the data is abnormal, such as missing data, it can be collected again. When recording experimental data, the environmental conditions during the experiment can be recorded in detail, including weather conditions, ground material, light intensity, etc., to determine the degree of influence of environmental factors on the test results in subsequent data analysis.

[0035] It should be noted that the above method of collecting sample echo data is only an example and is not used to limit the method of producing the training data set of this solution. In other words, the sample echo data obtained by ultrasonic radar detecting sample obstacles in this solution can be obtained not only through the above collection method, for example, the sample echo data can also be generated by vehicle simulation software.

[0036] The training data set used to train the highly classified model consists of multiple sample data. The data composition and construction method of each sample data are the same. Next, this manual introduces the construction method of any sample data: For any type of sample obstacle of known height, sample echo data is obtained by detecting the sample obstacle with ultrasonic radar. For example, experimental data of different types of sample obstacles collected in the above embodiments can be obtained, and each piece of experimental data includes sample echo data obtained by detecting the sample obstacle with ultrasonic radar and the height information of the sample obstacle.

[0037] After acquiring the sample echo data, sample echo parameters for describing the characteristics of the sample obstacles can be parsed from the sample echo data. The sample obstacle characteristics can be understood as information such as the size, shape, material, and height of the sample obstacles.

[0038] In an illustrated embodiment, the sample echo data is composed of at least one sample echo unit, and the sample echo parameters include the confidence of the sample echo unit, and also include the echo distance, echo height and / or echo width of the sample echo unit. For example, each time the ultrasonic radar transmits an ultrasonic signal, if the shape of the sample obstacle is complex (such as having multiple reflecting surfaces or surfaces with different angles), the edge of the sample obstacle or the uneven surface will produce independent reflections, so that the ultrasonic radar will receive multiple echo signals in time sequence, and each echo signal can be regarded as an echo unit. The sample echo data received each time is composed of at least one sample echo unit. Each sample echo unit is represented by the rising edge timestamp of the received echo signal, the falling edge timestamp of the echo signal, and the signal strength peak of the echo signal. Among them, the echo distance can be the distance between the sample obstacle and the ultrasonic radar calculated based on the time difference from the ultrasonic radar transmitting the ultrasonic signal to the ultrasonic radar receiving the ultrasonic signal; the echo height of the echo unit can be represented by the signal intensity peak of the echo signal of the echo unit; the echo width of the echo unit can be represented by the difference between the rising edge timestamp and the falling edge timestamp of the echo signal of the echo unit. The confidence of the sample echo unit can be the reliability score of the sample echo unit determined based on the signal strength consistency and noise level. Preferably, when calculating the echo distance, the ultrasonic signal emitted by other ultrasonic radars can be used, and the echo signal captured by the ultrasonic radar after being reflected by the sample obstacle can be used. This fusion of multipath signals enhances the feature richness of the echo parameters in the training data set.

[0039] In this embodiment, the echo distance in the sample echo parameters can be used to help the height classification model identify the spatial distribution of the sample obstacles, the echo height can be used to help the height classification model determine the material of the sample obstacles, the echo width can be used to help the height classification model determine the size and shape of the sample obstacles, and the confidence can be used to help the height classification model determine the reliability of the sample echo parameters. By combining multi-dimensional parameters such as confidence, echo height, echo distance, and echo width, the characteristics of the obstacles can be analyzed more comprehensively, so that the height attributes of the obstacles can be more accurately determined based on these characteristics.

[0040] In an illustrated embodiment, the sample echo data is composed of at least one sample echo unit. After the sample echo data is acquired, the first k sample echo units received first can be determined from the sample echo data in the order in which the sample echo units are received, where k is a positive integer. Then, the sample echo parameters of each of the first k sample echo units are determined respectively.

[0041] Specifically, assuming that the sample echo data is composed of 10 sample echo units, k is 3, and after determining the first three first sample echo units, the second sample echo units, and the third sample echo units received first from the sample echo data, the first sample echo parameters of the first sample echo unit, the second sample echo parameters of the second sample echo unit, and the third sample echo parameters of the third sample echo unit are determined respectively. The first sample echo parameters, the second sample echo parameters, and the third sample echo parameters together constitute the sample echo parameters corresponding to the sample echo data. For example, the sample echo parameters of each sample echo unit include the echo distance, echo height, echo width, and confidence of the sample echo unit. Then the sample echo parameter values ​​of each frame of sample echo data can be composed of three echo distance values, three echo height values, three echo width values, and three confidence values ​​of the first three sample echo units. There is a special case where, if the number of sample echo units in the sample echo data is less than k, all sample echo units in the sample echo data may be selected, and sample echo parameters of the missing sample echo units are filled with zero values.

[0042] In this embodiment, since the echo signals of sample echo units with earlier receiving times have less noise and interference and higher confidence, sample echo units with earlier receiving times in the sample echo data are selected, and sample echo parameters are extracted from these sample echo units as feature data, which can effectively avoid noise interference.

[0043] In an illustrated embodiment, when determining the first k sample echo units received from the sample echo data, invalid sample echo units whose echo distance is less than the distance threshold may be first removed from the sample echo data to obtain suspicious sample echo data, and then the first k sample echo units received from the suspicious sample echo data may be determined. For example, the echo distance of each sample echo unit in the sample echo data may be calculated and compared with the distance threshold respectively. If the echo distance is greater than the distance threshold, the sample echo unit may be retained. If the echo distance is not greater than the distance threshold, the sample echo unit may be removed from the sample echo data, and finally suspicious sample echo data may be obtained, and then the first k sample echo units received may be determined from the suspicious sample echo data. The distance threshold may be selected as 15 cm.

[0044] In this embodiment, by deleting sample echo data whose echo distance is less than the distance threshold, sample echo data of near-field interference can be excluded.

[0045] In another illustrated embodiment, when determining the first k sample echo units received first from the sample echo data, a first sample set of valid sample echo units whose confidence is greater than a confidence threshold can be determined from the sample echo data, and the first k first valid sample echo units received first can be determined from the first sample set. For example, the confidence of each sample echo unit in the sample echo data can be calculated first and compared with the confidence threshold respectively. If the confidence is greater than the confidence threshold, the sample echo unit can be placed in the first sample set. If the confidence is not greater than the confidence threshold, the sample echo unit can be not placed in the first sample set. Finally, the first sample set is obtained, and then the first k first valid sample echo units received first can be determined from the first sample set.

[0046] In this embodiment, by filtering the sample echo units whose confidence is lower than the confidence threshold according to the confidence, the sample echo units containing a large amount of noise interference can be excluded, thereby improving the quality of the training data.

[0047] In another illustrated embodiment, when determining the first k sample echo units received first from the sample echo data, invalid sample echo units with echo distances less than a distance threshold may be eliminated from the sample echo data to obtain suspicious sample echo data, a second sample set of valid sample echo units with confidence greater than the confidence threshold is determined from the suspicious sample echo data, and the first k second valid sample echo units received first are determined from the second sample set. Of course, filtering may also be performed based on confidence first and then based on echo distance.

[0048] In an illustrated embodiment, when acquiring sample echo data obtained by ultrasonic radar detecting sample obstacles, the original sample echo data obtained by ultrasonic radar detecting sample obstacles can be acquired, and then the original sample echo data is filtered to obtain sample echo data with noise removed. The filtering process can specifically be co-frequency interference processing.

[0049] like Figure 2 As shown, Figure 2 201 is a flowchart of a method for determining the height attribute of an obstacle according to an exemplary embodiment of the present specification. The method can be applied to a vehicle equipped with an on-board ultrasonic radar, and includes steps 201-202: Step 201: Acquire echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and parse the echo data to obtain echo parameters for describing the characteristics of the obstacles.

[0050] Step 202: Input the echo parameters into a height classification model for prediction to determine the target height category of the obstacle relative to a preset height.

[0051] In the prior art, vehicles mainly use the echo data detected by the on-board ultrasonic radar to determine the direction and distance between the obstacle and the vehicle, and make avoidance or warning operations accordingly. However, due to the propagation characteristics of ultrasonic radar, it lacks the perception ability to identify the height information of obstacles, that is, it is unable to parse the true height information of obstacles from the echo data.

[0052] This solution designs a training data set, in which the feature data in any sample data in the training data set includes sample echo data, and the label in the sample data is used to characterize the height category of the sample obstacle relative to the preset height. Figure 2 The training data set used by the highly classified model can be produced by the production method shown in the aforementioned embodiment, and of course, can also be produced by other production methods. This specification does not impose any limitation on the production method.

[0053] The height classification model is trained by the training data set, so that it can learn the correlation between the sample echo parameters in the sample echo data and the height category of the sample obstacle, and has the ability to predict the height category of the real obstacle according to the echo parameters in the new echo data, so as to make up for the defect that the real height information of the obstacle cannot be directly parsed from the echo data. Among them, the model architecture of the height classification model can be a classification model such as a support vector machine (SVM) and a random forest in the prior art. Of course, it can also be a model architecture customized by a technician in this field, and this specification does not impose any restrictions on this.

[0054] The height category may include two categories: high and low. The classification of the height category of the obstacle is related to the preset height. For example, the preset height may be the height of the vehicle chassis from the ground. Since the vehicle chassis heights of different vehicles are different, the actual height of the obstacle will be classified as a category with collision risk relative to some models, and will be classified as a category without collision risk relative to other models. For example, the chassis of a truck is much higher than that of a sports car. Although some obstacles are higher than the chassis of a sports car, they are lower than the chassis of a truck.

[0055] Therefore, the training data set can be suitable for training a height classification model for a certain vehicle model, and the height category label in each sample data can be obtained by comparing the actual height of the sample obstacle with the chassis height of the vehicle model. For example, if the actual height of the sample obstacle is not lower than the chassis height of the vehicle model, the height classification label of the sample obstacle can be marked as "high"; if the actual height of the sample obstacle is lower than the chassis height of the vehicle model, the height category label of the sample obstacle can be marked as "short". Of course, the training data set can be a height classification model applicable to all vehicle models. The preset height can select the lowest height that all vehicle models can accept.

[0056] In practical applications, the highly classified model obtained by training the training data set designed in this scheme is deployed in vehicles equipped with ultrasonic radars. When the vehicle is actually driving or stopped, the on-board ultrasonic radar detects the surrounding environment in real time, such as Figure 3 As shown, the vehicle obtains the echo data obtained by the on-board ultrasonic radar detecting the surrounding obstacles, and parses the echo parameters used to describe the characteristics of the obstacle from the echo data. Among them, the parameter type of the echo parameter and the sample echo parameter is the same, which will not be repeated here. Then, the echo parameter is input into the height classification model for prediction, and the height classification model will output the target height category of the obstacle relative to the preset height. When the target height category is high and the current distance between the obstacle and the vehicle is lower than the preset safety distance, the vehicle can output warning information for the obstacle. For example, the spatial position information between the obstacle and the vehicle is displayed on the central control screen of the vehicle, and the warning information is displayed to prompt the user to avoid the vehicle from scratching the obstacle. And / or, if the vehicle is in a driving state at the current moment, for example, the vehicle is in an automatic driving or intelligent assisted driving state at this moment, the braking system can be notified to control the vehicle to slow down or stop. When the target height category is short, it indicates that the height of the obstacle will not scratch the vehicle chassis, and the vehicle can ignore the obstacle.

[0057] By deploying the height classification model trained with the training data set designed according to this solution on the vehicle, the vehicle can determine whether the object is an obstacle with a collision risk based on the echo data obtained by ultrasonic radar detecting surrounding obstacles. Therefore, this solution can help more ordinary vehicles support the judgment of the collision risk of obstacles without obtaining the real height information of the obstacles.

[0058] In addition, the original ultrasonic signal characteristics in the original echo data obtained by the ultrasonic radar detecting the same obstacle will be affected by the configuration of the ultrasonic radar, such as the calibration parameters and installation location of the ultrasonic radar. The original ultrasonic signal characteristics obtained by ultrasonic radars with different configurations detecting the same obstacle are different, but the echo parameters parsed from the original echo data are the same. Therefore, this scheme uses the echo parameters parsed from the original echo data as the feature variables in the sample data, so that the data set produced by this scheme can be applied to vehicles with different configurations from the ultrasonic radar in the production environment, without the need to recalibrate the vehicle's ultrasonic radar to be consistent with the configuration of the ultrasonic radar in the production environment, thereby improving the universality of the training data set.

[0059] Optionally, the echo data is composed of at least one sample echo unit, and echo parameters used to describe obstacle characteristics are parsed from the echo data, including: determining the first k echo units received from the echo data in the order of receiving the echo units, where k is a positive integer; and determining the echo parameters of each of the first k echo units respectively.

[0060] Optionally, determining the first k echo units received first from the echo data includes: eliminating invalid echo units whose echo distance is less than a distance threshold from the echo data to obtain suspicious echo data, and determining the first k echo units received first from the suspicious echo data; and / or determining a first set of valid echo units whose confidence level is greater than a confidence threshold from the echo data, and determining the first k first valid echo units received first from the first set.

[0061] Optionally, the echo data is composed of at least one echo unit, and the echo parameters include the confidence of the echo unit and also include the echo distance, echo height and / or echo width of the echo unit.

[0062] Optionally, obtaining echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles includes: obtaining original echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and filtering the original echo data to obtain echo data with noise removed.

[0063] Corresponding to the embodiments of the aforementioned method, this specification also provides embodiments of a device and a terminal to which it is applied.

[0064] like Figure 4 As shown, Figure 4It is a structural diagram of an electronic device shown in this specification according to an exemplary embodiment. At the hardware level, the electronic device 400 includes a processor 402, an internal bus 404, a network interface 406, a memory 408 and a non-volatile memory 410, and of course may also include hardware required for other services. One or more embodiments of this specification can be implemented based on software, such as the processor 402 reading the corresponding computer program from the non-volatile memory 410 into the memory 408 and then running it. Of course, in addition to the software implementation, one or more embodiments of this specification do not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic module, but can also be hardware or logic devices.

[0065] like Figure 5 As shown, Figure 5 is a block diagram of a training data set preparation device according to an exemplary embodiment of the present specification. The device can be applied to Figure 4 The electronic device 400 shown in the figure is used to implement the technical solution of this specification. The device includes: The training data set preparation module 51 is used to construct a training data set for training a high-level classification model, wherein the training data set includes a plurality of sample data; wherein constructing any sample data includes: The sample echo parameter determination submodule 511 is used to obtain, for any type of sample obstacle of known height, sample echo data obtained by ultrasonic radar detecting the sample obstacle, and parse the sample echo parameters used to describe the characteristics of the sample obstacle from the sample echo data; The sample data construction submodule 512 is used to construct any one of the sample data based on the sample echo parameters, the characteristic data in the sample data includes the sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to a preset height.

[0066] Optionally, the sample echo data is composed of at least one sample echo unit, and the sample echo parameter determination submodule 511 is specifically used to determine the first k sample echo units received first from the sample echo data according to the order of receiving the sample echo units, where k is a positive integer; and to determine the sample echo parameters of each of the first k sample echo units respectively.

[0067] Optionally, the sample echo parameter determination submodule 511 is specifically used to eliminate invalid sample echo units whose echo distance is less than a distance threshold from the sample echo data to obtain suspicious sample echo data, and determine the first k sample echo units received first from the suspicious sample echo data; and / or, determine a first sample set of valid sample echo units whose confidence level is greater than a confidence threshold from the sample echo data, and determine the first k first valid sample echo units received first from the first sample set.

[0068] Optionally, the sample echo data is composed of at least one sample echo unit, and the sample echo parameters include the confidence of the sample echo unit and also include the echo distance, echo height and / or echo width of the sample echo unit.

[0069] Optionally, the sample echo parameter determination submodule 511 is specifically used to obtain original sample echo data obtained by detecting the sample obstacle with the ultrasonic radar; and filter the original sample echo data to obtain sample echo data with noise removed.

[0070] Optionally, the ultrasonic radar is a vehicle-mounted ultrasonic radar.

[0071] like Figure 6 As shown, Figure 6 is a block diagram of a device for determining the height attribute of an obstacle according to an exemplary embodiment of the present specification. The device can be applied to Figure 4 In the electronic device 400 shown, the device can be applied to a vehicle equipped with a vehicle-mounted ultrasonic radar to implement the technical solution of this specification. The device includes: The echo parameter determination module 602 is used to obtain echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and parse the echo parameters used to describe the characteristics of the obstacle from the echo data; The height category judgment module 604 is used to input the echo parameters into the height classification model for prediction to determine the target height category of the obstacle relative to the preset height, wherein the feature data in any sample data in the training data set includes sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to the preset height, and the training data set is used to train the height classification model.

[0072] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.

[0073] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this specification. A person of ordinary skill in the art can understand and implement it without paying creative labor.

[0074] The present specification also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the aforementioned methods for preparing a training data set or the method for determining the height attribute of an obstacle provided in the present application.

[0075] Specifically, computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including, for example, semiconductor memory devices (such as EPROM, EEPROM and flash memory devices), magnetic disks (such as internal hard disks or removable disks), magneto-optical disks, and CD ROM and DVD-ROM disks.

Claims

1. A method for preparing a training data set, characterized in that: The method comprises: Constructing a training data set for training a high-level classification model, wherein the training data set includes a plurality of sample data; Among them, constructing any sample data includes: For any type of sample obstacle of known height, sample echo data obtained by ultrasonic radar detecting the sample obstacle is obtained, and sample echo parameters used to describe characteristics of the sample obstacle are parsed from the sample echo data; The any one piece of sample data is constructed based on the sample echo parameters, the characteristic data in the piece of sample data includes the sample echo parameters, and the label in the piece of sample data is used to characterize the height category of the sample obstacle relative to a preset height.

2. The method according to claim 1, characterized in that The sample echo data is composed of at least one sample echo unit, and the parsing of the sample echo parameters for describing the characteristics of the sample obstacle from the sample echo data includes: According to the order of receiving the sample echo units, the first k sample echo units received first are determined from the sample echo data, where k is a positive integer; The sample echo parameters of each sample echo unit in the first k sample echo units are determined respectively.

3. The method according to claim 2, characterized in that The step of determining the first k sample echo units received first from the sample echo data comprises: Eliminating invalid sample echo units whose echo distance is less than a distance threshold from the sample echo data to obtain suspicious sample echo data, and determining first k sample echo units received first from the suspicious sample echo data; and / or, A first sample set of valid sample echo units having a confidence level greater than a confidence threshold is determined from the sample echo data, and first k first valid sample echo units received first are determined from the first sample set.

4. The method according to claim 1, characterized in that: The sample echo data is composed of at least one sample echo unit, and the sample echo parameters include the confidence of the sample echo unit, and also include the echo distance, echo height and / or echo width of the sample echo unit.

5. The method according to claim 1, characterized in that The acquiring of sample echo data obtained by detecting the sample obstacle by the ultrasonic radar includes: Acquire original sample echo data obtained by ultrasonic radar detecting the sample obstacle; The original sample echo data is filtered to obtain sample echo data from which noise is removed.

6. The method according to any one of claims 1 to 5, characterized in that The ultrasonic radar is a vehicle-mounted ultrasonic radar.

7. A method for determining the height attribute of an obstacle, characterized in that: The method is applied to a vehicle equipped with a vehicle-mounted ultrasonic radar, and the method comprises: Acquire echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and parse echo parameters used to describe the characteristics of the obstacle from the echo data; The echo parameters are input into a height classification model for prediction to determine a target height category of the obstacle relative to a preset height, wherein the feature data in any sample data in a training data set includes sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to the preset height, and the training data set is used to train the height classification model.

8. The method according to claim 7, characterized in that The height categories include tall and short, and the method further includes: When the target height category is high and the distance between the vehicle and the obstacle is lower than a preset safety distance, outputting warning information for the obstacle; and / or, If the vehicle is in a driving state at the current moment, the braking system is notified to control the vehicle to decelerate or stop.

9. A device for preparing a training data set, characterized in that: The device comprises: The training data set preparation module is used to construct a training data set for training a high-level classification model, wherein the training data set includes a plurality of sample data; wherein constructing any sample data includes: A sample echo parameter determination submodule is used to obtain, for any type of sample obstacle of known height, sample echo data obtained by ultrasonic radar detecting the sample obstacle, and parse the sample echo parameters used to describe the characteristics of the sample obstacle from the sample echo data; The sample data construction submodule is used to construct any one of the sample data based on the sample echo parameters, the characteristic data in the sample data includes the sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to a preset height.

10. A device for determining the height attribute of an obstacle, characterized in that: The device is applied to a vehicle equipped with a vehicle-mounted ultrasonic radar, and the device comprises: An echo parameter determination module is used to obtain echo data obtained by the vehicle-mounted ultrasonic radar detecting surrounding obstacles, and parse the echo parameters used to describe the characteristics of the obstacle from the echo data; The height category judgment module is used to input the echo parameters into the height classification model for prediction to determine the target height category of the obstacle relative to the preset height, wherein the feature data in any sample data in the training data set includes sample echo parameters, and the label in the sample data is used to characterize the height category of the sample obstacle relative to the preset height, and the training data set is used to train the height classification model.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images