A method, device, equipment and storage medium for obstacle determination

Abstract

The embodiments of the present application disclose a method, device, equipment and storage medium for obstacle determination, belonging to the technical field of vehicle driving, and can be applied to scenarios such as closed parks like ports and logistics, or urban traffic, or highways. The method is as follows: The detection results of the lidar and other target sensors for the driving space are fused by means of deep learning to achieve the confirmation of obstacles. During the implementation process, for the laser point cloud obtained by the lidar, not only the determination of whether it is noise is carried out at the point cloud level, but also the morphological characteristics of the driving space are determined at the global level, and the discrimination of whether the point position conforms to the temporal characteristics of the obstacle is carried out. The perception characteristics of whether there are obstacles in the driving space are determined through the target sensors. By discriminating the dynamic characteristics of the driving space and at least one perception characteristic, it is determined whether there are obstacles in the driving space, improving the accuracy of determining whether there are obstacles in the driving space.

Description

Technical Field

[0001] This application relates to the technical field of vehicle driving, and provides a method, device, equipment and storage medium for obstacle determination. Background Art

[0002] At present, lidar has been widely used in the field of autonomous driving, and many driverless systems are equipped with lidar. During the driving process of a vehicle, the lidar outputs feedback information such as the positions and intensities of points of 3D space objects. The accurate feedback information of obstacle positions in 3D space is crucial for the driverless system. Therefore, the driverless system relies heavily on lidar.

[0003] However, negligible obstacles such as rain, snow, and fog that may be encountered during driving will form noise points in the lidar return point cloud (i.e., obstacles that can be ignored during driving). The vehicle may misjudge such noise points as an obstacle during driving and control the driving system to perform an emergency brake, etc., thereby bringing an uncomfortable experience to passengers.

[0004] At present, there are generally two methods for processing the above-mentioned noise points during driving: one is to filter the laser noise points, that is, according to information such as the data wavelength and intensity returned by the lidar, perform noise filtering inside the lidar, and further perform point-by-point classification filtering on the point cloud; the other is to analyze the spatial form of the point cloud and perform classification filtering on the point cloud according to local point cloud information. Obviously, the above process of processing noise points only analyzes the local information of the lidar point cloud. Therefore, it is easy to misjudge noise points as real obstacles, thereby affecting the normal driving of the vehicle. Summary of the Invention

[0005] The embodiments of this application provide a method, device, equipment and storage medium for obstacle determination to improve the accuracy of determining obstacles in the driving space.

[0006] The specific technical solutions provided by this application are as follows:

[0007] In a first aspect, the embodiments of this application provide a method for obstacle determination, including:

[0008] Based on a point cloud feature model, determine the determination features of the points in the lidar point cloud as noise points, where the lidar point cloud is obtained by a lidar mounted on a vehicle detecting the driving space;

[0009] Based on the points in the lidar point cloud that are outside the ground, determine the morphological features of the driving space;

[0010] If both the determination feature and the morphological feature of the driving space conform to the preset obstacle time series feature, the points corresponding to the determination feature and the points corresponding to the morphological feature are integrated to obtain the dynamic feature of the driving space;

[0011] Based on at least one target sensor, determine the perception feature of whether there is an obstacle in the driving space, where the target sensor is a sensor installed on the vehicle;

[0012] By discriminating the dynamic feature of the driving space and at least one perception feature, determine whether there is an obstacle in the driving space.

[0013] Optionally, based on the point cloud feature model, determine the determination feature that the point in the lidar point cloud is a noise point, including:

[0014] For each point in the lidar point cloud, use the point cloud feature model to judge the probability that the point is a noise point. If the probability that the point is a noise point is less than the historical probability, it is determined that the point is not a noise point; if the probability that the point is a noise point is not less than the historical probability, it is determined that the point is a noise point, and the result of determining the point as a noise point is used as the determination feature;

[0015] Among them, the point cloud feature model is trained according to historical sample data, the historical sample data carries the historical probability that each point in the driving space is determined as a noise point, and the historical probability is determined based on the point feature of each point and the type information of the lidar.

[0016] Optionally, based on the points in the lidar point cloud that are outside the ground, determine the morphological feature of the driving space, including:

[0017] Based on the position information of each point in the lidar point cloud, extract the points outside the ground from the lidar point cloud;

[0018] Cluster the extracted points outside the ground to obtain unknown obstacles, and use the clustered unknown obstacles as the morphological feature of the driving space.

[0019] Optionally, if both the determination feature and the morphological feature of the driving space conform to the preset obstacle time series feature, the points corresponding to the determination feature and the points corresponding to the morphological feature are integrated to obtain the dynamic feature of the driving space, including:

[0020] Determine the first time series corresponding to the determination feature, where the first time series includes the determination moment when the point is determined as a noise point;

[0021] Determine the second time series corresponding to the morphological feature of the driving space, and determine the tracking feature corresponding to the morphological feature of the driving space based on the point identifier of the point determined as a noise point, where the first time series and the second time series partially overlap, and the duration of the second time series is greater than the duration of the first time series;

[0022] If, within the continuous duration corresponding to the first time sequence, it is determined that the determination moment corresponding to the determination feature conforms to the preset obstacle time sequence feature, and within the continuous duration corresponding to the second time sequence, the morphological feature of the driving space conforms to the preset obstacle time sequence feature, then the point corresponding to the determination feature and the point corresponding to the morphological feature are integrated based on their positions in the lidar point cloud to obtain the dynamic feature of the driving space, where the continuous duration is determined based on the determination moment corresponding to the point and the preset time fluctuation threshold, and the preset obstacle time sequence feature corresponding to each moment includes a tracking feature.

[0023] Optionally, the perception feature for determining whether there is an obstacle in the driving space based on at least one target sensor includes:

[0024] Obtain the category information of the target sensor installed on the vehicle;

[0025] For each type of target sensor with category information, perform the following: Obtain the result data after the target sensor perceives the driving space. If the result data is greater than or equal to the confidence level of the target sensor, it is determined that there is an obstacle in the driving space; if the result data is less than the confidence level of the target sensor, it is determined that there is no obstacle in the driving space, where the confidence level is determined based on the prior data after the target sensor perceives the driving space and the category information of the target sensor.

[0026] Optionally, before determining whether there is an obstacle in the driving space by discriminating the dynamic feature of the driving space and at least one perception feature, it further includes:

[0027] Input the dynamic feature and perception feature of the original driving space in the sample data into the original deep learning model to determine whether there is an obstacle in the original driving space. If the training result of whether there is an obstacle in the original driving space determined this time is different from the expected result, adjust the parameters of the original deep learning model, and continue to input the dynamic feature and perception feature of the driving space in the sample data into the original deep learning model with adjusted parameters until the training result of whether there is an obstacle in the original driving space obtained is the same as the expected result. Take the original deep learning model with the same training result and expected result as the deep learning model;

[0028] Among them, the dynamic feature of the driving space carries the type information of the lidar, the perception feature carries the category information of the corresponding target sensor, and the original deep learning model is trained based on the dynamic feature obtained in advance by the lidar with type information and the perception feature obtained in advance by at least one type of target sensor with category information.

[0029] In a second aspect, an embodiment of the present application further provides an obstacle determination device, including:

[0030] A noise point determination unit, configured to determine a determination feature for a point position in a lidar point cloud to be a noise point based on a point cloud feature model, where the lidar point cloud is obtained by a lidar mounted on a vehicle detecting a driving space;

[0031] A shape determination unit, configured to determine a shape feature of a driving space based on point positions outside the ground in the lidar point cloud;

[0032] A timing discrimination unit, configured to integrate the point positions corresponding to the determination feature and the point positions corresponding to the shape feature of the driving space to obtain a dynamic feature of the driving space if both the determination feature and the shape feature of the driving space conform to a preset obstacle timing feature;

[0033] A sensor determination unit, configured to determine a perception feature of whether there is an obstacle in a driving space based on at least one target sensor, where the target sensor is a sensor mounted on a vehicle;

[0034] An obstacle determination unit, configured to determine whether there is an obstacle in the driving space by discriminating the dynamic feature of the driving space and at least one perception feature.

[0035] Optionally, based on a point cloud feature model, to determine a determination feature for a point position in a lidar point cloud to be a noise point, the noise point determination unit is configured to:

[0036] For each point position in the lidar point cloud, use the point cloud feature model to judge the probability that the point position is a noise point. If the probability that the point position is a noise point is less than the historical probability, it is determined that the point position is not a noise point; if the probability that the point position is a noise point is not less than the historical probability, it is determined that the point position is a noise point, and the result of determining the point position as a noise point is used as the determination feature;

[0037] Wherein, the point cloud feature model is trained according to historical sample data, the historical sample data carries the historical probability of each point position in the driving space being determined as a noise point, and the historical probability is determined based on the point position feature of each point position and the type information of the lidar.

[0038] Optionally, based on point positions outside the ground in the lidar point cloud, to determine a shape feature of the driving space, the shape determination unit is configured to:

[0039] Extract point positions outside the ground from the lidar point cloud based on the position information of each point position in the lidar point cloud;

[0040] Cluster the extracted point positions outside the ground to obtain unknown obstacles, and use the clustered unknown obstacles as the shape feature of the driving space.

[0041] Optionally, if both the determination feature and the morphological feature of the driving space conform to the preset obstacle time series feature, the points corresponding to the determination feature and the points corresponding to the morphological feature are integrated to obtain the dynamic feature of the driving space. The time series discrimination unit is used for:

[0042] Determine the first time series corresponding to the determination feature, where the first time series includes the determination moment when the point is determined to be a noise point;

[0043] Determine the second time series corresponding to the morphological feature of the driving space, and determine the tracking feature corresponding to the morphological feature of the driving space based on the point identifier of the point determined to be a noise point. The first time series and the second time series partially overlap, and the duration of the second time series is greater than the duration of the first time series;

[0044] If within the continuous duration corresponding to the first time series, the determination moment corresponding to the determination feature conforms to the preset obstacle time series feature, and within the continuous duration corresponding to the second time series, the morphological feature of the driving space conforms to the preset obstacle time series feature, then the points corresponding to the determination feature and the points corresponding to the morphological feature are integrated based on their positions in the laser point cloud to obtain the dynamic feature of the driving space, where the continuous duration is determined based on the determination moment corresponding to the point and the preset time fluctuation threshold, and the preset obstacle time series feature corresponding to each moment includes the tracking feature.

[0045] Optionally, based on at least one target sensor, determine the perception feature of whether there is an obstacle in the driving space. The sensor determination unit is used for:

[0046] Obtain the category information of the target sensor installed on the vehicle;

[0047] For each type of target sensor with category information, execute: obtain the result data after the target sensor perceives the driving space. If the result data is greater than or equal to the confidence level of the target sensor, determine that there is an obstacle in the driving space; if the result data is less than the confidence level of the target sensor, determine that there is no obstacle in the driving space, where the confidence level is determined based on the prior data after the target sensor perceives the driving space and the category information of the target sensor.

[0048] Optionally, before determining whether there is an obstacle in the driving space by discriminating the dynamic feature of the driving space and at least one perception feature, it further includes:

[0049] Input the dynamic features and perception features of the original driving space in the sample data into the original deep learning model to determine whether there are obstacles in the original driving space. If the training result of whether there are obstacles in the original driving space determined this time is different from the expected result, adjust the parameters of the original deep learning model, and continue to input the dynamic features and perception features of the driving space in the sample data into the original deep learning model with adjusted parameters until the training result of whether there are obstacles in the obtained original driving space is the same as the expected result. Take the original deep learning model with the training result the same as the expected result as the deep learning model;

[0050] Among them, the dynamic features of the driving space carry the type information of the lidar, the perception features carry the category information of the corresponding target sensor, and the original deep learning model is trained based on the dynamic features obtained in advance by the lidar based on the type information and the perception features obtained in advance by the target sensors of at least one category information.

[0051] In a third aspect, an electronic device includes:

[0052] A memory for storing executable instructions;

[0053] A processor for reading and executing the executable instructions stored in the memory to implement the method according to any one of the first aspects.

[0054] In a fourth aspect, a computer-readable storage medium, when the instructions in the storage medium are executed by a processor, enables the processor to execute the method according to any one of the first aspects.

[0055] In summary, in the embodiments of the present application, a method, device, equipment and storage medium for obstacle determination are provided. The method includes: fusing the detection results of the lidar and at least one other target sensor for the driving space by using deep learning to confirm whether there are obstacles in the driving space. In the specific implementation process, for the laser point cloud obtained by the lidar, not only the determination of whether it is a noise point is performed at the point cloud level, but also the morphological features of the driving space are determined at the global level. The processing results of the above laser point cloud will be used to discriminate whether the points meet the timing characteristics of the obstacles. At the same time, the perception features of whether there are obstacles in the driving space are determined by the target sensor, and by discriminating the dynamic features of the driving space and at least one perception feature, it is finally determined whether there are obstacles in the driving space. The above method of fusing the detection results of the lidar and the target sensor for the driving space improves the accuracy of determining whether there are obstacles in the driving space. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of an application scenario provided by an embodiment of the present application;

[0057] Figure 2 It is a schematic diagram of the overall process of an obstacle determination method provided by an embodiment of the present application;

[0058] Figure 3 It is a schematic diagram of the process of determining the morphological characteristics of the driving space based on points outside the ground provided by an embodiment of the present application;

[0059] Figure 4 It is a schematic diagram of the process of determining the dynamic characteristics of the driving space provided by an embodiment of the present application;

[0060] Figure 5 It is a schematic diagram of the process of determining whether there are obstacles in the driving space based on the target sensor provided by an embodiment of the present application;

[0061] Figure 6 It is a schematic diagram of the logical architecture of an obstacle determination device provided by an embodiment of the present application;

[0062] Figure 7 It is a schematic diagram of the physical architecture of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0063] To make the objectives, technical solutions and advantages of the present application clearer and more understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without making creative efforts fall within the protection scope of the present application. Without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other arbitrarily. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0064] The terms "first" and "second" in the embodiments of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the term "including" and any of its variations are intended to cover non-exclusive protection. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices. "Multiple" in the present application may mean at least two, for example, it may be two, three or more, and the embodiments of the present application do not make limitations.

[0065] In the technical solutions of the present application, the collection, dissemination, use, etc. of data all comply with the requirements of relevant national laws and regulations.

[0066] First, some concepts involved in the embodiments of the present application will be introduced below.

[0067] 1. Lidar: It emits detection signals to a target, compares the received signals reflected from the target with the transmitted signals, and thus obtains information about the target, such as parameters like the target distance, azimuth, altitude, speed, attitude, and even shape, so as to detect, track, and identify the target.

[0068] 2. Camera: It is a video input device with basic functions such as video shooting and static image capture. After the lens captures an image, the photosensitive component and control component inside the camera process the image and further convert it into a digital signal that can be recognized by a computer. Then, it is input into the computer via a parallel port or a Universal Serial BUS (USB) connection, etc., and the image is restored by software.

[0069] 3. Millimeter-wave radar: It is a radar that operates in the millimeter-wave band, featuring small size, light weight, and high spatial resolution. The millimeter-wave seeker has strong ability to penetrate fog, smoke, and dust, and has the characteristics of all-weather (except heavy rain) and all-time. In addition, the millimeter-wave radar can identify very small targets and can also identify multiple targets simultaneously, with strong imaging ability, small size, good mobility, and concealment.

[0070] 4. Sonar sensor: A sensor developed using the characteristics of ultrasonic waves. During the working process, it emits acoustic wave signals outward. When encountering an object, the above acoustic wave signals will reflect back corresponding signals, and the distance and position of the target object are calculated based on the reflection time and waveform.

[0071] 5. Random forest: It refers to a classifier that uses multiple trees to train and predict samples. In machine learning, a random forest is a classifier containing multiple decision trees, and the output class is determined by the mode of the classes output by individual trees.

[0072] 6. XGBoost: It is an optimized distributed gradient boosting library designed to be efficient, flexible, and portable. Before training, XGBoost sorts the data according to features, then saves it in a block structure, and adopts a sparse matrix storage format for storage in each block structure. The block structure is repeatedly used during the subsequent training process, thus greatly reducing the computational amount.

[0073] 7. Random Sample Consensus Algorithm: Estimates the parameters of a mathematical model iteratively from a set of observed data containing outliers. In this algorithm, it is assumed that the data contains correct data and abnormal data. The correct data is denoted as inliers, and the abnormal data is denoted as outliers. At the same time, this algorithm also assumes that given a set of correct data, there is a method to calculate the model parameters that fit these data.

[0074] As mentioned above, during the driving process of a vehicle, rain, snow, fog, etc. will form noise points in the laser point cloud obtained by the lidar. When the above noise points are determined to be obstacles, operations such as emergency braking of the vehicle will be triggered. Other ways to process the noise points include noise filtering inside the lidar and classification filtering based on local point cloud information. However, the above ways to process the noise points are likely to determine the negligible obstacles of the noise points as real obstacles, thus affecting the normal driving of the vehicle.

[0075] To solve the above problems, the embodiment of the present application provides a method for obstacle determination, which combines the lidar and other sensors (i.e., at least one target sensor) to comprehensively determine whether there are obstacles in the driving space. Specifically, the laser point cloud obtained by the lidar is analyzed from two aspects: the point level and the morphological level to determine whether there are obstacles, so as to obtain the dynamic characteristics of the driving space. Further, the above dynamic characteristics and the perception characteristics of the target sensor are discriminated to finally determine whether there are obstacles in the driving space.

[0076] Refer to Figure 1 shown, which is a schematic diagram of an application scenario corresponding to an embodiment of the present application.

[0077] As Figure 1 shown, this application scenario may include a vehicle driving on a road, and the vehicle is equipped with a lidar and at least one target sensor. Among them, the above target sensor can be one or several of a camera, a millimeter-wave radar, a lidar, and a sonar sensor. During the driving process of the vehicle, the driving space is used to represent the area corresponding to the field of view of the vehicle during the driving process (i.e., the space sensed by one or more of the above sensors in any direction for the road). During the driving process of the vehicle, obstacles may appear at any time. The obstacles here include but are not limited to other vehicles, objects on the road, and objects on both sides of the road that may affect the driving of the vehicle.

[0078] In Figure 1In the illustrated application scenario, the vehicle may be affected by special weather conditions (e.g., dust, rain, fog, etc.). When using lidar for detection, the above special weather conditions may correspondingly create noise points in the lidar point cloud. Generally, the above noise points are negligible obstacles for the vehicle's driving. However, if only relying on the results shown by the lidar point cloud obtained by the lidar, it is very likely to determine the above noise points as real obstacles (i.e., obstacles) affecting the vehicle's driving, thereby triggering operations such as emergency braking of the vehicle. Therefore, in the embodiments of the present application, the detection results of the lidar and at least one target sensor installed on the vehicle are combined to make a comprehensive determination of whether it is an obstacle. The following is a specific introduction.

[0079] Referring to Figure 2 As shown, in the embodiments of the present application, the specific process of obstacle determination is as follows:

[0080] Since a lidar is installed on the vehicle, during the driving process of the vehicle, the lidar will detect the driving space and obtain the returned detection result, namely the lidar point cloud. In the embodiments of the present application, the lidar point cloud obtained by the lidar is processed at the point level and the global level respectively.

[0081] Step 201: Based on the point cloud feature model, determine the determination features of the points in the lidar point cloud as noise points, where the lidar point cloud is obtained by the lidar installed on the vehicle detecting the driving space.

[0082] It should be noted that before processing the lidar point cloud at the point level, a point cloud feature model needs to be set inside the lidar. Among them, the point cloud feature model is trained according to historical sample data. The historical sample data carries the historical probability of determining noise points at each point in the driving space. The historical probability is determined based on the point features at each point and the type information of the lidar.

[0083] That is, during the training process of the above point cloud feature model, historical sample data is input. The above historical sample data is the lidar point cloud obtained by the lidar, and the above historical sample data carries the historical probability of determining noise points at each point in the driving space. For example, the historical probability of determining point A as a noise point is X1, the historical probability of determining point B as a noise point is X2, and so on.

[0084] In the process of processing the lidar-derived point cloud at the point level, in order to obtain more accurate determination features during the use of the point cloud feature model, first, the point features of each point in the lidar point cloud during the training process are consistent with those during the use process. For example, the distribution shape of the points, the density of the points, etc.; second, the type of lidar used during the training process is consistent with the type of lidar during the use process. Usually, the numerical value of the above historical probability is related to the point features of each point and the type information of the lidar.

[0085] Specifically, the point cloud feature model can be any one of the above-mentioned random forest or XGBoost.

[0086] During the implementation process, for each point in the lidar point cloud, the point cloud feature model is used to judge the probability that the point is a noise point. If the probability that the point is a noise point is less than the historical probability, it is determined that the point is not a noise point. If the probability that the point is a noise point is not less than the historical probability, it is determined that the point is a noise point, and the result of determining the point as a noise point is used as the determination feature.

[0087] For example, when the above point cloud feature model is a random forest, the random forest judges the probability that each point in the lidar point cloud is a noise point in the form of a decision tree. For example, for point A, the probability that the point is a noise point calculated by the random forest is a%. Further compare whether the probability a% is less than the historical probability b%. If a% is less than b%, it is determined that point A is not a noise point; if a% is not less than b%, it is determined that point A is a noise point, and the result of determining the point as a noise point (i.e., point A, a%) is used as the determination feature.

[0088] Another example is when the above point cloud feature model is XGBoost, it will judge the probability that each point in the lidar point cloud is a noise point in the form of a distributed gradient boosting library. For example, for point C, the probability that the point is a noise point calculated by the random forest is c%. Further compare whether the probability c% is less than the historical probability d%. If c% is less than d%, it is determined that point C is not a noise point; if c% is not less than d%, it is determined that point C is a noise point, and the result of determining the point as a noise point (i.e., point C, c%) is used as the determination feature.

[0089] Step 202: Based on the points in the lidar point cloud that are outside the ground, determine the morphological features of the driving space.

[0090] The following introduces the process of processing the lidar-derived point cloud at the global level. During the driving process of the vehicle, if all the points in the lidar point cloud fall on the ground, that is, the display height of all points is zero, it means that there are no obstacles in the driving space, and only the road surface for the vehicle to drive exists in the current driving space.

[0091] In the embodiments of the present application, based on the points outside the ground in the laser point cloud, the morphological characteristics of the driving space are determined. Refer to Figure 3 as shown, which specifically includes:

[0092] Step 2021: Based on the position information of each point in the laser point cloud, extract the points outside the ground from the laser point cloud.

[0093] During the implementation process, determine the position information of each point in the laser point cloud. For example, method (1) can be used to first obtain the actual position information of a certain point in the laser point cloud, and then obtain the relative position information between other points and the above-mentioned certain point, and determine the position information of each point according to the above actual position information and relative position information; or method (2) to obtain the true position information of a certain reference plane, for example, the ground, and then determine the relative position information of each point in the laser point cloud with respect to the reference plane. The position information of each point in the laser point cloud is determined by the above method (1), method (2) or other methods.

[0094] After determining the position information of each point, compare each position information with the position represented by the ground. If the position information of a certain point exceeds the position represented by the ground, extract the point outside the ground from the laser point cloud. Alternatively, the above-mentioned normal vector method or random sample consensus algorithm can be used to extract the points on the ground in the laser point cloud. In this way, the remaining points in the laser point cloud are the points outside the ground.

[0095] Step 2022: Cluster the points outside the ground extracted to obtain unknown obstacles, and use the clustered unknown obstacles as the morphological characteristics of the driving space.

[0096] Since raindrops about to fall on the car window in the driving space, road signs on the ground in the driving space, etc. will all be extracted as points outside the ground in the above processing. Therefore, after extracting all the points outside the ground, continue to cluster the points, that is, obtain the obstacle categories represented by all the above points, obtain unknown obstacles from the process of point clustering, and further use the clustered unknown obstacles as the morphological characteristics of the driving space.

[0097] It should be added that the above morphological characteristics of the driving space only exist for the current acquisition moment of the laser point cloud, and at the next acquisition moment of the laser point cloud, the points outside the ground need to be re-extracted and clustered using the above method.

[0098] Step 203: If both the determination feature and the morphological characteristics of the driving space meet the preset obstacle time sequence characteristics, integrate the points corresponding to the determination feature and the points corresponding to the morphological characteristics to obtain the dynamic characteristics of the driving space.

[0099] Considering that the noise points formed by factors such as raindrops and dust may only last for a short period of time, that is, such noise points are negligible obstacles to vehicle driving, and the point positions obtained in the above steps do not have temporality for both the determination features of noise points and the morphological features determined based on the point positions outside the ground. Therefore, in the implementation process, a tracking filter can be introduced, and then the above determination features and morphological features can be further determined according to the temporal features of the tracking filter, that is, the tracking filter is used to perform temporal discrimination on the determination features and the morphological features of the driving space. For the specific steps, refer to Figure 4 as shown.

[0100] Step 2031: Determine the first time sequence corresponding to the determination feature, where the first time sequence includes the determination moment when the point position is determined as a noise point.

[0101] In order to measure whether the noise points at the point level have temporal persistence, correspondingly, in the implementation process, a tracking filter such as is used to determine the first time sequence corresponding to the determination feature. Usually, the above first time sequence is a continuous period of time, and, in this first time sequence, it includes the determination moment when the point position is determined as a noise point, that is, the above first time sequence is a period of time including the noise point determination moment.

[0102] For example, when the determination moment when the point position E is determined as a noise point is 8:20, the tracking filter Z can set the above first time sequence to 8:00 - 8:40 according to the pre-configuration.

[0103] It should be noted that common tracking filters are one of the Kalman filter, the Extended Kalman filter, or the Unscented Kalman filter.

[0104] Step 2032: Determine the second time sequence corresponding to the morphological feature of the driving space, and, based on the point position identifier of the point position determined as a noise point, determine the tracking feature corresponding to the morphological feature of the driving space, where the first time sequence and the second time sequence partially overlap, and the duration of the second time sequence is greater than the duration of the first time sequence.

[0105] Correspondingly, in order to measure whether the morphological feature of the driving space at the global level has temporal persistence, in the implementation process, the same tracking filter mentioned above, or a tracking filter of the same model as the tracking filter mentioned above can be used to determine the second time sequence corresponding to the morphological feature of the driving space. Similarly, the above second time sequence is also a continuous period of time, that is, the duration of the clustering of the point positions outside the ground into unknown obstacles is determined by the tracking filter. For example, the second time sequence of the clustering of the point positions A, N, and M outside the ground into unknown obstacles is 8:00 - 9:00.

[0106] In order to correspond the laser points at the point level and the global level, the tracking filter extracts the point identifiers of the points determined to be noise points, and further determines the tracking features corresponding to the morphological features of the driving space based on the point identifiers. Usually, the tracking features include the contour of the morphological features and their corresponding appearance times, etc.

[0107] Similarly, in order to correspond the laser points at the point level and the global level, the above first time sequence partially overlaps with the second time sequence, and considering that the process of processing the points at the global level is relatively cumbersome, the duration of the second time sequence here is greater than the duration of the above first time sequence.

[0108] Step 2033: If within the continuous duration corresponding to the first time sequence, the determination moment corresponding to the determination feature conforms to the preset obstacle time sequence feature, and within the continuous duration corresponding to the second time sequence, the morphological feature of the driving space conforms to the preset obstacle time sequence feature, then the points corresponding to the determination feature and the points corresponding to the morphological feature are integrated based on their positions in the laser point cloud to obtain the dynamic feature of the driving space, where the continuous duration is determined based on the determination moment corresponding to the point and the preset time fluctuation threshold, and the preset obstacle time sequence feature corresponding to each moment includes the tracking feature.

[0109] It should be noted that during the determination process of the dynamic feature, preset obstacle time sequence features will be set. For example, features such as the appearance duration of the obstacle exceeding 1 minute and the morphology of the obstacle remaining unchanged within 30 seconds. During the implementation process, the determination of the determination feature and the morphological feature is respectively performed within the continuous durations corresponding to the first time sequence and the second time sequence. The specific values of the continuous durations corresponding to the above first time sequence and the second time sequence are both determined based on the determination moment corresponding to the point and the preset time fluctuation threshold, that is, the specific start moments of the first time sequence and the second time sequence are jointly determined by the determination moment and the preset time fluctuation threshold, and the value of the above time fluctuation threshold can be flexibly set according to the usage scenario.

[0110] During the implementation process, only when the determination moment corresponding to the above determination feature conforms to the preset obstacle time sequence feature within the continuous duration corresponding to the first time sequence, and the morphological feature of the above driving space also conforms to the preset obstacle time sequence feature within the continuous duration corresponding to the second time sequence, will the dynamic feature of the driving space be obtained, that is, the point corresponding to the determination feature and the point corresponding to the morphological feature are integrated based on their positions in the laser point cloud. Specifically, if there is a point corresponding to the determination feature and a point corresponding to the morphological feature at the same position in the laser point cloud, any one of the points can be removed; if the points corresponding to the determination feature and the morphological feature partially overlap at the same position in the laser point cloud, the union of the points corresponding to the determination feature and the morphological feature is taken; if the points corresponding to the determination feature and the morphological feature are within the preset integration range but do not overlap, the points corresponding to the determination feature and the morphological feature are retained; if the points corresponding to the determination feature and the morphological feature are not within the preset integration range, the points corresponding to the determination feature and the morphological feature are deleted.

[0111] For example, the preset obstacle time sequence feature is that the appearance duration of the obstacle exceeds 20 seconds and the morphology of the obstacle does not change within 10 seconds. Then, within the continuous duration corresponding to the first time sequence (assumed to be 60 seconds), it is detected whether the continuous appearance duration of the determination feature exceeds 20 seconds. If the determination moment corresponding to the determination feature is from 8:20:10 to 8:20:40, that is, there is a situation exceeding 20 seconds, the points corresponding to the determination feature at 8:20:10 and the points corresponding to the determination feature at 8:20:40 are respectively obtained. If the morphological feature of the driving space conforms to the rule of not changing within 10 seconds, the points corresponding to the morphological feature are obtained. Assume that the duration corresponding to the points corresponding to the morphological feature is from 8:20:10 to 8:20:21. Assume that within the time period from 8:20:10 to 8:20:21, the points corresponding to the determination feature and the points corresponding to the morphological feature overlap. Then, integration is performed by removing the points corresponding to the above determination feature, and the points corresponding to the morphological feature are used as the dynamic feature of the driving space.

[0112] Step 204: Determine the perception feature of whether there is an obstacle in the driving space based on at least one target sensor, where the target sensor is a sensor installed on the vehicle.

[0113] It should be noted that at least one target sensor is also installed on the vehicle. Specifically, the target sensor can be one or several of the above camera, millimeter wave radar, and sonar sensor.

[0114] Refer to Figure 5 As shown, the steps of determining whether there is an obstacle in the driving space through the target sensor specifically include:

[0115] Step 2041: Obtain the category information of the target sensor mounted on the vehicle.

[0116] Considering that the category of the target sensor actually mounted on the vehicle is uncertain, in order to accurately obtain the result of whether there are obstacles in the driving space, during the implementation process, it is necessary to first obtain the category information of the target sensor mounted on the vehicle, that is, to determine which one of the above-mentioned cameras, millimeter-wave radars, and sonar sensors the target sensor is.

[0117] Step 2042: For the target sensor of each category information, perform the following operations: Obtain the result data after the target sensor senses the driving space. If the result data is greater than or equal to the confidence level of the target sensor, it is determined that there are obstacles in the driving space. If the result data is less than the confidence level of the target sensor, it is determined that there are no obstacles in the driving space, where the confidence level is determined based on the prior data after the target sensor senses the driving space and the category information of the target sensor.

[0118] After determining the category information of the target sensor mounted on the vehicle, the confidence level of the target sensor with this category information is obtained in advance. This confidence level is used to represent the probability that the target sensor determines that there are obstacles in the driving space. Usually, the confidence levels corresponding to target sensors with different category information are also different. The confidence level is obtained by using target sensors with different category information to sense the driving space in advance to obtain prior data, and obtaining the probability value determined as an obstacle that matches the category information based on the prior data.

[0119] For the target sensor of each category information, the following operations are performed: During the driving process of the vehicle, the driving space is sensed by the target sensor to obtain result data. The result data specifically includes: obtaining image data of the driving space through a camera, detecting target data of the driving space through a millimeter-wave radar, and obtaining target objects in the driving space through acoustic wave signals emitted by a sonar sensor, etc.

[0120] After obtaining the result data, the result data is compared with the confidence level of the target sensor of the same category information. If the result data is greater than or equal to the confidence level of the target sensor, that is, the result data after the target sensor senses the driving space is greater than the probability of determining that there are obstacles, it is determined that there are obstacles in the driving space. If the result data is less than the confidence level of the target sensor, that is, the result data after the target sensor senses the driving space is less than the probability of determining that there are obstacles, it is determined that there are no obstacles in the driving space.

[0121] Step 205: Determine whether there are obstacles in the driving space by discriminating the dynamic characteristics and at least one sensing characteristic of the driving space.

[0122] During the implementation process, after obtaining the dynamic features of the driving space through a lidar and at least one perception feature through a target sensor, the above-mentioned dynamic features and perception features are discriminated to determine whether there are obstacles in the driving space. Preferably, the above-mentioned dynamic features and perception features are input into a deep learning model for discrimination.

[0123] During the implementation process, before determining whether there are obstacles in the driving space by discriminating the dynamic features of the driving space and at least one perception feature, it further includes: inputting the dynamic features and perception features of the original driving space in the sample data into the original deep learning model to determine whether there are obstacles in the original driving space. If the training result of whether there are obstacles in the original driving space determined this time is different from the expected result, the parameters of the original deep learning model are adjusted, and the dynamic features and perception features of the driving space in the sample data are continuously input into the original deep learning model with adjusted parameters until the training result of whether there are obstacles in the original driving space obtained is the same as the expected result. The original deep learning model with the training result the same as the expected result is used as the deep learning model.

[0124] Obviously, in the embodiment of the present application, the training process of the original deep learning model takes the dynamic features and perception features of the original driving space in the sample data as inputs. On this basis, it is compared whether the training result of the obstacle is the same as the expected result. If they are the same, the training continues; if they are different, the parameters of the model are adjusted and then the training continues until the latest obtained training result is the same as the expected result and the training stops, and the original deep learning model is used as the deep learning model.

[0125] It should be noted that although the lidar is used to detect the driving space to obtain the laser point cloud, and the target sensor is used to obtain the original perception data after perceiving the driving space, the above-mentioned original data such as the laser point cloud and the result data will not be directly input into the deep learning model for training and learning. Instead, the lidar preliminarily processes the laser point cloud to obtain the intermediate result of the dynamic features, and the target sensor preliminarily processes the above-mentioned original perception data to obtain the intermediate result of the perception feature of whether there are obstacles, and then the above-mentioned intermediate results, that is, the dynamic features and perception features, are input into the deep learning model for discrimination.

[0126] In addition, the dynamic features of the driving space carry the type information of the lidar, the perception features carry the category information of the corresponding target sensor, and the original deep learning model is trained based on the dynamic features obtained in advance by the lidar with type information and the perception features obtained in advance by at least one target sensor with category information.

[0127] That is, to ensure the accuracy of the determination result, during the training process and the usage process of the deep learning model, the type information of the lidar corresponding to them is consistent, and the category information of the target sensor corresponding to them is also consistent. In this way, during the implementation process, the dynamic features of the driving space and at least one perception feature obtained in real time are input into the deep learning model for discrimination. If the discrimination result is the same as the expected result, it is determined that there is an obstacle in the driving space; if the discrimination result is different from the expected result, it is determined that there is no obstacle in the driving space.

[0128] Those skilled in the art can understand that in the above method of the specific implementation manner, the writing order of each step does not mean a strict execution order and does not constitute any limitation to the implementation process. The specific execution order of each step should be determined according to its function and possible internal logic.

[0129] Refer to Figure 6 As shown, an obstacle determination device is provided in an embodiment of the present application, including:

[0130] A noise point determination unit 601, configured to determine the determination feature that a point position in the laser point cloud is a noise point based on the point cloud feature model, where the laser point cloud is obtained by a lidar mounted on a vehicle detecting the driving space;

[0131] A shape determination unit 602, configured to determine the shape feature of the driving space based on the point positions outside the ground in the laser point cloud;

[0132] A timing discrimination unit 603, configured to, if both the determination feature and the shape feature of the driving space conform to the preset obstacle timing feature, integrate the point position corresponding to the determination feature and the point position corresponding to the shape feature to obtain the dynamic feature of the driving space;

[0133] A sensor determination unit 604, configured to determine the perception feature of whether there is an obstacle in the driving space based on at least one target sensor, where the target sensor is a sensor mounted on the vehicle;

[0134] An obstacle determination unit 605, configured to determine whether there is an obstacle in the driving space by discriminating the dynamic feature of the driving space and at least one perception feature.

[0135] Optionally, based on the point cloud feature model, to determine the determination feature that a point position in the laser point cloud is a noise point, the noise point determination unit 601 is configured to:

[0136] For each point position in the laser point cloud, use the point cloud feature model to judge the probability that the point position is a noise point. If the probability that the point position is a noise point is less than the historical probability, it is determined that the point position is not a noise point; if the probability that the point position is a noise point is not less than the historical probability, it is determined that the point position is a noise point, and the result of determining the point position as a noise point is used as the determination feature;

[0137] Among them, the point cloud feature model is trained according to historical sample data. The historical sample data carries the historical probability that each point in the driving space is determined to be a noise point, and the historical probability is determined based on the point features of each point and the type information of the lidar.

[0138] Optionally, based on the points outside the ground in the lidar point cloud, determine the morphological features of the driving space. The morphology determination unit 602 is used for:

[0139] Extract the points outside the ground from the lidar point cloud based on the position information of each point in the lidar point cloud;

[0140] Cluster the extracted points outside the ground to obtain unknown obstacles, and use the clustered unknown obstacles as the morphological features of the driving space.

[0141] Optionally, if both the determination feature and the morphological feature of the driving space conform to the preset obstacle time series feature, then integrate the points corresponding to the determination feature and the points corresponding to the morphological feature to obtain the dynamic feature of the driving space. The time series discrimination unit 603 is used for:

[0142] Determine the first time series corresponding to the determination feature, where the first time series includes the determination moment when the point is determined to be a noise point;

[0143] Determine the second time series corresponding to the morphological feature of the driving space, and determine the tracking feature corresponding to the morphological feature of the driving space based on the point identifier of the point determined to be a noise point. Among them, the first time series and the second time series partially overlap, and the duration of the second time series is greater than the duration of the first time series;

[0144] If within the continuous duration corresponding to the first time series, the determination moment corresponding to the determination feature conforms to the preset obstacle time series feature, and within the continuous duration corresponding to the second time series, the morphological feature of the driving space conforms to the preset obstacle time series feature, then integrate the points corresponding to the determination feature and the points corresponding to the morphological feature based on their positions in the lidar point cloud to obtain the dynamic feature of the driving space, where the continuous duration is determined based on the determination moment corresponding to the point and the preset time fluctuation threshold, and the preset obstacle time series feature corresponding to each moment includes the tracking feature.

[0145] Optionally, determine the perception feature of whether there is an obstacle in the driving space based on at least one target sensor. The sensor determination unit 604 is used for:

[0146] Obtain the category information of the target sensors installed on the vehicle;

[0147] Execute for the target sensor corresponding to each type of category information: Obtain the result data after the target sensor senses the driving space. If the result data is greater than or equal to the confidence level of the target sensor, it is determined that there is an obstacle in the driving space; if the result data is less than the confidence level of the target sensor, it is determined that there is no obstacle in the driving space. Here, the confidence level is determined based on the prior data after the target sensor senses the driving space and the category information of the target sensor.

[0148] Optionally, before determining whether there is an obstacle in the driving space by discriminating the dynamic features and at least one sensing feature of the driving space, it further includes:

[0149] Input the dynamic features and sensing features of the original driving space in the sample data into the original deep learning model to determine whether there is an obstacle in the original driving space. If the training result of whether there is an obstacle in the original driving space determined this time is different from the expected result, adjust the parameters of the original deep learning model, and continue to input the dynamic features and sensing features of the driving space in the sample data into the original deep learning model with adjusted parameters until the training result of whether there is an obstacle in the original driving space obtained is the same as the expected result. Take the original deep learning model with the same training result and expected result as the deep learning model;

[0150] Among them, the dynamic features of the driving space carry the type information of the lidar, the sensing features carry the category information of the corresponding target sensor, and the original deep learning model is trained based on the dynamic features pre-obtained by the lidar based on the type information and the sensing features pre-obtained by at least one category information of the target sensor.

[0151] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0152] Refer to Figure 7 As shown, the embodiments of the present application provide an electronic device, including: a memory 701 for storing executable instructions; a processor 702 for reading and executing the executable instructions stored in the memory and executing any one of the methods in the above first aspect.

[0153] Based on the same inventive concept, the embodiments of the present application provide a computer-readable storage medium. When the instructions in the storage medium are executed by the processor, the processor can execute any one of the methods described in the above first aspect.

[0154] In summary, in the embodiments of the present application, a method, device, equipment, and storage medium for obstacle determination are provided. The method includes: fusing the detection results of a lidar and at least one other target sensor for the driving space by using deep learning to confirm whether there are obstacles in the driving space. In the specific implementation process, for the lidar point cloud obtained, not only the determination of whether it is noise is made at the point cloud level, but also the morphological characteristics of the driving space are determined at the global level. The processing results of the above lidar point cloud will be used to determine whether the points meet the temporal characteristics of obstacles. At the same time, the perception characteristics of whether there are obstacles in the driving space are determined through the target sensor, and by discriminating the dynamic characteristics of the driving space and at least one perception characteristic, it is finally determined whether there are obstacles in the driving space. The above method of fusing the detection results of the lidar and the target sensor for the driving space by using a deep learning model improves the accuracy of determining whether there are obstacles in the driving space.

[0155] In various embodiments of the present application, if there is no special explanation and logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.

[0156] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, system, or computer program product system. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product system 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 application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program product systems according to the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes 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 processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes 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 apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means embodying the function specified in the flowchart(s) Figure 1 a flowchart or flowcharts and / or block(s) Figure 1 a block or blocks.

[0159] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the function specified in the flowchart(s) Figure 1 a flowchart or flowcharts and / or block(s) Figure 1 a block or blocks.

[0160] It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to cover these modifications and variations.

Claims

1. A method for obstacle determination, characterized in that, The method includes: Based on a point cloud feature model, determining a determination feature for a point position in a lidar point cloud to be a noise point, where the lidar point cloud is obtained by a lidar mounted on a vehicle detecting a driving space, and the driving space is used to represent the area corresponding to the field of view of the vehicle during driving; Based on the point positions outside the ground in the lidar point cloud, determining a morphological feature of the driving space; Determining a first time sequence corresponding to the determination feature, where the first time sequence includes a determination moment when the point position is determined to be a noise point, determining a second time sequence corresponding to the morphological feature of the driving space, and determining a tracking feature corresponding to the morphological feature of the driving space based on the point position identifier of the point position determined to be a noise point, where the first time sequence and the second time sequence partially overlap, and the duration of the second time sequence is greater than the duration of the first time sequence. If, within the continuous duration corresponding to the first time sequence, the determination moment corresponding to the determination feature conforms to a preset obstacle time sequence feature, and within the continuous duration corresponding to the second time sequence, the morphological feature of the driving space conforms to a preset obstacle time sequence feature, then integrating the point position corresponding to the determination feature and the point position corresponding to the morphological feature based on their positions in the lidar point cloud to obtain the dynamic feature of the driving space, where the continuous duration is determined based on the determination moment corresponding to the point position and a preset time fluctuation threshold, and the preset obstacle time sequence feature corresponding to each moment includes the tracking feature; Based on at least one target sensor, determining a perception feature of whether there is an obstacle in the driving space, where the target sensor is a sensor mounted on the vehicle; By discriminating the dynamic feature of the driving space and at least one of the perception features, determining whether there is an obstacle in the driving space.

2. The method according to claim 1, characterized in that, The determining, based on the point cloud feature model, the determination feature for a point position in the lidar point cloud to be a noise point includes: For each point position in the lidar point cloud, using the point cloud feature model to judge the probability of the point position being a noise point. If the probability of the point position being a noise point is less than the historical probability, then determining that the point position is not a noise point; if the probability of the point position being a noise point is not less than the historical probability, then determining that the point position is a noise point, and using the result of determining the point position to be a noise point as the determination feature; Wherein, the point cloud feature model is trained according to historical sample data, the historical sample data carries the historical probability of each point position in the driving space being determined to be a noise point, and the historical probability is determined based on the point position feature of each point position and the type information of the lidar.

3. The method according to claim 1, characterized in that, The determining, based on the point positions outside the ground in the lidar point cloud, the morphological feature of the driving space includes: Based on the position information of each of the point positions in the lidar point cloud, extracting the point positions outside the ground from the lidar point cloud; Clustering the extracted point positions outside the ground to obtain unknown obstacles, and using the clustered unknown obstacles as the morphological feature of the driving space.

4. The method according to claim 1, characterized in that, The determining, based on at least one target sensor, the perception feature of whether there is an obstacle in the driving space includes: Obtain the category information of the target sensor mounted on the vehicle; For each target sensor with the category information, perform the following: Obtain the result data after the target sensor senses the driving space. If the result data is greater than or equal to the confidence level of the target sensor, it is determined that there is an obstacle in the driving space; if the result data is less than the confidence level of the target sensor, it is determined that there is no obstacle in the driving space, where the confidence level is determined based on the prior data after the target sensor senses the driving space and the category information of the target sensor.

5. The method according to claim 1, characterized in that, Before determining whether there is an obstacle in the driving space by discriminating the dynamic characteristics of the driving space and at least one of the sensing characteristics, it further includes: Input the dynamic characteristics and sensing characteristics of the original driving space in the sample data into the original deep learning model to determine whether there is an obstacle in the original driving space. If the training result of whether there is an obstacle in the original driving space determined this time is different from the expected result, adjust the parameters of the original deep learning model, and continue to input the dynamic characteristics and sensing characteristics of the driving space in the sample data into the original deep learning model with adjusted parameters until the training result of whether there is an obstacle in the original driving space obtained is the same as the expected result. Take the original deep learning model with the training result the same as the expected result as the deep learning model; Wherein, the dynamic characteristics of the driving space carry the type information of the lidar, the sensing characteristics carry the category information of the corresponding target sensor, and the original deep learning model is trained based on the dynamic characteristics obtained in advance by the lidar based on the type information and the sensing characteristics obtained in advance by the target sensor of at least one category information.

6. An apparatus for obstacle determination, characterized in that, It includes: A noise determination unit for determining the determination characteristics of the points in the laser point cloud as noise points based on the point cloud feature model, where the laser point cloud is obtained by the lidar mounted on the vehicle detecting the driving space, and the driving space is used to represent the area corresponding to the vehicle's field of view during driving; A morphology determination unit for determining the morphological characteristics of the driving space based on the points in the laser point cloud located outside the ground; A timing discrimination unit, configured to determine a first timing corresponding to the determination feature, where the first timing includes a determination moment when the point is determined to be a noise point, determine a second timing corresponding to the morphological feature of the driving space, and determine a tracking feature corresponding to the morphological feature of the driving space based on the point identifier of the point determined to be a noise point. The first timing and the second timing partially overlap, and the duration of the second timing is greater than the duration of the first timing. If, within the continuous duration corresponding to the first timing, the determination moment corresponding to the determination feature conforms to a preset obstacle timing feature, and within the continuous duration corresponding to the second timing, the morphological feature of the driving space conforms to a preset obstacle timing feature, then the point corresponding to the determination feature and the point corresponding to the morphological feature are integrated based on their positions in the lidar point cloud to obtain the dynamic feature of the driving space. The continuous duration is determined based on the determination moment corresponding to the point and a preset time fluctuation threshold, and the preset obstacle timing feature corresponding to each moment includes the tracking feature; A sensor determination unit, configured to determine a perception feature of whether there is an obstacle in the driving space based on at least one target sensor, where the target sensor is a sensor mounted on the vehicle; An obstacle determination unit, configured to determine whether there is an obstacle in the driving space by discriminating the dynamic feature of the driving space and at least one of the perception features; 7. The apparatus according to claim 6, characterized in that, Based on the point cloud feature model, a determination feature for determining that a point in the lidar point cloud is a noise point, and the noise point determination unit is configured to: For each point in the lidar point cloud, use the point cloud feature model to judge the probability that the point is a noise point. If the probability that the point is a noise point is less than the historical probability, it is determined that the point is not a noise point; if the probability that the point is a noise point is not less than the historical probability, it is determined that the point is a noise point, and the result of determining the point as a noise point is used as the determination feature; Wherein, the point cloud feature model is trained according to historical sample data, the historical sample data carries the historical probability that each point in the driving space is determined to be a noise point, and the historical probability is determined based on the point features of each point and the type information of the lidar; 8. An electronic device, characterized in that, Comprising: A memory, configured to store executable instructions; A processor, configured to read and execute the executable instructions stored in the memory to implement the method according to any one of claims 1-5; 9. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor, the processor is enabled to execute the method according to any one of claims 1-5.

Images