Scene screening method, device, equipment and storage medium based on lane matching

Abstract

The present invention relates to the field of autonomous driving technology, and discloses a scene screening method, device, equipment and storage medium based on lane matching. The method collects road test data, analyzes the road test data, determines the position point with the smallest distance of the main vehicle on the semantic map based on the obtained basic vehicle information and semantic map data, and judges whether the main vehicle is at an intersection in the current frame based on the position point. If so, the current frame is marked as a target frame, and the straight lane matching pairs entering and exiting the intersection in the target frame are determined according to the semantic map information and the position information of the main vehicle at the current intersection, and it is judged whether the number of matching pairs is greater than a preset threshold. If so, it is determined that the target frame corresponds to a crossroads scene. The automatic screening algorithm for crossroads scenes based on the analysis of road test data and lane matching solves the technical problems of low crossroads scene screening efficiency and recognition accuracy in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a scene screening method, device, equipment and storage medium based on lane matching. Background Art

[0002] As a basic scenario, the intersection scenario plays an important role in both the analysis of drive test data and algorithm iteration testing. When establishing a test scenario set containing intersections or counting the number of times this scenario appears in a drive test, you first need to have the ability to filter this scenario.

[0003] Currently, there are two main methods: manual annotation and algorithm automatic screening. The former has low screening efficiency due to the huge amount of data; while the mainstream method of the latter only considers simple screening logic - intersection (Hub) filtering, but as a general type of map annotation area, the scene set covered by the intersection is too broad, including on and off ramps, merging / exiting, U-turns, etc., resulting in a high false alarm rate. Therefore, how to improve the screening and recognition efficiency of intersection scenes has become a technical problem that technicians in this field need to solve. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problems existing in the prior art of low efficiency in screening intersection scenes and low accuracy in identifying intersection scenes.

[0005] A first aspect of the present invention provides a scene screening method based on lane matching, comprising: collecting road test data of a main vehicle traveling in a driving scene, wherein the road test data includes traffic scene data collected when the main vehicle is tested on a real road, and the traffic scene data includes multiple frames of vehicle driving images; parsing the road test data to obtain basic vehicle information and semantic map data of the main vehicle; based on the basic vehicle information and the semantic map data, determining a target position point with the smallest distance of the main vehicle on a preset semantic map, and judging whether the main vehicle is at an intersection in a current frame based on the target position point; if so, determining that the main vehicle is located at an intersection in the current frame, and marking the current frame as a target frame; determining straight lane matching pairs entering and exiting the intersection in the target frame according to the semantic map information and the position information of the main vehicle at the current intersection; judging whether the number of straight lane matching pairs is greater than a preset threshold, and if so, determining that the target frame corresponds to an intersection scene.

[0006] Optionally, in a first implementation method of the first aspect of the present invention, the collection of road test data of the main vehicle traveling in a driving scenario includes: metadata in the road test data transmitted based on a preset wireless network; determining the full data corresponding to the metadata; and fusing the metadata and the full data to obtain the road test data of the main vehicle traveling in the driving scenario.

[0007] Optionally, in a second implementation method of the first aspect of the present invention, the target position point of the main vehicle with the shortest distance on a preset semantic map is determined based on the basic vehicle information and the semantic map data, and whether the main vehicle is at an intersection in the current frame is judged based on the target position point, including: determining the position coordinate data and a first orientation angle of the main vehicle based on the basic vehicle information and the semantic map data; determining the orientation of the main vehicle and the target road point closest to the main vehicle based on the position coordinate data and the first orientation angle; taking the target road point as the center of the circle and the preset distance as the radius, obtaining a semantic map within the radius; and judging whether the main vehicle is at an intersection in the current frame based on the semantic map.

[0008] Optionally, in a third implementation manner of the first aspect of the present invention, determining a matching pair of straight lanes entering and exiting the intersection within the target frame based on the semantic map information and the position information of the main vehicle at the current intersection includes: determining all lanes entering and exiting the intersection based on the semantic map information and the position information of the main vehicle at the current intersection; acquiring road points on all the lanes, and judging whether all the lanes are straight lanes based on the road points.

[0009] Optionally, in a fourth implementation method of the first aspect of the present invention, the acquiring of road points on all lanes and determining whether all lanes are straight lanes based on the road points include: calculating the standard deviation of the second orientation angle corresponding to the road points; determining whether the standard deviation is less than or equal to a preset threshold; if so, the lane is a straight lane.

[0010] Optionally, in a fifth implementation manner of the first aspect of the present invention, determining a matching pair of straight lanes entering and exiting the intersection within the target frame based on the semantic map information and the position information of the main vehicle at the current intersection also includes: acquiring all boundaries of the current intersection; traversing all boundaries, and summarizing the orientation angles of lanes corresponding to all boundaries to obtain a third orientation angle of all boundaries; based on the third orientation angle, matching all lanes entering and exiting the intersection to obtain a matching result; and determining a matching pair of straight lanes entering and exiting the intersection within the target frame based on the matching result.

[0011] The second aspect of the present invention provides a scene screening device based on lane matching, comprising: a collection module, used to collect road test data of a main vehicle traveling in a driving scene, wherein the road test data includes traffic scene data collected when the main vehicle is tested on a real road, and the traffic scene data includes multiple frames of vehicle driving images; a parsing module, used to parse the road test data to obtain basic vehicle information and semantic map data of the main vehicle; a judgment module, used to determine the target position point with the smallest distance on a preset semantic map of the main vehicle based on the basic vehicle information and the semantic map data, and judge whether the main vehicle is at an intersection in the current frame based on the target position point; a marking module, used to determine that the main vehicle is located at an intersection in the current frame and mark the current frame as a target frame if so; a first determination module, used to determine the straight lane matching pairs entering and exiting the intersection in the target frame according to the semantic map information and the position information of the main vehicle at the current intersection; a second determination module, used to judge whether the number of the straight lane matching pairs is greater than a preset threshold, and if so, determine that the target frame corresponds to a crossroads scene.

[0012] Optionally, in a first implementation method of the second aspect of the present invention, the acquisition module is specifically used to: obtain metadata in road test data transmitted based on a preset wireless network; determine the full data corresponding to the metadata; and fuse the metadata and the full data to obtain road test data of the main vehicle traveling in a driving scenario.

[0013] Optionally, in a second implementation method of the second aspect of the present invention, the judgment module is specifically used to: determine the position coordinate data and a first orientation angle of the main vehicle based on the vehicle basic information and the semantic map data; determine the orientation of the main vehicle and the target road point closest to the main vehicle based on the position coordinate data and the first orientation angle; obtain a semantic map within the radius with the target road point as the center and a preset distance as the radius; and determine whether the main vehicle is at an intersection in the current frame based on the semantic map.

[0014] Optionally, in a third implementation manner of the second aspect of the present invention, the first determination module includes: a determination unit, used to determine all lanes entering and exiting the intersection based on the semantic map information and the position information of the main vehicle at the current intersection; a judgment unit, used to obtain road points on all the lanes, and judge whether all the lanes are straight lanes based on the road points.

[0015] Optionally, in a fourth implementation manner of the second aspect of the present invention, the judgment unit is specifically used to: calculate the standard deviation of the second orientation angle corresponding to the road point; determine whether the standard deviation is less than or equal to a preset threshold; if so, the lane is a straight lane.

[0016] Optionally, in a fifth implementation manner of the second aspect of the present invention, the first determination module is specifically used to: obtain all boundaries of the current intersection; traverse all boundaries, and summarize the orientation angles of lanes corresponding to all boundaries to obtain a third orientation angle of all boundaries; based on the third orientation angle, match all lanes entering and exiting the intersection to obtain a matching result; based on the matching result, determine a matching pair of straight lanes entering and exiting the intersection in the target frame.

[0017] A third aspect of the present invention provides a scene screening device based on lane matching, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via a line;

[0018] The at least one processor calls the instructions in the memory to enable the lane matching-based scene screening device to perform each step of the above-mentioned lane matching-based scene screening method.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the various steps of the above-mentioned lane matching-based scene screening method.

[0020] In the technical solution provided by the present invention, the road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of the matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to a crossroads scene. Through the analysis of the road test data and the automatic screening algorithm for the crossroads scene of the lane matching, the technical problems of low crossroads scene screening efficiency and low recognition accuracy in the prior art are solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a first embodiment of a scene screening method based on lane matching provided by the present invention;

[0022] Figure 2 A schematic diagram of a second embodiment of the scene screening method based on lane matching provided by the present invention;

[0023] Figure 3 A schematic diagram of a third embodiment of the scene screening method based on lane matching provided by the present invention;

[0024] Figure 4 A schematic diagram of a first embodiment of a scene screening device based on lane matching provided by the present invention;

[0025] Figure 5 A schematic diagram of a second embodiment of a scene screening device based on lane matching provided by the present invention;

[0026] Figure 6 A schematic diagram of an embodiment of a lane matching-based scene screening device provided by the present invention. DETAILED DESCRIPTION

[0027] The embodiment of the present invention provides a scene screening method, device, equipment and storage medium based on lane matching. In the technical solution of the present invention, firstly, road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to an intersection scene. The automatic intersection scene screening algorithm based on the analysis of the road test data and lane matching solves the technical problems of low intersection scene screening efficiency and recognition accuracy in the prior art.

[0028] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , a first embodiment of a scene screening method based on lane matching in an embodiment of the present invention includes:

[0030] 101. Collecting road test data of the main vehicle in the driving scene;

[0031] In this embodiment, the road test data may specifically include data obtained from multiple autonomous driving systems during the road test. For example, sensor data in the perception system of autonomous driving, or satellite positioning data in the positioning system of autonomous driving, or processor operation data in the planning system of autonomous driving, etc. Of course, the road test data listed above is only a schematic illustration. During specific implementation, according to specific processing scenarios and user needs, other types of parameter data related to road testing of autonomous driving can be introduced as road test data in addition to the road test data listed above. This specification does not limit this.

[0032] Specifically, the preset database may store road test data of multiple autonomous driving systems related to road testing of autonomous driving vehicles.

[0033] In one embodiment, the above-mentioned multiple autonomous driving systems may specifically include: an autonomous driving perception system, an autonomous driving positioning system, an autonomous driving planning system, and an autonomous driving control system. Of course, the above-mentioned systems are only a schematic illustration. In specific implementation, according to the specific characteristics of the autonomous driving algorithm model involved, the above-mentioned system may also include other types of autonomous driving systems. For example, an autonomous driving emergency response system, etc. This specification does not limit this.

[0034] In one embodiment, the server may be connected to a preset database. The server may save the road test data records of each autonomous driving system received each time in a preset database. When saving, the server may also record the data name or number of the road test data of each autonomous driving system, the corresponding system and other basic information in the preset database. Furthermore, the server may also record additional information such as the collection time of each road test data, the name or number of the road test data that has an association relationship with each road test data, and the specific association method in the preset database. In addition, the server may also store the data processing results of the road test data obtained in different scenarios in the preset database.

[0035] 102. Analyze the road test data to obtain basic vehicle information and semantic map data of the main vehicle;

[0036] In this embodiment, according to the preset data processing rules, the drive test data is automatically processed accordingly (for example, data screening, etc.), so as to obtain the data processing results of the drive test data required by the user, that is, the basic vehicle information and semantic map data of the main vehicle. In specific implementation, according to the specific situation and the target processing rules, other types of data processing can also be performed on the drive test data. For example, feature extraction processing, indicator evaluation processing, data calculation, etc. This specification does not limit this.

[0037] In one embodiment, considering that when processing the road test data of the autonomous driving system, it is often necessary to use the road test data of other autonomous driving systems in combination, the method may also include: identifying and extracting at least one second road test data related to the road test data from the preset database as associated data. According to the associated data, determining at least one second processing rule related to the road test data. Obtaining the target processing rule by performing logical calculation on the at least one first data processing rule and the at least one second processing rule. Using the target processing rule to perform data screening on the road test data to obtain a data processing result of the road test data.

[0038] In one embodiment, the above-mentioned associated data can be specifically understood as a type of data that will be directly or indirectly used in the subsequent specific processing of the road test data. Specifically, the above-mentioned associated data can be intermediate data that needs to be used when calculating a certain result data based on the road test data. For example, the above-mentioned associated data can be time data that needs to be used when calculating speed data based on distance data. The above-mentioned associated data can also be a reference data that can be selected when tracking and comparing the calculation results based on the road test data. For example, the above-mentioned associated data can be a calculation result obtained historically based on the same road test data. Of course, the above-mentioned associated data listed is only a schematic illustration. In specific implementation, according to specific circumstances and processing requirements, the above-mentioned associated data can also include other data related to the road test data. This specification does not limit this.

[0039] The associated data may be road test data belonging to a different autonomous driving system than the road test data, or may be road test data belonging to the same autonomous driving system as the road test data.

[0040] Furthermore, the server may determine the road test data related to the drive test data from the preset database as the associated data according to the additional information of the found drive test data in the preset database, or according to the associated relationship automatically matched based on the drive test data.

[0041] 103. Based on the basic vehicle information and the semantic map data, determine the target position point with the shortest distance from the host vehicle on the preset semantic map, and determine whether the host vehicle is at the intersection in the current frame based on the target position point;

[0042] In this embodiment, the target position point with the shortest distance from the main vehicle on the preset semantic map is determined, and based on the target position point, it is judged whether the main vehicle is at the intersection. It should be noted that in the semantic map, the main vehicle coordinates are used as the center of the circle, and an initial area is selected according to the preset selection radius as the basis for the selection range. Optionally, the initial area can be a prototype or other shapes, such as a square, which is not limited in this embodiment. It should be noted that the above-mentioned semantic map is a high-precision map that annotates road points and various actual road information. Therefore, the above-mentioned initial area can include the road distribution status of the area where the main vehicle is located, and its distribution status is characterized by multiple road points.

[0043] In this embodiment, the computer device can use a preset screening rule to screen the multiple road points in the above initial area. Optionally, it can be a one-time screening or a step-by-step screening, which is not limited in this embodiment. Since the above screening rule is a rule for screening road points based on at least one of the driving speed of the host vehicle, the direction of the road point, the distance between the road point and the host vehicle coordinates, the distance between the road points, and the connectivity relationship between the road points, the computer device uses the screening rule to screen the multiple roads in the above initial area to obtain the screened target location point.

[0044] 104. If yes, determine that the host vehicle is located at the intersection in the current frame, and mark the current frame as the target frame;

[0045] In this embodiment, due to the complex lane information at the intersection, this method makes a case-by-case judgment on the main vehicle at the intersection and the main vehicle not at the intersection. In this step, the intersection map information of the nearest position point of the main vehicle on the semantic map is extracted to judge whether the main vehicle is at the intersection. When the main vehicle is at the intersection, it is determined that the main vehicle is at the intersection in the current frame, and the current frame is marked as the target frame.

[0046] 105. Determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the semantic map information and the position information of the host vehicle at the current intersection;

[0047] In this embodiment, the trajectory information corresponding to the intersection is obtained. It should be noted that there are no excessive restrictions on the category of the intersection, the category of the trajectory information, etc. The trajectory information may include multiple trajectory points. The intersection may include a composite intersection and a crossroad. It should be noted that a composite intersection refers to an intersection composed of at least two nodes and / or at least two sections, including but not limited to a crossroad, a T-junction, etc.

[0048] Specifically, the trajectory information corresponding to the intersection may include the trajectory information within the set area corresponding to the intersection, and the shape and size of the set area are not excessively limited. For example, the set area may be an area formed by spreading outward according to the set value with the location of the intersection as the center. For example, the set area may be a rectangular area with the location of the intersection as the center. According to the semantic map information and the location information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined.

[0049] Specifically, the target frame image is binarized, and then the Sobe1 edge detection algorithm is used to detect the edge information of the lane lines in the image. According to the width factor of the lane lines, the useless edge information in the image is eliminated, leaving the edge information of the lane lines, and then the Surf algorithm is used to extract the feature points of the straight lane line image.

[0050] 106. Determine whether the number of straight lane matching pairs is greater than a preset threshold. If so, determine that the target frame corresponds to an intersection scene.

[0051] In this embodiment, for frame images that meet the conditions, the in-out lane pair is searched for according to the map information and the position of the main vehicle at the current intersection. The steps mainly include the following two steps: the judgment of the straight lane, that is, the standard deviation s of the orientation angles of all road points on each lane is calculated. If it exceeds the given threshold T1, it is regarded as a curve, otherwise it is a straight road. At the same time, for the lane matching of the in-out intersection, each boundary of the current intersection is traversed, and the orientation angles of all lanes on the boundary are summarized. The method includes taking the average, taking the median, etc., and taking it as the orientation angle agg_bound_yaw of the boundary; matching all agg_bound_yaw, if the difference is less than the given threshold T2, it is regarded as finding a set of lane pairs entering and exiting the intersection.

[0052] Furthermore, when the number of lane matching pairs is 4 or more, it is determined that the target frame image is an intersection scene.

[0053] In the embodiment of the present invention, the road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to an intersection scene. The automatic intersection scene screening algorithm of the analysis of the road test data and lane matching solves the technical problems of low intersection scene screening efficiency and low recognition accuracy in the prior art.

[0054] See also Figure 2 , a second embodiment of the scene screening method based on lane matching in an embodiment of the present invention includes:

[0055] 201. Metadata in drive test data transmitted based on a preset wireless network;

[0056] In this embodiment, an index interface is used to receive metadata in the autonomous driving road test data transmitted by the wireless cellular network. By receiving metadata through the index interface, the user can index the corresponding full data according to the metadata in the data center to meet the user's diversified functional usage needs.

[0057] Specifically, the metadata includes the accident identifier, test route, test version, accident level, accident label, and the accident description voice of the safety officer. The full amount of data includes lidar data, camera data, map data, positioning data, perception result data, planning result data and control output data.

[0058] 202. Determine the full amount of data corresponding to the metadata;

[0059] In this embodiment, by receiving metadata in the autonomous driving road test data transmitted by the wireless cellular network, since the amount of metadata is small, the metadata can be received in real time during the road test using the wireless cellular network, which can help the data center to timely count the road test data and shorten the analysis iteration cycle. However, due to its huge amount of data, the full amount of data and its corresponding metadata need to be saved through the on-board hard disk after the autonomous driving vehicle completes the road test, and then it can be transmitted to the data center after the autonomous driving vehicle returns to the data upload point, thereby helping to reduce the cost of data backhaul.

[0060] Specifically, the metadata and the full data corresponding to the metadata in the autonomous driving road test data are downloaded from the vehicle hard disk; the metadata and the full data corresponding to the metadata downloaded from the vehicle hard disk are uploaded to the upload service device. The upload service device is controlled to upload the metadata and the full data corresponding to the metadata to the data center.

[0061] 203. Fusing the metadata and the full data to obtain the road test data of the main vehicle in the driving scene;

[0062] In this embodiment, data fusion refers to an information processing technology that analyzes, integrates, and combines data from multiple sources (hereinafter referred to as multi-source data) to complete the required decision-making and evaluation tasks. Its purpose is to fuse multiple originally scattered and independent data together to discover data patterns and trends and enhance data value.

[0063] Specifically, dimension relationship records are extracted from a pre-established dimension relationship record table based on the dimension relationship record identifier of each data to be fused. The dimension relationship record table is used to characterize the mapping relationship between dimensions, dimension relationship records, and dimension relationship record identifiers. The dimension relationship record table includes one or more dimensions. The dimension values ​​corresponding to each dimension in the one or more dimensions constitute a dimension relationship record. Each dimension relationship record corresponds to a dimension relationship record identifier that identifies the dimension relationship record. Based on the dimension values ​​included in the extracted dimension relationship records, data fusion is performed on the multi-source data to be fused.

[0064] In this embodiment, based on the dimension relationship record identifier included in each data in the multi-source data to be fused, the dimension relationship record is extracted from the dimension relationship record table, and each dimension relationship record includes one or more dimension values. Based on the dimension values ​​included in the extracted dimension relationship record, the multi-source data to be fused is fused, and the data fusion process is fast and concise, and there is no need to store a data table with equal field value association relationships for each data to be fused, thereby reducing storage occupancy.

[0065] 204. Analyze the road test data to obtain basic vehicle information and semantic map data of the main vehicle;

[0066] 205. Determine the position coordinate data and the first orientation angle of the host vehicle based on the basic vehicle information and the semantic map data;

[0067] In this embodiment, based on the basic vehicle information and semantic map data, the first yaw angle of the speed direction of the main vehicle in all frames in the driving scene and the position coordinate data of the main vehicle in the lane travel direction of the target position point are extracted. Specifically, for the main vehicle, the first yaw angle of the speed direction of each frame of the main vehicle and the position coordinate data of the lane travel direction of the nearest position point of the main vehicle on the semantic map are first extracted.

[0068] 206. Determine the orientation of the host vehicle and the target road point closest to the host vehicle based on the position coordinate data and the first orientation angle;

[0069] In this embodiment, based on the position coordinate data and the first orientation angle, the preset parking orientation of the parking space where the host vehicle is located is determined, wherein the preset parking orientation, i.e., the host vehicle orientation, refers to the preset parking direction of the target parking space;

[0070] Specifically, before the vehicle determination step, there is a step of establishing a database, wherein the database includes the range of vehicles within the camera coverage area, semantic map data of the camera coverage area, and preset parking space orientations of parking spaces within the camera coverage area.

[0071] After obtaining the position coordinate data and the first orientation angle of the host vehicle, the position coordinate range of the area where the host vehicle is located is searched in the database, and the preset parking orientation of the parking space and the target road point closest to the host vehicle are obtained.

[0072] 207. Taking the target road point as the center and the preset distance as the radius, obtain a semantic map within the radius;

[0073] In this embodiment, the coordinates of the target road point are taken as the center of the circle, and an initial area is selected according to the preset selection radius as the basis for the selection range, that is, the semantic map within the radius. Optionally, the semantic map within the radius can be circular or in other shapes, such as a square, which is not limited in this embodiment. It should be noted that the semantic map is a high-precision map that annotates road points and various road actual information. Therefore, the semantic map within the radius can include the road distribution status of the area where the target road point is located, and its distribution status is characterized by multiple road points.

[0074] 208. Based on the semantic map, determine whether the main vehicle is at an intersection in the current frame;

[0075] In this embodiment, the computer device can use a preset screening rule to screen multiple road points in the semantic map within the above radius. Optionally, it can be a one-time screening or a step-by-step screening, which is not limited in this embodiment. Since the above screening rule is a rule for screening road points based on at least one of the driving speed of the main vehicle, the direction of the road point, the distance between the road point and the main vehicle coordinates, the distance between the road points, and the connectivity relationship between the road points, the computer device uses the screening rule to screen multiple roads in the semantic map within the above radius to obtain the screened target location point.

[0076] In this embodiment, due to the complex lane information at the intersection, this method makes a case-by-case judgment on the main vehicle at the intersection and the main vehicle not at the intersection. In this step, the intersection map information of the nearest position point of the target main vehicle on the semantic map is extracted to judge whether the main vehicle is at the intersection.

[0077] 209. If yes, determine that the host vehicle is located at the intersection in the current frame, and mark the current frame as the target frame;

[0078] 210. Determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the semantic map information and the position information of the host vehicle at the current intersection;

[0079] 211. Determine whether the number of straight lane matching pairs is greater than a preset threshold. If so, determine that the target frame corresponds to an intersection scene.

[0080] Steps 204, 209-211 in this embodiment are similar to steps 102, 104-106 in the first embodiment, and are not described again here.

[0081] In the embodiment of the present invention, the road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to an intersection scene. The automatic intersection scene screening algorithm based on the analysis of the road test data and lane matching solves the technical problems of low intersection scene screening efficiency and low recognition accuracy in the prior art.

[0082] See also Figure 3 , a third embodiment of the scene screening method based on lane matching in the embodiment of the present invention includes:

[0083] 301. Collecting road test data of the main vehicle in the driving scene;

[0084] 302. Analyze the road test data to obtain basic vehicle information and semantic map data of the main vehicle;

[0085] 303. Based on the basic vehicle information and the semantic map data, determine the target position point with the shortest distance from the host vehicle on the preset semantic map, and determine whether the host vehicle is at the exit in the current frame based on the target position point;

[0086] 304. If yes, determine that the host vehicle is located at the intersection in the current frame, and mark the current frame as the target frame;

[0087] 305. Determine all lanes entering and exiting the intersection based on the semantic map information and the location information of the host vehicle at the current intersection;

[0088] In this embodiment, when identifying lane lines, it is necessary to first obtain the target frame image containing lane lines collected at the current moment, and then use the pre-trained lane line recognition model to recognize the target frame image containing lane lines, so as to obtain the lane recognition result at the current moment, which may include one or more lane lines. According to the semantic map information and the position information of the main vehicle at the current intersection, the lane line recognition result obtained at the current moment is uniformly sampled for each lane line, so that multiple sampling points corresponding to each lane line can be obtained, and the information of these multiple sampling points is used as the basis for subsequent identification of lane line color and lane line type, thereby reducing the probability of incorrect classification of lane line color or type due to inaccurate recognition due to changes in lighting, large-area wear or occlusion of lane lines, etc.

[0089] Specifically, the recognition of lane line types may be affected by vehicle occlusion and other situations. Therefore, in order to further improve the accuracy of lane line recognition, it is also possible to obtain information such as the lane line type at historical moments such as the previous moment, and use this as a priori condition to constrain the recognition result of the lane line type at the current moment.

[0090] It should be noted here that, since the recognition of lane line color is generally not affected by vehicle occlusion, the lane line color constraint at the historical moment can be used as an optional condition when performing lane line color recognition. In addition, if the current moment is the first moment, the lane line recognition result at the historical moment can be set to null. According to the multiple sampling points corresponding to the lane line at the current moment and the lane line recognition results at the historical moment, all lanes entering and exiting the intersection are determined.

[0091] 306. Calculate the standard deviation of the road point corresponding to the second orientation angle, and determine whether the standard deviation is less than or equal to a preset threshold. If so, the lane is a straight lane.

[0092] In this embodiment, based on the straight lane area in the image, the intersection difference between each lane line and each vehicle key point is obtained; the product of the intersection difference between adjacent lane lines is calculated; the standard deviation of the second orientation angle corresponding to the road point is calculated, and it is determined whether the standard deviation is less than or equal to a preset threshold. If so, the lane is a straight lane.

[0093] 307. Obtain all boundaries of the current intersection, traverse all boundaries, and summarize the orientation angles of lanes corresponding to all boundaries to obtain the third orientation angles of all boundaries;

[0094] In this embodiment, the initial road boundary point cloud is obtained based on the preset laser radar equipment, that is, the positional relationship between the point clouds is used to extract mutation points that meet the road boundary characteristics. Some of these mutation points are not actually real road boundaries, but may be interference from other obstacles, so it is called coarse extraction.

[0095] The road boundary model around the main vehicle in the semantic map is obtained according to the position coordinate data of the main vehicle; wherein the vehicle positioning information is obtained by the vehicle's own tracking algorithm. The initial road boundary point cloud and the road boundary model are fitted in the same coordinate system to obtain rotation and translation parameters. The road boundary point cloud extracted by rough extraction is only a discrete point cloud, some of which are real road boundary points and some are point clouds near the road boundary. Therefore, the real road boundary model is obtained by fitting, and then the real road boundary is identified based on this model.

[0096] According to the rotation and translation parameters, a road boundary model of the radar point cloud is calculated to realize road boundary recognition based on the radar point cloud to obtain all boundaries of the current intersection, traverse all boundaries, and summarize the orientation angles of lanes corresponding to all boundaries to obtain the third orientation angle of all boundaries.

[0097] 308. Based on the third orientation angle, all lanes entering and exiting the intersection are matched to obtain a matching result;

[0098] In this embodiment, based on the target frame that meets the conditions, the in-out lane pair that enters and exits the intersection is searched according to the map information and the position of the main vehicle at the current intersection, which mainly includes the following two steps: the judgment of the straight lane, that is, calculating the standard deviation s of the orientation angles of all road points on each lane. If it exceeds the given threshold T1, it is regarded as a curve, otherwise it is a straight road. The lane matching of the in-out intersection traverses each boundary of the current intersection and summarizes the orientation angles of all lanes on the boundary. The method includes taking the average, taking the median, etc., and taking it as the orientation angle agg_bound_yaw of the boundary; matching all agg_bound_yaw, if the difference is less than the given threshold T2, it is regarded as finding a set of lane pairs entering and exiting the intersection.

[0099] 309. Determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the matching result;

[0100] In this embodiment, the standard deviation s of the orientation angles of all road points on each lane is calculated. If it exceeds a given threshold T1, it is considered a curve, otherwise it is a straight road. For lane matching at the entrance and exit of the intersection, each boundary of the current intersection is traversed, and the orientation angles of all lanes on the boundary are summarized. The method includes taking the average, taking the median, etc., and taking it as the orientation angle agg_bound_yaw of the boundary; matching all agg_bound_yaw, if the difference is less than a given threshold T2, it is considered that a set of lane pairs entering and exiting the intersection is found. Furthermore, according to the matching results, the matching pairs of straight lanes entering and exiting the intersection in the target frame are determined.

[0101] 310. Determine whether the number of straight lane matching pairs is greater than a preset threshold. If so, determine that the target frame corresponds to an intersection scene.

[0102] Steps 301 - 304 and 310 in this embodiment are similar to steps 101 - 104 and 106 in the first embodiment, and are not described in detail here.

[0103] In the embodiment of the present invention, the road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to an intersection scene. The automatic intersection scene screening algorithm of the analysis of the road test data and lane matching solves the technical problems of low intersection scene screening efficiency and low recognition accuracy in the prior art.

[0104] The above describes the scene screening method based on lane matching in the embodiment of the present invention. The following describes the scene screening device based on lane matching in the embodiment of the present invention. Figure 4 , a first embodiment of a scene screening device based on lane matching in an embodiment of the present invention includes:

[0105] The acquisition module 401 is used to acquire the drive test data of the main vehicle in the driving scene, wherein the drive test data includes the traffic scene data acquired when the main vehicle is driving on a real road, and the traffic scene data includes multiple frames of vehicle driving images;

[0106] The parsing module 402 is used to parse the drive test data to obtain basic vehicle information and semantic map data of the host vehicle;

[0107] A judgment module 403 is used to determine the target position point with the shortest distance from the host vehicle on the preset semantic map based on the basic vehicle information and the semantic map data, and to judge whether the host vehicle is at an intersection in the current frame based on the target position point;

[0108] A marking module 404 is used for determining that the host vehicle in the current frame is located at an intersection and marking the current frame as a target frame;

[0109] A first determination module 405 is used to determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the semantic map information and the position information of the host vehicle at the current intersection;

[0110] The second determination module 406 is used to determine whether the number of the straight lane matching pairs is greater than a preset threshold, and if so, determine that the target frame corresponds to an intersection scene.

[0111] In the embodiment of the present invention, the road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to an intersection scene. The automatic intersection scene screening algorithm based on the analysis of the road test data and lane matching solves the technical problems of low intersection scene screening efficiency and low recognition accuracy in the prior art.

[0112] See also Figure 5 , a second embodiment of a scene screening device based on lane matching in an embodiment of the present invention, the scene screening device based on lane matching specifically comprises:

[0113] The acquisition module 401 is used to acquire the drive test data of the main vehicle in the driving scene, wherein the drive test data includes the traffic scene data acquired when the main vehicle is driving on a real road, and the traffic scene data includes multiple frames of vehicle driving images;

[0114] The parsing module 402 is used to parse the drive test data to obtain basic vehicle information and semantic map data of the host vehicle;

[0115] A judgment module 403 is used to determine the target position point with the shortest distance from the host vehicle on the preset semantic map based on the basic vehicle information and the semantic map data, and to judge whether the host vehicle is at an intersection in the current frame based on the target position point;

[0116] A marking module 404 is used for determining that the host vehicle in the current frame is located at an intersection and marking the current frame as a target frame;

[0117] A first determination module 405 is used to determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the semantic map information and the position information of the host vehicle at the current intersection;

[0118] The second determination module 406 is used to determine whether the number of the straight lane matching pairs is greater than a preset threshold, and if so, determine that the target frame corresponds to an intersection scene.

[0119] In this embodiment, the acquisition module 401 is specifically used for:

[0120] Metadata in the drive test data transmitted based on a preset wireless network;

[0121] Determining the full amount of data corresponding to the metadata;

[0122] The metadata and the full data are fused to obtain the road test data of the main vehicle traveling in the driving scene.

[0123] In this embodiment, the determination module 403 is specifically used for:

[0124] Determining the position coordinate data and the first orientation angle of the host vehicle based on the basic vehicle information and the semantic map data;

[0125] Determine the host vehicle orientation and the target road point closest to the host vehicle based on the position coordinate data and the first orientation angle;

[0126] Taking the target road point as the center and a preset distance as the radius, obtaining a semantic map within the radius;

[0127] Based on the semantic map, it is determined whether the host vehicle is at an intersection in a current frame.

[0128] In this embodiment, the first determining module 405 includes:

[0129] A determination unit 4051 is used to determine all lanes entering and exiting the intersection according to the semantic map information and the position information of the host vehicle at the current intersection;

[0130] The judging unit 4052 is configured to obtain road points on all lanes, and judge whether all lanes are straight lanes based on the road points.

[0131] In this embodiment, the determining unit 4052 is specifically used for:

[0132] Calculate the standard deviation of the second orientation angle corresponding to the road point;

[0133] Determining whether the standard deviation is less than or equal to a preset threshold;

[0134] If so, the lane is a straight lane.

[0135] In this embodiment, the first determining module 405 is further configured to:

[0136] Get all boundaries of the current intersection;

[0137] Traversing all the boundaries, and summing up the orientation angles of lanes corresponding to all the boundaries to obtain third orientation angles of all the boundaries;

[0138] Based on the third orientation angle, all lanes entering and exiting the intersection are matched to obtain a matching result;

[0139] According to the matching result, a matching pair of straight lanes entering and exiting the intersection in the target frame is determined.

[0140] In the embodiment of the present invention, the road test data is collected; the road test data is parsed, and based on the obtained basic vehicle information and semantic map data, the position point with the smallest distance of the main vehicle on the semantic map is determined, and based on the position point, it is determined whether the main vehicle is at the intersection in the current frame; if so, the current frame is marked as the target frame; according to the semantic map information and the position information of the main vehicle at the current intersection, the straight lane matching pairs entering and exiting the intersection in the target frame are determined; it is determined whether the number of matching pairs is greater than a preset threshold, and if so, it is determined that the target frame corresponds to an intersection scene. The automatic intersection scene screening algorithm based on the analysis of the road test data and lane matching solves the technical problems of low intersection scene screening efficiency and low recognition accuracy in the prior art.

[0141] above Figure 4 and Figure 5 The scene screening device based on lane matching in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The scene screening device based on lane matching in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0142] Figure 6 6 is a schematic diagram of a scene screening device based on lane matching provided by an embodiment of the present invention. The scene screening device based on lane matching 600 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 610 (for example, one or more processors) and a memory 620, and one or more storage media 630 (for example, one or more mass storage devices) storing application programs 633 or data 632. Among them, the memory 620 and the storage medium 630 may be short-term storage or permanent storage. The program stored in the storage medium 630 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the scene screening device based on lane matching 600. Furthermore, the processor 610 may be configured to communicate with the storage medium 630, and execute a series of instruction operations in the storage medium 630 on the scene screening device based on lane matching 600 to implement the steps of the scene screening method based on lane matching provided by the above-mentioned method embodiments.

[0143] The scene screening device 600 based on lane matching may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input and output interfaces 660, and / or one or more operating systems 631, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 6 The lane matching-based scene screening device structure shown does not constitute a limitation on the lane matching-based scene screening device provided by the present application, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0144] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the above-mentioned scene screening method based on lane matching.

[0145] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0146] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0147] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A scene screening method based on lane matching, characterized in that: The scene screening method based on lane matching includes: Collecting road test data of the main vehicle traveling in a driving scene, wherein the road test data includes traffic scene data collected when the main vehicle is tested on a real road, and the traffic scene data includes multiple frames of vehicle driving images; Parsing the drive test data to obtain basic vehicle information and semantic map data of the host vehicle; Based on the basic vehicle information and the semantic map data, determine the target position point of the main vehicle with the smallest distance on the preset semantic map, and determine whether the main vehicle is at an intersection in the current frame based on the target position point; If yes, determine that the host vehicle in the current frame is located at an intersection, and mark the current frame as a target frame; Determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the semantic map information and the position information of the host vehicle at the current intersection; Determining whether the number of the straight lane matching pairs is greater than a preset threshold, and if so, determining that the target frame corresponds to an intersection scene; The determining whether the host vehicle is at an intersection in the current frame based on the target position point includes: Determining the position coordinate data and the first orientation angle of the host vehicle based on the basic vehicle information and the semantic map data; Determine the host vehicle orientation and the target road point closest to the host vehicle based on the position coordinate data and the first orientation angle; Taking the target road point as the center and a preset distance as the radius, obtaining a semantic map within the radius; Based on the semantic map, it is determined whether the host vehicle is at an intersection in a current frame.

2. The scene screening method based on lane matching according to claim 1, characterized in that: The collecting of the road test data of the main vehicle in the driving scene includes: Metadata in the drive test data transmitted based on a preset wireless network; Determining the full amount of data corresponding to the metadata; The metadata and the full data are fused to obtain the road test data of the main vehicle traveling in the driving scene.

3. The scene screening method based on lane matching according to claim 1, characterized in that: The determining, according to the semantic map information and the position information of the host vehicle at the current intersection, a matching pair of straight lanes entering and exiting the intersection in the target frame includes: Determine all lanes entering and exiting the intersection according to the semantic map information and the position information of the host vehicle at the current intersection; Road points on all the lanes are acquired, and whether all the lanes are straight lanes is determined based on the road points.

4. The scene screening method based on lane matching according to claim 3 is characterized in that: The acquiring of the road points on all the lanes, and judging whether all the lanes are straight lanes based on the road points, comprises: Calculate the standard deviation of the second orientation angle corresponding to the road point; Determining whether the standard deviation is less than or equal to a preset threshold; If so, the lane is a straight lane.

5. The scene screening method based on lane matching according to claim 1, characterized in that: The determining, according to the semantic map information and the position information of the host vehicle at the current intersection, a matching pair of straight lanes entering and exiting the intersection in the target frame further includes: Get all boundaries of the current intersection; Traversing all the boundaries, and summing up the orientation angles of lanes corresponding to all the boundaries to obtain third orientation angles of all the boundaries; Based on the third orientation angle, all lanes entering and exiting the intersection are matched to obtain a matching result; According to the matching result, a matching pair of straight lanes entering and exiting the intersection in the target frame is determined.

6. A scene screening device based on lane matching, characterized in that: The scene screening device based on lane matching includes: A collection module, used to collect the drive test data of the main vehicle in the driving scene, wherein the drive test data includes the traffic scene data collected when the main vehicle is tested on a real road, and the traffic scene data includes multiple frames of vehicle driving images; A parsing module, used to parse the drive test data to obtain basic vehicle information and semantic map data of the host vehicle; A judgment module, used to determine the target position point with the shortest distance from the main vehicle on the preset semantic map based on the basic vehicle information and the semantic map data, and to judge whether the main vehicle is at an intersection in a current frame based on the target position point; a marking module, configured to determine that the host vehicle in the current frame is located at an intersection and mark the current frame as a target frame; A first determination module is used to determine a matching pair of straight lanes entering and exiting the intersection in the target frame according to the semantic map information and the position information of the host vehicle at the current intersection; A second determination module is used to determine whether the number of the straight lane matching pairs is greater than a preset threshold, and if so, determine that the target frame corresponds to a crossroads scene; The judgment module is specifically used to: determine the position coordinate data and the first orientation angle of the main vehicle based on the vehicle basic information and the semantic map data; determine the orientation of the main vehicle and the target road point closest to the main vehicle based on the position coordinate data and the first orientation angle; obtain a semantic map within the radius with the target road point as the center and a preset distance as the radius; and determine whether the main vehicle is at an intersection in the current frame based on the semantic map.

7. The scene screening device based on lane matching according to claim 6, characterized in that: The acquisition module is specifically used for: Metadata in the drive test data transmitted based on a preset wireless network; Determining the full amount of data corresponding to the metadata; The metadata and the full data are fused to obtain the road test data of the main vehicle traveling in the driving scene.

8. The scene screening device based on lane matching according to claim 6, characterized in that: The first determination module includes: a determination unit, used to determine all lanes entering and exiting the intersection according to the semantic map information and the position information of the main vehicle at the current intersection; a judgment unit, used to obtain road points on all the lanes, and judge whether all the lanes are straight lanes based on the road points.

9. A scene screening device based on lane matching, characterized in that: The lane matching-based scene screening device comprises: a memory and at least one processor, the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor calls the instructions in the memory to enable the lane matching-based scene screening device to perform each step of the lane matching-based scene screening method as described in any one of claims 1-5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the scene screening method based on lane matching as described in any one of claims 1 to 5 is implemented.

Images