A local dynamic map generation method and system

Abstract

The application discloses a local dynamic map generation method and system, and relates to the field of intelligent traffic management.The method comprises the following steps: performing semantic segmentation and target detection on road images collected at different sampling frequencies to obtain static road semantic information and dynamic target information on the road, so that more real-time road target information can be obtained, static geographic semantic information can be displayed, and dynamic road data such as motor vehicle information, non-motor vehicle information, pedestrian information, regional traffic information and regional accident information can be provided; meanwhile, since the second collection frequency is very short, the collected data is more real-time, and therefore a higher map data update frequency can be realized; and in the application, after hardware facilities are installed, map data update is automatically completed based on software each time, and a large amount of manpower and energy is not required.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent traffic management, and in particular to a local dynamic map generation method and system. BACKGROUND

[0002] In recent years, thanks to the development and application of artificial intelligence technology, sensor technology, 5G communication, and big data technology, intelligent road networks and unmanned driving technology have developed rapidly. High-precision maps are an important software infrastructure in related application fields such as unmanned driving, vehicle-road cooperation, and intelligent network intersection. In intelligent transportation systems, local dynamic maps are one of the core key technologies. Maps can be divided into two modes: weak maps and strong maps. The former is only used for navigation and positioning, and relies on vehicle sensors for environmental perception during driving. The latter can be understood as a container for road sensors, and more detailed road information and road sensor information are integrated into the map. High-precision maps are a strong map construction mode that includes detailed road information such as lane lines, lane centerlines, dashed lines, solid lines, no-entry areas, drivable areas, road slopes, curvatures, obstacles, traffic signs, and other information to meet the needs of autonomous driving programs.

[0003] Currently, many domestic and foreign manufacturers have their own high-precision map collection and production methods: such as using laser radar and camera combinations to periodically survey and update maps through map collection vehicles; or using vehicle-mounted cameras and GPS to obtain road data, and processing in the cloud through spatial generation technology to finally generate a map.

[0004] However, the construction of high-precision maps is done through several map collection vehicles equipped with complex and expensive sensors, including multiple cameras, laser radars, GPS, inertial measurement units, and odometers. Due to the limited number of map collection vehicles, which are special vehicles, the sensors installed on the vehicles are expensive, have high manufacturing and maintenance costs, and the total number of vehicles is limited. Additionally, map updates take a long time: only after the map collection vehicle drives through and records changes can road information be updated, which may take months after the changes occur, limiting the application of scenarios sensitive to map changes. At the same time, map updates are costly, requiring dedicated map collection personnel and engineers to continuously drive map collection vehicles to collect maps, resulting in high map collection costs. Finally, due to the reliance on map collection vehicles for high-precision map generation, it is not possible to incorporate dynamic road information such as vehicle information, pedestrian information, signal light states, intersection queue conditions, road occupancy conditions, temporary road changes and restrictions, and temporary traffic signs. SUMMARY

[0005] To solve the above technical problems, the application provides a local dynamic map generation method and system, which can solve the problems of high construction cost, poor real-time updating and low precision of existing high-precision maps.

[0006] To achieve the above-mentioned purpose, in one aspect, the application provides a local dynamic map generation method, which comprises:

[0007] According to the relative displacement amount obtained from the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position, the corresponding region in the cloud point cloud full set is incrementally updated according to the relative displacement amount.

[0008] The road images collected by the camera according to the first collection frequency are subjected to semantic segmentation to obtain static road semantic information.

[0009] The road images collected by the camera according to the second collection frequency are subjected to target detection to obtain dynamic target information on the road.

[0010] According to the configuration parameter information of the camera, the static road semantic information and the dynamic target information, the spatial coordinates of the semantic pixel points are obtained.

[0011] According to the incrementally updated three-dimensional point cloud data and the spatial coordinates of the semantic pixel points, the local dynamic map is constructed.

[0012] Further, the step of obtaining the relative displacement amount from the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position comprises:

[0013] According to the preset time, the road segment point cloud data collected by each radar sensor is subjected to point cloud data accumulation respectively.

[0014] The accumulated point cloud data is subjected to feature point extraction, and the relative displacement amount is obtained according to the extracted feature point position and the preset initial feature position.

[0015] According to the relative displacement amount, the road segment point cloud data collected by each radar sensor is compensated.

[0016] Further, the step of incrementally updating the three-dimensional point cloud of the corresponding region in the cloud point cloud full set according to the relative displacement amount comprises:

[0017] The road segment point cloud data collected by each radar sensor is subjected to point cloud splicing.

[0018] The spliced data is subjected to motion object filtering through a preset motion detection algorithm.

[0019] According to the relative displacement amount, the corresponding region in the filtered road segment point cloud data is incrementally updated according to the relative displacement amount.

[0020] Further, the step of performing semantic segmentation on the road image collected by the camera at the first collection frequency to obtain static road semantic information comprises:

[0021] The static road semantic information is obtained by performing semantic segmentation on the road image collected by the camera at the first collection frequency through the preset SAM model.

[0022] Further, the step of performing target detection on the road image collected by the camera at the second collection frequency to obtain dynamic target information on the road comprises:

[0023] The dynamic target information on the road is obtained by performing target detection on the road image collected by the camera at the second collection frequency through the preset YOLOv5 model.

[0024] In another aspect, the application provides a local dynamic map generation system, comprising: an acquisition unit configured to acquire a relative displacement amount from a feature point position extracted from point cloud data collected by a radar sensor and a preset initial position, and perform incremental updating of a corresponding region in a cloud point cloud set according to the relative displacement amount;

[0025] A semantic segmentation unit is configured to perform semantic segmentation on a road image collected by a camera at a first collection frequency to obtain static road semantic information.

[0026] The acquisition unit is configured to perform target detection on a road image collected by a camera at a second collection frequency to obtain dynamic target information on the road.

[0027] The acquisition unit is further configured to acquire spatial coordinates of semantic pixel points according to configuration parameter information of the camera, static road semantic information, and dynamic target information.

[0028] A construction unit is configured to construct the local dynamic map according to the incrementally updated three-dimensional point cloud data and the spatial coordinates of the semantic pixel points.

[0029] Further, the acquisition unit is specifically configured to accumulate point cloud data collected by each radar sensor according to a preset time; extract feature points from the accumulated point cloud data, and acquire a relative displacement amount from a feature point position and a preset initial feature position; and compensate the road segment point cloud data collected by each radar sensor according to the relative displacement amount.

[0030] Further, the acquisition unit is specifically further configured to perform point cloud splicing on the road section point cloud data collected by each radar sensor, perform motion object filtering on the spliced data through a preset motion detection algorithm, and perform incremental updating of three-dimensional point clouds on the corresponding region in the filtered road section point cloud data according to the relative displacement amount.

[0031] Further, the semantic segmentation unit is specifically configured to perform semantic segmentation on the road images collected by the camera at the first collection frequency through a preset SAM model to obtain static road semantic information.

[0032] Further, the acquisition unit is specifically further configured to perform target detection on the road images collected by the camera at the second collection frequency through a preset YOLOv5 model to obtain dynamic target information on the road.

[0033] The local dynamic map generation method and system provided by the application can obtain more real-time road target information by performing semantic segmentation and target detection on road images collected at different sampling frequencies to obtain static road semantic information and dynamic target information on the road, can not only display static geographic semantic information but also provide road dynamic data such as motor vehicles, non-motor vehicles, pedestrian information, regional traffic information, and regional accident information, can achieve a higher map data update frequency because the second collection frequency is very short and the collected data is more real-time, and each map data update is automatically completed based on software without additional manpower and effort after the hardware facilities are deployed and installed. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a flowchart of a local dynamic map generation method provided by the application;

[0035] Figure 2 is a structural schematic diagram of a local dynamic map generation system provided by the application. DETAILED DESCRIPTION

[0036] The technical solutions of the application will be further described in detail below with reference to the drawings and embodiments.

[0037] Before the method steps are described in detail, the hardware layout scenario adopted by the application is first described. In the local dynamic map construction method described in the application, local refers to a specific road adjacent area rather than a global map, dynamic refers to that the local map contains highly dynamic data and is a three-dimensional map structure divided into three layers: a static three-dimensional map layer, a road identification data layer, and a real-time perception data layer, and the system component includes a camera, a laser radar, a roadside broadcasting unit, an edge calculator, a switch, and a cloud server.

[0038] The camera and the laser radar can be combined into one hardware through hardware assembly, and the functions and effects are the same as when they are installed separately. The roadside broadcasting unit is used to broadcast information to passing vehicles in real time. The edge computer is used to aggregate information synchronization data and is responsible for communication with the cloud server. The cloud server aggregates all road data, constructs and stores a local dynamic map, and provides services to the outside. The laser radar and the camera are deployed at an interval of 300 to 500 meters, and are preferably deployed in the center of the road, so that their coverage range can cover both directions of the road to save costs. The roadside broadcasting unit is installed and deployed according to its coverage range. All roadside devices are connected through a switch and a cloud server. The system modules include a perception module, incremental road three-dimensional reconstruction, real-time perception data aggregation, a local dynamic map database, and a map data interface and a roadside broadcasting protocol. The perception module includes a camera and a laser radar, which are installed on the roadside. The roadside broadcasting unit is responsible for implementing the data broadcasting protocol. The incremental road three-dimensional reconstruction and the real-time perception data aggregation are completed in the cloud server, and the data is stored in the cloud database, which is also responsible for providing the map data interface. The perception module uses the laser radar to capture point cloud data of the road and the bridge regularly or on demand, and the camera captures road images in real time to identify motor vehicle flow, non-motor vehicle flow, pedestrians, road signs, obstacles, and other information, which are collectively referred to as camera perception data.

[0039] It should be noted that each laser radar is only responsible for capturing point cloud data of the road and the bridge near its installation location. The point cloud data of multiple laser radars on a road is uploaded to the cloud for incremental three-dimensional reconstruction of road geographic information. The perception data is uploaded to the cloud and stored in the local dynamic map database, and is also synchronized to several adjacent roadside broadcasting devices, which are responsible for broadcasting the perception data. The cloud server maintains a complete set of local dynamic maps for the road area. Each roadside perception device contributes part of the information of the map. After the cloud server aggregates the point cloud data and perception data uploaded by each laser radar device, it updates the corresponding areas in the local dynamic map incrementally. The system provides a bottom road three-dimensional map service and dynamic perception data to the outside through the map data interface, broadcasts dynamic perception data through the roadside broadcasting unit according to the broadcasting protocol, and the connected cars within the broadcast coverage range of the roadside device receive the broadcast to obtain real-time dynamic perception data.

[0040] The method steps are described in detail as follows. Figure 1 As shown in the method steps, the local dynamic map generation method provided by the embodiment of the present application includes the following steps:

[0041] 101、According to the relative displacement amount obtained from the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position, the corresponding region in the cloud point cloud full set is incrementally updated.

[0042] For the embodiment of the application, step 101 can specifically include: accumulating the road section point cloud data collected by each radar sensor respectively according to a preset time; extracting feature points from the accumulated point cloud data, and obtaining a relative displacement amount according to the feature point position and the preset initial feature position; compensating the road section point cloud data collected by each radar sensor according to the relative displacement amount. The road section point cloud data collected by each radar sensor is spliced into point clouds; the spliced data is filtered through a preset motion detection algorithm; and the corresponding region in the filtered road section point cloud data is incrementally updated according to the relative displacement amount.

[0043] Specifically, for example, the layer is a static three-dimensional map layer. By distributing laser radars on the road at a certain interval, each laser radar needs to be positioned and the attitude angle measured using a global positioning system such as Beidou or GPS when installed, and the road feature data is initialized and the positioning result initial feature position is recorded. The scanning of data is generally performed at a fixed period or manually according to needs. The fixed period is performed once a day at night. After starting the scanning, the laser radar uses a double reflection mode to scan the road section covered thereby at a speed of generating at least 400,000 points per second, and the scanning is ended after ten minutes, the current point cloud accumulation is completed once per minute, then the feature points are extracted, the feature point position is compared with the initial feature position to obtain a relative displacement amount, and the relative displacement amount is compensated into the point cloud, and after completion, the result is uploaded to a cloud server. After the cloud server collects the data uploaded by each laser radar, the point cloud is spliced, the redundancy is removed, the moving objects are filtered through motion detection, and the moving objects are incrementally updated into the corresponding region in the cloud point cloud full set, and the incrementally updated three-dimensional point cloud is completed.

[0044] 102、According to the relative displacement amount obtained from the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position, the corresponding region in the cloud point cloud full set is incrementally updated.

[0045] For the embodiment of the application, step 101 can specifically include: accumulating the road section point cloud data collected by each radar sensor respectively according to a preset time; extracting feature points from the accumulated point cloud data, and obtaining a relative displacement amount according to the feature point position and the preset initial feature position; compensating the road section point cloud data collected by each radar sensor according to the relative displacement amount. The road section point cloud data collected by each radar sensor is spliced into point clouds; the spliced data is filtered through a preset motion detection algorithm; and the corresponding region in the filtered road section point cloud data is incrementally updated according to the relative displacement amount.

[0046] Specifically, for example, the layer is a road sign data layer, the update frequency can be fixed period or manually on demand through the photos captured by the camera, and the fixed period is performed once a day when the vehicle is less. The captured image is subjected to semantic segmentation through a SAM model, and the SAM model includes an image encoder and a fast prompt encoder or a mask decoder, and the process is as follows: first, the original image is input into a pre-trained ViT-H image encoder to calculate image embedding, prompt embedding is obtained through a prompt word encoder, and finally the mask decoder maps the two and output token to the mask, then the image embedding is up-sampled, and the output token is mapped to the classifier, then the foreground probability to the mask can be calculated to obtain the segmentation result.

[0047] It should be noted that the SAM model used in the embodiment of the present application greatly improves the semantic segmentation efficiency of the image on the basis of ensuring the segmentation effect, without using expensive GPU chips, and can run on a common CPU, greatly reducing the system application cost, and can quickly update the semantic information on the map on a large scale. After obtaining the segmentation result, useless information is filtered out, and only the map semantic information, such as lane lines, zebra crossings, traffic lights, and driving direction signs, is left, and then the spatial coordinates of the semantic pixel points are calculated according to the focal length, camera positioning coordinates and state angle.

[0048] 103. Perform target detection on the road images collected by the camera at the second collection frequency to obtain dynamic target information on the road.

[0049] For the embodiment of the present application, step 103 can specifically include: performing target detection on the road images collected by the camera at the second collection frequency through a pre-set YOLOv5 model to obtain dynamic target information on the road.

[0050] Specifically, for example, the layer is a real-time perception data layer. This layer still uses photo data captured by the camera, and the update frequency is at least 10Hz, that is, 10 photos per second, and then a YOLOv5 model is used for target detection for each captured photo to obtain dynamic target information on the road, such as motor vehicles, non-motor vehicles, pedestrians and the like, and then the spatial coordinates of the semantic pixel points are calculated according to the focal length, camera positioning coordinates and state angle.

[0051] 104. Obtain the spatial coordinates of the semantic pixel points according to the configuration parameter information of the camera, the static road semantic information and the dynamic target information.

[0052] 105. Construct the local dynamic map according to the incrementally updated three-dimensional point cloud data and the spatial coordinates of the semantic pixel points.

[0053] It should be noted that the three layers of data are executed according to the respective update frequency and actual update demand, and the finally constructed map layer data is uploaded to the local dynamic map database in the cloud server. Among them, the static three-dimensional map layer and the road identification data layer provide upgrade service for vehicles through OTA mode, generally trigger upgrade when the map changes, but the upgrade interval is not less than 1 month, and the vehicle triggering the map upgrade downloads the latest map data through the system interface; the real-time sensing data layer sends data to the nearest several roadside devices in real time, and broadcasts to the passing vehicles through them in time, and uploads the sensing data to the cloud server in real time, which can realize real-time road condition display at the vehicle and pedestrian level on the cloud platform level.

[0054] The local dynamic map generation method provided by the embodiment of the application can obtain more real-time road target information by performing semantic segmentation and target detection on road images collected at different sampling frequencies to obtain static road semantic information and dynamic target information on the road, can not only display static geographic semantic information, but also can provide road dynamic data such as motor vehicle, non-motor vehicle, pedestrian information, regional traffic information, and regional accident information; at the same time, since the second collection frequency is very short, the collected data is more real-time, and therefore a higher map data update frequency can be realized; and after the hardware facilities are deployed and installed in the application, each map data update is automatically completed based on software, without additional manpower and effort.

[0055] To realize the method provided by the embodiment of the application, the embodiment of the application provides a local dynamic map generation system, as shown in Figure 2 The system comprises an acquisition unit 21, a semantic segmentation unit 22, and a construction unit 23.

[0056] The acquisition unit 21 is configured to acquire a relative displacement amount from a feature point position extracted from point cloud data collected by a radar sensor and a preset initial position, and perform incremental update of three-dimensional point cloud in a corresponding region in a cloud point cloud set according to the relative displacement amount.

[0057] The semantic segmentation unit 22 is configured to perform semantic segmentation on a road image collected by a camera at a first collection frequency to obtain static road semantic information.

[0058] The acquisition unit 21 is configured to perform target detection on a road image collected by a camera at a second collection frequency to obtain dynamic target information on the road.

[0059] The acquisition unit 21 is further configured to acquire a spatial coordinate of a semantic pixel point according to configuration parameter information of the camera, the static road semantic information, and the dynamic target information.

[0060] The constructing unit 23 is configured to construct the local dynamic map according to the incrementally updated three-dimensional point cloud data and the spatial coordinates of the semantic pixel points.

[0061] Further, the acquisition unit 21 is specifically configured to accumulate the road section point cloud data collected by each radar sensor in a point cloud data manner according to a preset time; extract feature points from the accumulated point cloud data, and obtain a relative displacement amount according to the positions of the extracted feature points and preset initial feature positions; and compensate the road section point cloud data collected by each radar sensor according to the relative displacement amount.

[0062] Further, the acquisition unit 21 is specifically configured to further perform point cloud splicing on the road section point cloud data collected by each radar sensor; perform motion object filtering on the spliced data by using a preset motion detection algorithm; and perform incrementally update of three-dimensional point cloud on a corresponding region in the filtered road section point cloud data according to the relative displacement amount.

[0063] Further, the semantic segmentation unit 22 is specifically configured to perform semantic segmentation on road images collected by the camera in a first collection frequency by using a preset SAM model, to obtain static road semantic information.

[0064] Further, the acquisition unit 21 is specifically configured to further perform target detection on road images collected by the camera in a second collection frequency by using a preset YOLOv5 model, to obtain dynamic target information on the road.

[0065] The local dynamic map generation system provided by the application can obtain more real-time road target information by performing semantic segmentation and target detection on road images collected in different sampling frequencies, respectively obtaining static road semantic information and dynamic target information on the road, and can not only display static geographic semantic information, but also provide road dynamic data such as motor vehicles, non-motor vehicles, pedestrian information, regional traffic information, and regional accident information; at the same time, since the second collection frequency is very short, the real-time performance of the collected data is stronger, so that a higher map data update frequency can be realized; and in the application, once the hardware facilities are deployed and installed, the map data update is automatically completed based on software, without the need for additional manpower and effort.

[0066] It should be understood that the specific order or hierarchy of steps in the processes disclosed should not be taken as a limitation of the example methods. One having ordinary skill in the art will understand that the specific order or hierarchy of steps in the processes can be re-arranged while remaining within the scope of the present disclosure. The accompanying method claims present elements of the various steps in a sample order, and are not meant to be limited to the specific order or hierarchy presented.

[0067] In the above detailed description, various features are grouped together in single embodiments for the purpose of streamlining the disclosure. This method of disclosure, however, is not to be interpreted as reflecting a necessity or desire to group the claimed embodiments together in such a way. In addition, each of the various embodiments described for a single embodiment can be used in combination with another embodiment; i.e., each stated embodiment is an alternative to the other embodiments, and can be combined with other embodiments to produce yet further embodiments. Thus, the claims are not to be interpreted as being limited to the singular embodiments contained in the specific embodiments, but rather are to be interpreted to cover all compatible combinations of the embodiments included within the scope of the claims.

[0068] The disclosed embodiments herein can be described with respect to the following example embodiments. One skilled in the art will readily recognize that the example embodiments can be practiced with materials, methods and equivalents not specifically referred to herein, and that the present disclosure is not limited to the example embodiments set forth herein. The following examples are intended to further clarify the application but are not intended to limit its scope in any way.

[0069] The above description includes example embodiments. Of course, not all possible combinations of components or method steps are described, but one of ordinary skill in the art will recognize that further combinations and permutations of each embodiments are also possible. The description herein is intended to cover all such alternatives, modifications and variations as are included within the scope of the appended claims. Furthermore, the use of the term "comprising" in the description or claims is not intended to exclude any embodiments except those in which the only components are as specifically recited in the claim. In addition, the use of the term "or" is intended to mean an inclusive "or" rather than an exclusive "or" unless explicitly stated otherwise. Furthermore, the use of the term "a" or "an" is intended to mean "one or more" unless explicitly stated otherwise.

[0070] Those skilled in the art will further appreciate that the various illustrative logical blocks, modules, and steps described in connection with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans can implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present embodiments.

[0071] The various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. The various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality, without limitation. Depending upon the implementation, the various illustrative blocks, modules, circuits, and steps could be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of them. The software implemented as a computer program, software, or firmware in electronic form, which includes but is not limited to software

[0072] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as distinct components in a user terminal. Exemplary components for a user terminal include, but are not limited to, mobile computing devices, netbooks, notebooks, ultrabooks, laptops, desktops, handheld devices, tablets, televisions, cameras, personal digital assistants, media players, game consoles, or any other computing device.

[0073] In one or more exemplary designs, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. Storage media can be any available media that can be accessed by a computer. By way of example, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general-purpose or special-purpose processor. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or other wire-based, fiber- optic based, or wireless based communications, then the coaxial cable, fiber optic cable, twisted pair, DSL, or other wire-based, fiber-optic based, or wireless based communications are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, DVD, floppy disk, and Blu-ray® disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0074] The specific implementation described above is illustrative for purposes of teaching the present application. The application should not be limited, however, to these specific implementation. Any modifications, equivalents, or improvements, as long as they are within the spirit and scope of the present application are included in the scope of the present application.

Claims

1. A method for local dynamic map generation, characterized in that, The method comprises: According to the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position, the relative displacement amount is obtained, and the corresponding region in the cloud end point cloud full set is incrementally updated according to the relative displacement amount. The road images collected by the camera at the first collection frequency are subjected to semantic segmentation to obtain static road semantic information. The road images collected by the camera at the second collection frequency are subjected to target detection to obtain dynamic target information on the road. According to the configuration parameter information of the camera, the static road semantic information and the dynamic target information, the spatial coordinates of the semantic pixel points are obtained. The local dynamic map is constructed according to the incrementally updated three-dimensional point cloud data and the spatial coordinates of the semantic pixel points. The step of collecting the road images by the camera at the first collection frequency and subjecting them to semantic segmentation to obtain static road semantic information comprises: The road images collected by the camera at the first collection frequency are subjected to semantic segmentation by a preset SAM model to obtain static road semantic information. The step of collecting the road images by the camera at the second collection frequency and subjecting them to target detection to obtain dynamic target information on the road comprises: The road images collected by the camera at the second collection frequency are subjected to target detection by a preset YOLOv5 model to obtain dynamic target information on the road.

2. The method of claim 1, wherein, The step of obtaining the relative displacement amount according to the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position comprises: The point cloud data collected by each radar sensor is accumulated according to a preset time. The accumulated point cloud data is subjected to feature point extraction, and the relative displacement amount is obtained according to the extracted feature point position and the preset initial feature position. The road section point cloud data collected by each radar sensor is compensated according to the relative displacement amount.

3. The method of claim 1 or 2, wherein, The step of incrementally updating the three-dimensional point cloud according to the relative displacement amount in the corresponding region in the cloud end point cloud full set comprises: The road section point cloud data collected by each radar sensor is spliced. The spliced data is subjected to motion object filtering by a preset motion detection algorithm. The corresponding region in the filtered road section point cloud data is incrementally updated according to the relative displacement amount.

4. A local dynamic map generation system characterized by, The system comprises: An acquisition unit is configured to obtain the relative displacement amount according to the feature point position extracted from the point cloud data collected by the radar sensor and the preset initial position, and to incrementally update the three-dimensional point cloud in the corresponding region in the cloud end point cloud full set according to the relative displacement amount. A semantic segmentation unit is configured to subject the road images collected by the camera at the first collection frequency to semantic segmentation to obtain static road semantic information. The acquisition unit is configured to subject the road images collected by the camera at the second collection frequency to target detection to obtain dynamic target information on the road. The acquisition unit is further configured to obtain the spatial coordinates of the semantic pixel points according to the configuration parameter information of the camera, the static road semantic information and the dynamic target information. A construction unit is configured to construct the local dynamic map according to the incrementally updated three-dimensional point cloud data and the spatial coordinates of the semantic pixel points. The semantic segmentation unit is specifically configured to perform semantic segmentation on road images collected by the camera at a first collection frequency by using a preset SAM model, to obtain static road semantic information. The acquisition unit is specifically configured to perform target detection on road images collected by the camera at a second collection frequency by using a preset YOLOv5 model, to obtain dynamic target information on the road.

5. The local dynamic map generation system according to claim 4, wherein the acquisition unit is specifically configured to accumulate the road segment point cloud data collected by each radar sensor according to a preset time; extract feature points from the accumulated point cloud data, and obtain a relative displacement amount according to the position of the extracted feature points and a preset initial feature position; and compensate the road segment point cloud data collected by each radar sensor according to the relative displacement amount.

6. The local dynamic map generation system according to claim 4 or 5, wherein the acquisition unit is specifically configured to splice the road segment point cloud data collected by each radar sensor; filter the spliced data by using a preset motion detection algorithm; and perform incremental updating of three-dimensional point cloud on a corresponding region in the filtered road segment point cloud data according to the relative displacement amount. ​ ​

Images