WO2021128297A1 - 三维点云地图构建方法、系统和设备 - Google Patents

三维点云地图构建方法、系统和设备 Download PDF

Info

Publication number
WO2021128297A1
WO2021128297A1 PCT/CN2019/129307 CN2019129307W WO2021128297A1 WO 2021128297 A1 WO2021128297 A1 WO 2021128297A1 CN 2019129307 W CN2019129307 W CN 2019129307W WO 2021128297 A1 WO2021128297 A1 WO 2021128297A1
Authority
WO
WIPO (PCT)
Prior art keywords
point cloud
data
pose data
dimensional point
cloud map
Prior art date
Application number
PCT/CN2019/129307
Other languages
English (en)
French (fr)
Inventor
钟阳
周游
朱振宇
Original Assignee
深圳市大疆创新科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市大疆创新科技有限公司 filed Critical 深圳市大疆创新科技有限公司
Priority to CN201980091861.6A priority Critical patent/CN113424232B/zh
Priority to PCT/CN2019/129307 priority patent/WO2021128297A1/zh
Publication of WO2021128297A1 publication Critical patent/WO2021128297A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects

Definitions

  • the present disclosure relates to the technical field of high-precision maps, and in particular to methods, systems and equipment for constructing a three-dimensional point cloud map.
  • High-precision electronic maps are also called high-resolution maps (HD Map, High Definition Map), which is a map dedicated to unmanned driving services.
  • HD Map High-resolution maps
  • Map construction is the first step of the high-precision map module, and its construction quality directly affects the accuracy of downstream high-precision map labeling and high-precision map positioning.
  • the previous map construction methods have low accuracy. Therefore, it is necessary to optimize the map construction process.
  • the embodiments of the present disclosure propose a method, system and device for constructing a three-dimensional point cloud map to solve the technical problem of low accuracy in related technologies.
  • a method for constructing a three-dimensional point cloud map includes: acquiring multiple frames of initial point cloud data collected by a lidar system mounted on a movable platform, and acquiring the laser
  • the radar system collects the pose data of multiple frames of the initial point cloud data; determines at least one loop constraint condition, and the loop constraint condition is based on the difference between the spatial poses of the two frames within a first preset range and is not continuous in time sequence
  • the pose data is determined; the pose data is processed according to the at least one loop constraint condition to obtain the optimized pose data of the initial point cloud data for each frame; according to the optimized pose data, based on Multiple frames of the initial point cloud data construct a three-dimensional point cloud map.
  • a three-dimensional point cloud map construction system includes: a processing device and a lidar system mounted on a movable platform; In collecting multiple frames of initial point cloud data; the processor is configured to determine at least one loopback constraint condition; according to the at least one loopback constraint condition, perform pose data when the lidar system collects multiple frames of the initial point cloud data Processing to obtain the optimized pose data of the initial point cloud data for each frame; according to the optimized pose data, a three-dimensional point cloud map is constructed based on multiple frames of the initial point cloud data; the loop constraint condition is based on two frames of spatial position The pose difference is determined by the pose data that is within the first preset range and is not continuous in time sequence.
  • a device for constructing a three-dimensional point cloud map includes a memory, a processor, and a computer program stored in the memory and running on the processor. The method described in any embodiment of the present disclosure is implemented when the program is executed.
  • a computer-readable storage medium is provided, and a number of computer instructions are stored on the readable storage medium, and when the computer instructions are executed, the steps of the method described in any of the embodiments are implemented.
  • Figure 1A is a schematic diagram of a conventional map of some embodiments.
  • Figure 1B is a schematic diagram of a high-precision map of some embodiments.
  • Fig. 2A is a schematic diagram of a navigation line of a conventional map according to some embodiments.
  • Fig. 2B is a schematic diagram of a navigation line of a high-precision map of some embodiments.
  • Fig. 3 is a flowchart of a method for constructing a three-dimensional point cloud map according to the present disclosure.
  • Fig. 4 is a schematic diagram showing the adjustment logic of loopback constraint conditions according to an embodiment of the present disclosure.
  • Fig. 5 is a schematic structural diagram of a system for constructing a three-dimensional point cloud map according to another embodiment of the present disclosure.
  • Fig. 6 is a block diagram of a device for constructing a three-dimensional point cloud map according to an embodiment of the present disclosure.
  • first, second, third, etc. may be used in this specification to describe various information, the information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other.
  • first information may also be referred to as second information, and similarly, the second information may also be referred to as first information.
  • word “if” as used herein can be interpreted as "when” or “when” or "in response to determination”.
  • FIGS. 1A and 1B are schematic diagrams of a traditional map and a high-precision map of some embodiments, respectively.
  • High-precision electronic maps are also called high-definition maps (HD Map, High Definition Map), which are very suitable for unmanned driving services.
  • HD Map High Definition Map
  • high-precision maps can provide not only road-level navigation information, but also lane-level navigation information.
  • 2A and 2B are schematic diagrams of navigation lines of traditional maps and high-precision maps of some embodiments, respectively.
  • the navigation information of the traditional map only includes information on navigation roads (shown in R1 to R4 in the figure) and information on road nodes (shown in N1 and N2 in the figure).
  • FIG. 2A the navigation information of the traditional map only includes information on navigation roads (shown in R1 to R4 in the figure) and information on road nodes (shown in N1 and N2 in the figure).
  • FIG. 1A and 1B are schematic diagrams of a traditional map and a high-precision map
  • the navigation information of the high-precision map includes not only the navigation road information, but also the lane information on each road (shown in Lane1 to Lane10 in the figure). It can be seen that high-precision maps are much higher than traditional navigation maps in terms of information richness and information accuracy. High-precision maps can provide accurate data that traditional maps cannot provide, thereby providing more accurate navigation guidance information for unmanned vehicles.
  • Map construction is the first step of the high-precision map module, and its construction quality directly affects the accuracy of downstream high-precision map labeling and high-precision map positioning.
  • a common map construction method can be: the vehicle is equipped with lidar, and the worker drives the vehicle in the area that needs to build a high-precision map, during which the lidar on the vehicle scans the surrounding environment; the lidar is collected after the driving is completed Scan the point cloud data of the surface, and process it to get the constructed map.
  • the point cloud data collected by lidar needs to be fused and superimposed in the later stage to form a high-precision map with sufficient point cloud density, it depends on the pose data provided by the navigation system on the vehicle for processing.
  • the accuracy and stability of the navigation system depends on the positioning signal. It usually has a higher accuracy in an open space, and in some more complicated scenes, such as an urban environment with high-rise buildings, or the signal is blocked In the tunnel, the positioning accuracy is so low that the pose data provided by the navigation system can sometimes not be directly used to construct a high-precision map.
  • the embodiments of the present disclosure provide a method, system and device for constructing a three-dimensional point cloud map.
  • the movable platform can be a device with mobile capabilities such as vehicles, drones, and mobile robots.
  • the target area is the area where a three-dimensional point cloud map needs to be built.
  • the area can be a road, a small area, or an indoor space, etc.
  • the initial point cloud data can be collected in the target area through the lidar system mounted on the movable platform .
  • the loop closure constraint refers to the pose constraint when the lidar system collects point cloud data at the same place at different times. In the following, a case where the movable platform is a vehicle is taken as an example for description.
  • the method for constructing a three-dimensional point cloud map of an embodiment of the present disclosure may include:
  • Step 301 Acquire multiple frames of initial point cloud data collected by a lidar system mounted on a movable platform, and acquire pose data when the lidar system collects multiple frames of the initial point cloud data;
  • Step 302 Determine at least one loopback constraint condition, the loopback constraint condition being determined based on the pose data whose spatial pose difference between two frames is within a first preset range and is not continuous in time sequence;
  • Step 303 Process multiple frames of the pose data according to the at least one loopback constraint condition to obtain optimized pose data of the initial point cloud data for each frame;
  • Step 304 Construct a three-dimensional point cloud map based on the multiple frames of the initial point cloud data according to the optimized pose data.
  • step 301 it is first necessary to obtain multi-frame initial point cloud data of the target area.
  • the vehicle can be driven to the target area, and the laser radar system mounted on the vehicle collects multiple frames of initial point cloud data in the target area.
  • the operation of collecting initial point cloud data can be performed twice or more, that is, the vehicle is repeatedly driven under the target area for many times, and each time the vehicle is driven, it is collected by the lidar system mounted on the vehicle One-time initial point cloud data.
  • the initial point cloud data can be collected at different time intervals each time. For example, in the first collection, collect the initial point cloud data under the target area A from 8:00 to 9:00, and collect the initial point cloud data under the target area A from 13:00 to 14:00 during the second collection. Cloud data, in the third collection, collect the initial point cloud data under the target area A from 17:00 to 18:00, and so on. Since there may be moving objects (for example, vehicles or pedestrians) in the target area, multiple collections are required to ensure a more comprehensive coverage of the target area; in addition, some target areas have a minimum speed limit, and the vehicle cannot drive too slowly. The number of point cloud data collected during the collection process may not be enough to build a three-dimensional point cloud map.
  • the pose data when the lidar system collects multiple frames of the initial point cloud data may be determined by the pose data of the vehicle equipped with the lidar system.
  • the vehicle is also equipped with a navigation system, and the pose data is obtained based on the navigation data of the navigation system.
  • the navigation system may be GNSS (Global Navigation Satellite System), inertial navigation system (INS, Inertial Navigation System), visual inertial navigation system (VINS, Visual-Inertial Navigation System), real-time dynamic (Real-Time) Kinematic, RTK) carrier phase differential positioning system (RTK system), etc.
  • GNSS can include GPS (Global Positioning System, global positioning system) navigation system, Galileo satellite positioning system, BeiDou Navigation Satellite System (BDS) or Navigation systems such as GLONASS.
  • the pose data may include position data and attitude data.
  • the position data is the three-dimensional coordinate position of the vehicle (latitude and longitude coordinates may be used), and the attitude data includes the roll angle, heading angle, and pitch angle of the vehicle.
  • the navigation system includes a first navigation system and a second navigation system; the pose data is based on the first navigation data output by the first navigation system and the second navigation system output by the second navigation system. Obtained by navigation data. Obtaining the pose data through the integrated navigation system including the first navigation system and the second navigation system can improve the positioning accuracy, and the constructed three-dimensional point cloud map has higher accuracy.
  • both the first navigation system and the second navigation system can be any type of navigation system, and the types of the first navigation system and the second navigation system can be the same or different.
  • the first navigation system may be an RTK system
  • the second navigation system may be an inertial navigation system
  • the pose data is generated based on fusion of RTK system data and inertial navigation system data.
  • different weights may be set for the first navigation system and the second navigation system, and the pose data may be a weighted average of the first navigation data and the second navigation data. It's worth getting.
  • the weight may be dynamically set according to the signal strength of the respective positioning signals of the first navigation system and the second navigation system. When the signal strengths of the positioning signals of the two navigation systems differ greatly, set a larger weight for the navigation system with stronger signal strength, and set a smaller weight for the navigation system with weaker signal strength; when the two navigation systems When the signal strength of the positioning signal is close, set the same or similar weights for the two navigation systems.
  • the pose data and the initial point cloud data need to be aligned to ensure
  • the initial point cloud data of each frame and the pose data when the initial point cloud data of the frame are collected have a one-to-one correspondence and matching.
  • the pose data can be aligned with the initial point cloud data based on the global time stamp of the navigation system and the global time stamp of the lidar system, so that the initial point cloud data of each frame is aligned with One frame of pose data corresponds to it.
  • the frequency of the lidar system can be in the tens of hertz, such as 5-20 Hz, that is, 5 to 20 frames of initial point cloud data are collected per second; the frequency of the navigation system can be higher than that of the lidar system
  • the frequency for example, when using an inertial navigation system and/or an RTK system can be in the hundreds of hertz. Therefore, a frame of pose data can be obtained by aligning each frame of initial point cloud data in time based on the global timestamp.
  • the frequency of the lidar system can also be higher than the frequency of the navigation system.
  • the initial point cloud data of a specific frame can be selected to be removed.
  • the interpolation method can also be used to increase the number of frames of the pose data to realize the time alignment of the initial point cloud data and the pose data.
  • the initial point cloud data collected by the lidar system may include moving objects such as vehicles and pedestrians, and the constructed three-dimensional point cloud map generally does not retain moving objects, therefore, moving objects are filtered out from the initial point cloud data.
  • the point cloud data corresponding to the object is removed after the point cloud data corresponding to the moving object is removed, and then the remaining point cloud data is used for the construction of a three-dimensional point cloud map.
  • the initial point cloud data collected on a road may not only include the initial point cloud data corresponding to moving objects such as pedestrians and vehicles, but also the buildings on both sides of the road, traffic signs around the road, road guardrails, etc.
  • the multiple frames of the initial point cloud data may include the initial point cloud data collected by the lidar system when the movable platform passes through the same place at least twice.
  • the two frames of initial point cloud data collected when the movable platform passes through the same place twice can be used to establish a loop constraint condition; when the vehicle passes through the same place three or more times, multiple loop constraint conditions can be established .
  • the vehicle when the vehicle is equipped with a lidar system for map data collection, it can pass through the same location twice or multiple times to increase the density of the point cloud for the final high-precision map construction at that location, and it can also provide more information at the same time. Loopback constraints.
  • the loopback constraint condition can also be set in a second preset based on the difference between the two frames of spatial pose It is determined by the pose data within the range and the opposite direction.
  • the pose data whose spatial pose difference is within the second preset range and facing the opposite direction generally refers to the pose data of the vehicle on the forward and reverse two-way roads corresponding to the same place, for example, the vehicle is on the two-way road 1
  • the pose data in which the difference between the two frames of spatial pose is within a second preset range and the orientation is opposite may be determined by target objects that can be detected in two-way lanes such as street signs, peripheral buildings, tunnel entrances, and the like.
  • Establishing loop constraints can reduce the cumulative error in the pose calculation process, improve the optimization accuracy of the pose data, and thereby improve the accuracy of the three-dimensional point cloud map.
  • the number of loopback constraints can be dynamically adjusted.
  • the number of loop constraint conditions can be adjusted according to the confidence of the pose data; for another example, the number of loop constraint conditions can be adjusted according to the three-dimensional size information of at least one target object in the three-dimensional point cloud map.
  • the confidence level of the pose data is used to characterize the credibility of the acquired pose data.
  • the confidence level of the pose data is related to the signal strength of the positioning signal used to locate the vehicle and the accuracy of the three-dimensional point cloud map. Therefore, the confidence level of the pose data can be determined according to any one of the strength of the positioning signal for positioning the vehicle and the accuracy of the three-dimensional point cloud map.
  • the signal strength of the positioning signal is strong, the confidence level of the pose data can be considered to be high; conversely, if the signal strength of the positioning signal is weak, the confidence level of the pose data can be considered to be low.
  • the signal strength of the GPS signal is often strong, the positioning accuracy is high, and the credibility of the acquired pose data is also high.
  • GPS signals are blocked by buildings, and the signal strength is often weak, so that the positioning error is large, and the obtained pose data is less reliable. .
  • the accuracy of the 3D point cloud map can be determined.
  • Determine the confidence level of the pose data the accuracy of the three-dimensional point cloud map can be determined by the deviation between the three-dimensional coordinates of a point in the three-dimensional point cloud map and the actual coordinates of the point. The smaller the deviation, the higher the accuracy; conversely, the larger the deviation, the lower the accuracy. If the accuracy of the three-dimensional point cloud map is high, the confidence of the pose data can be considered to be high; conversely, if the accuracy of the three-dimensional point cloud map is low, the confidence of the pose data can be considered low.
  • the accuracy of the three-dimensional point cloud map can be determined according to the following method: sampling the three-dimensional point cloud map within a certain range at intervals on the moving track of the movable platform; calculating the sampled three-dimensional point cloud map The height difference of each ground point; the accuracy of the three-dimensional point cloud map is determined according to the height difference. Further, if the height difference is greater than a preset height threshold, it is determined that the accuracy of the three-dimensional point cloud map is lower than the preset accuracy threshold. If the height difference is less than or equal to the height threshold, it is determined that the accuracy of the three-dimensional point cloud map is not lower than a preset accuracy threshold.
  • the points within a square range of 1 meter * 1 meter in the three-dimensional point cloud map can be sampled every 200 meters, and the height difference d of each ground point cloud within the square range can be calculated. It can be considered that this square range is on a plane. In theory, the ground points in this square range should be on the same level with a height difference of 0. However, if there are ground points with a height difference d greater than a certain threshold, it is considered to be within the square range.
  • the constructed three-dimensional point cloud map has low accuracy, and the pose data corresponding to the three-dimensional point cloud data within this square area needs to increase the number of loop matching (that is, increase the number of loop constraint conditions).
  • the number of loop constraints can be adjusted according to the confidence. If the confidence is less than the preset confidence threshold, the number of loop constraints can be increased. For example, increase the number of loopback constraints from 3 to 5. If the confidence is greater than or equal to the preset confidence threshold, the number of loop constraints can be reduced. For example, during data collection, the initial point cloud data may be repeatedly collected five times in the same target area, but only the initial point cloud data collected three times are used to establish the loopback constraint.
  • the target object may be a landmark object in the target area, such as traffic signs, traffic lights, and lamp posts.
  • the reference three-dimensional size information length, width, height
  • the difference between the three-dimensional size information and the reference three-dimensional size information of the target object may be calculated; and the number of the loop constraint conditions may be adjusted according to the difference.
  • the difference between the three-dimensional size information of the target object in the constructed three-dimensional point cloud map and the reference three-dimensional size information is greater than the preset size threshold, it can be considered that the constructed three-dimensional point cloud map has low accuracy, and the number of loop constraints needs to be increased.
  • a standard size of a triangular traffic sign is 70 cm in length. If the size of a triangular traffic sign in the three-dimensional point cloud map differs greatly from the above-mentioned standard size, the number of loop constraints needs to be increased.
  • Fig. 4 is a schematic diagram showing the adjustment logic of loopback constraint conditions according to an embodiment of the present disclosure.
  • the number adjustment strategy of the loopback constraint conditions in the embodiments of the present disclosure is to spend more computing resources and more optimization times in regions with poor accuracy, while performing two-way road detection to reduce accumulated errors.
  • the pose data may be processed according to the at least one loopback constraint condition to obtain optimized pose data of the initial point cloud data for each frame.
  • a graph-based optimization method may be used to solve the optimized pose data.
  • the graph is composed of nodes and edges.
  • SLAM Graph-based Simultaneous Localization And Mapping, synchronous positioning and mapping
  • the pose is a node or vertex, and the relationship between the poses constitutes the edge (edge).
  • edge the edge
  • the geometric relationship between time t+1 and time t constitutes an edge
  • a pose conversion matrix calculated by vision can also constitute an edge.
  • the processing the pose data according to the at least one loop constraint condition to obtain the optimized pose data of the initial point cloud data of each frame includes: taking the pose data as the vertex of the graph and using The loop constraint condition is used as the edge of the graph, and the optimization objective function of the graph (for example, the residual sum of squares is the smallest) is solved to obtain the optimized pose data. Further, the relative pose relationship between the loop constraint condition and the pose data of adjacent frames can also be used as the edges of the graph to solve the graph optimization.
  • all 3D point cloud data is traversed according to the time sequence, and the inter-frame registration in the time sequence can be used for the inter-frame registration.
  • NSS normal space sampling
  • ICP processing will be performed to obtain the two frames before and after.
  • the relative pose relationship of the pose data that is, the inter-frame constraint, is used as the edge of the graph.
  • the initial point cloud data of the previous frame P ⁇ p 1 ,p 2 ,...,p n ⁇ is collected in the first pose
  • the initial point cloud data of the next frame Q ⁇ q 1 ,q 2 ,...,q n ⁇
  • the coordinates of P and Q correspond to the coordinate system in the first pose and the coordinate system in the second pose, respectively.
  • the ith point in P and the ith point in Q correspond to the same point in three-dimensional space.
  • p 99 and q 99 correspond to the same point in three-dimensional space.
  • the formula for converting from the P coordinate system to the Q coordinate system is:
  • R and t are the rotation matrix and translation matrix from the first pose to the second pose, respectively. False matches due to noise and (p i and q i is not actually correspond to the same point in space, but wherein the two matching algorithm mistakenly believe that the same point) is present, the equation (1) is not always established, therefore, desirable to minimize
  • the objective function is:
  • n is the number of point cloud data per frame.
  • step 303 adds a new loop constraint when solving the optimized pose data, and according to the added loop constraint, step 303 is executed again to proceed. Optimize the solution of the pose data to obtain higher-precision optimized pose data. Repeat the above steps several times until the accuracy of the optimized pose data reaches the standard.
  • the loop detection error may be large, which makes the constructed three-dimensional point cloud map have a large error.
  • continuous multiple frames of initial point cloud data can be processed as one frame of initial point cloud data, that is, the initial point cloud data includes consecutive multiple frames of point cloud data that have been registered. Since the pose of each frame of initial point cloud data may be different, it is necessary to register the multiple frames of initial point cloud data first. The registration process can be performed according to the optimized pose data.
  • the initial point cloud data of a certain distance range (such as 100m) can be accumulated according to the optimized pose data, and then the loop registration of the point set can be performed to improve the accuracy of the loop registration.
  • all 3D point cloud data can be used to optimize the pose data according to a certain distance, and the time series can be accumulated, and the accumulated 3D point cloud data can be used as the input for registration, and the graph optimization can be performed in the above manner to further improve Accuracy of pose data.
  • the optimized pose data can be searched from a certain range of pose data; wherein, the search range of the optimized pose data is based on the signal strength of the positioning signal for positioning the movable platform and/or the Optimize the confidence level of the pose data and set it.
  • the search range of the optimized pose data is dynamically adjusted. For example, when the signal strength of the positioning signal is poor, the search range of the optimized pose data is expanded, and conversely, the search range of the optimized pose data can be reduced. When the confidence level is low, the search range of the optimized pose data can be expanded, and vice versa, the search range of the optimized pose data can be reduced. In actual operation, you can center on the pose data before optimization, and search for optimized pose data in the neighborhood of the pose data. When the signal strength of the positioning signal is poor, expand the radius of the neighborhood; when the signal strength of the positioning signal is strong, reduce the radius of the neighborhood.
  • the optimized pose data may be verified, and the three-dimensional point cloud map may be constructed based on the multiple frames of the initial point cloud data according to the optimized pose data that is successfully verified.
  • the trajectory of the initial point cloud data collected by the vehicle should be smooth. Therefore, the optimized pose data can be verified according to the smoothness of the optimized pose data.
  • the optimized pose data and the optimized pose data of the previous frame or the previous frames of the optimized pose data, and the optimized pose data of the next frame or the next few frames of the optimized pose data may be calculated The average value of the data. If the difference between the optimized pose data and the average is greater than the preset value, it means that the optimized pose data has undergone a relatively large change compared to the adjacent optimized pose data (ie , Not smooth), thereby determining that the optimized pose data verification fails.
  • the optimized pose data of the kth frame the average value of the optimized pose data of the kth frame, the optimized data of the k-1th frame, and the optimized pose data of the k+1th frame can be taken, and the optimized pose data of the kth frame Compare with this average. If the difference between the optimized pose data of the kth frame and the average value is greater than the preset value, it is determined that the verification of the optimized pose data fails.
  • the heading angle in the optimized pose data may be used to verify the optimized pose data. For example, for the optimized pose data of the kth frame, the heading angle of the optimized pose data of the kth frame and the heading angle of the optimized pose data of the previous frame (k-1th frame) and the heading angle of the next frame (th frame) are calculated. k+1 frame) the average value of the heading angle of the optimized pose data. If the difference between the heading angle of the optimized pose data in the k-th frame and the average value of the heading angle is greater than a certain threshold (for example, 1°), it is considered The verification of the optimized pose data failed. In this case, the number of loopback constraints can be increased.
  • a certain threshold for example, 1°
  • the verification of the optimized pose data is performed.
  • the optimized pose data that fails the verification has a lower confidence level, and the search range of the optimized pose data needs to be expanded, and/ Or increase the number of loopback constraints.
  • the method of the embodiment of the present disclosure can perform centimeter-level three-dimensional reconstruction of the target area, and obtain accurate map information.
  • map construction can also be completed smoothly.
  • GPU can be used to perform parallel processing on multiple frames of initial point cloud data, which has good parallelism and requires less time to build a map than traditional methods.
  • an embodiment of the present disclosure further provides a three-dimensional point cloud map construction system.
  • the three-dimensional point cloud map construction system includes:
  • Processing device 501 and lidar system 502 mounted on a movable platform;
  • the lidar system 502 is used to collect multiple frames of initial point cloud data
  • the processor 501 is configured to determine at least one loopback constraint condition; process multiple frames of the pose data according to the at least one loopback constraint condition to obtain optimized pose data of the initial point cloud data for each frame, and obtain each Frame the optimized pose data of the initial point cloud data; according to the optimized pose data, construct a three-dimensional point cloud map based on multiple frames of the initial point cloud data; the loop constraint condition is based on the difference between the two frames of spatial pose Determined by the pose data within the first preset range and not continuous in time sequence.
  • the three-dimensional point cloud map construction system further includes: a navigation system 503 mounted on the movable platform; the navigation system 503 is used to obtain the initial points collected by the lidar system for multiple frames Pose data in cloud data.
  • the navigation system 503 includes: a first navigation system and a second navigation system; the first navigation system is used to obtain the first navigation system when the lidar system collects multiple frames of the initial point cloud data. Pose data; the second navigation system is used to obtain second pose data when the lidar system collects multiple frames of the initial point cloud data; the processing device is also used to: based on the first navigation data And the second navigation data to obtain the pose data.
  • the processing device 501 is further configured to: align the pose data with the initial point cloud data.
  • the processing device 501 aligning the pose data with the initial point cloud data includes: comparing all the data based on the global time stamp of the navigation system and the global time stamp of the lidar system. The pose data is aligned with the initial point cloud data.
  • the multiple frames of the initial point cloud data include initial point cloud data collected by the lidar system when the movable platform passes through the same place at least twice.
  • the processing device 501 is further configured to: adjust the number of loopback constraint conditions.
  • the processing device 501 adjusts the number of loop constraint conditions, including: adjusting the number of loop constraint conditions according to the confidence of the pose data; and/or according to the three-dimensional point
  • the three-dimensional size information of at least one target object in the cloud map adjusts the number of the loop constraint conditions.
  • the confidence of the pose data is determined according to the strength of the positioning signal for positioning the movable platform and/or the accuracy of the three-dimensional point cloud map.
  • the processing device 501 is further configured to: sample a three-dimensional point cloud map within a certain range at intervals on the moving track of the movable platform; calculate each ground point in the sampled three-dimensional point cloud map The height difference; the accuracy of the three-dimensional point cloud map is determined according to the height difference.
  • the processing device 501 determines the accuracy of the three-dimensional point cloud map according to the height difference, including: if the height difference is greater than a preset height threshold, determining the accuracy of the three-dimensional point cloud map The degree of accuracy is lower than a preset accuracy threshold; and/or if the height difference is less than or equal to the height threshold, it is determined that the accuracy of the three-dimensional point cloud map is not lower than the preset accuracy threshold.
  • the processing device 501 adjusts the number of loop constraint conditions according to the confidence of the pose data, including: if the confidence is less than a preset confidence threshold, increasing the loop constraint And/or if the confidence is greater than or equal to a preset confidence threshold, reduce the number of loop constraints.
  • the processing device 501 adjusts the number of loop constraint conditions according to the three-dimensional size information of at least one target object in the three-dimensional point cloud map, including: calculating the three-dimensional size information and the three-dimensional size information of the target object. Refer to the difference between the three-dimensional size information; adjust the number of the loop constraint conditions according to the difference.
  • the loopback constraint condition is further determined based on the pose data of which the difference between the two frames of spatial pose is within a second preset range and the orientation is opposite.
  • the initial point cloud data includes registered continuous multi-frame point cloud data.
  • the processing device 501 is further configured to filter out the point cloud data corresponding to the moving object from the initial point cloud data.
  • the processing device 501 processes the pose data according to the at least one loop constraint condition to obtain optimized pose data of the initial point cloud data for each frame, including: The data is taken as the vertices of the graph, and the loop constraint conditions are taken as the edges of the graph, and the optimization objective function of the graph is solved to obtain the optimized pose data.
  • the processing device 501 uses the pose data as the vertices of the graph and uses the loop constraint conditions as the edges of the graph to solve the optimization objective function of the graph to obtain the optimized pose data, including : Taking the pose data as the vertices of the graph, and taking the loop constraint conditions and the relative pose relationships between the pose data of adjacent frames as the edges of the graph, solving the optimization objective function of the graph to obtain the optimization position Pose data.
  • the optimized pose data is retrieved from the pose data within a certain range; wherein, the search range of the optimized pose data is based on the strength of the positioning signal for positioning the movable platform and /Or the confidence level of the optimized pose data is set.
  • the processing device 501 constructs a three-dimensional point cloud map based on the multiple frames of the initial point cloud data according to the optimized pose data, including: verifying the optimized pose data; The pose data is successfully optimized, and a three-dimensional point cloud map is constructed based on the initial point cloud data of multiple frames.
  • the processing device 501 checks the optimized pose data, including: calculating the average value of the optimized pose data and the optimized pose data of several adjacent frames; calculating the optimized pose data The difference between the pose data and the average value; the optimized pose data is checked according to the difference.
  • the processing device 501 checks the optimized pose data according to the difference, including: if the difference is greater than a preset value, determining that the optimized pose data is successfully checked; And/or if the difference is less than or equal to the preset value, it is determined that the verification of the optimized pose data has failed.
  • the processing device 501 constructs a three-dimensional point cloud map based on the multiple frames of the initial point cloud data according to the optimized pose data, including: comparing the multiple frames of the initial points according to the optimized pose data The cloud data is registered; the three-dimensional point cloud map is constructed according to the registered multiple frames of the initial point cloud data.
  • the present disclosure also provides a three-dimensional point cloud map building device, the three-dimensional point cloud map building device includes a memory, a processor, and a computer program stored in the memory and running on the processor.
  • the processor executes the program, the steps of the method described in any one of the embodiments are implemented.
  • FIG. 6 shows a schematic diagram of the hardware structure of a more specific three-dimensional point cloud map construction device provided by an embodiment of the present disclosure.
  • the device may include: a processor 601, a memory 602, an input/output interface 603, and a communication interface 604 And bus 605.
  • the processor 601, the memory 602, the input/output interface 603, and the communication interface 604 realize the communication connection between each other in the device through the bus 605.
  • the processor 601 can be implemented by a general CPU (Central Processing Unit, central processing unit), microprocessor, application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits for execution related Program to implement the technical solutions provided in the embodiments of this specification.
  • CPU Central Processing Unit
  • microprocessor microprocessor
  • ASIC Application Specific Integrated Circuit
  • the memory 602 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory, random access memory), static storage device, dynamic storage device, etc.
  • the memory 602 may store an operating system and other application programs.
  • related program codes are stored in the memory 602 and called and executed by the processor 602.
  • the input/output interface 603 is used to connect an input/output module to realize information input and output.
  • the input/output/module can be configured in the device as a component (not shown in the figure), or can be connected to the device to provide corresponding functions.
  • the input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and an output device may include a display, a speaker, a vibrator, an indicator light, and the like.
  • the communication interface 604 is used to connect a communication module (not shown in the figure) to realize the communication interaction between the device and other devices.
  • the communication module can realize communication through wired means (such as USB, network cable, etc.), or through wireless means (such as mobile network, WIFI, Bluetooth, etc.).
  • the bus 605 includes a path for transmitting information between various components of the device (for example, the processor 601, the memory 602, the input/output interface 603, and the communication interface 604).
  • the device may also include the equipment necessary for normal operation. Other components.
  • the above-mentioned device may also include only the components necessary to implement the solutions of the embodiments of the present specification, and not necessarily include all the components shown in the figures.
  • the present disclosure also provides a computer-readable storage medium having a number of computer instructions stored on the readable storage medium, and when the computer instructions are executed, the steps of the method described in any one of the embodiments are implemented.
  • Computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology.
  • the information can be computer-readable instructions, data structures, program modules, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, CD-ROM, digital versatile disc (DVD) or other optical storage, Magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices. According to the definition in this article, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.
  • the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments.
  • the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment.
  • the device embodiments described above are only illustrative, and the modules described as separate components may or may not be physically separated.
  • the functions of the modules can be combined in the same way when implementing the solutions of the embodiments of this specification Or multiple software and/or hardware implementations. Some or all of the modules can also be selected according to actual needs to achieve the objectives of the solutions of the embodiments. Those of ordinary skill in the art can understand and implement it without creative work.

Abstract

一种三维点云地图构建方法、系统和设备，所述方法包括：获取搭载于可移动平台上的激光雷达系统采集的多帧初始点云数据，并获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据（301）；确定至少一个回环约束条件，所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定（302）；根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据（303）；根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图（304）。

Description

三维点云地图构建方法、系统和设备 技术领域
本公开涉及高精度地图技术领域，具体而言，涉及三维点云地图构建方法、系统和设备。
背景技术
高精度电子地图也称为高分辨率地图(HD Map，High Definition Map)，是一种专门为无人驾驶服务的地图。与传统导航地图不同的是，高精度地图除了能提供的道路(Road)级别的导航信息外，还能够提供车道(Lane)级别的导航信息。无论是在信息的丰富度还是信息的精度方面，都是远远高于传统导航地图的。地图构建作为高精度地图模块的先行步骤，其构建质量直接影响到下游的高精度地图标注和高精度地图定位的精度，然而，以往的地图构建方式精确度较低。因此，有必要对地图构建过程进行优化。
发明内容
有鉴于此，本公开的实施例提出了三维点云地图构建方法、系统和设备，以解决相关技术中精确度较低的技术问题。
根据本公开实施例的第一方面，提出一种三维点云地图构建方法，所述方法包括：获取搭载于可移动平台上的激光雷达系统采集的多帧初始点云数据，并获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据；确定至少一个回环约束条件，所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定；根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据；根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。
根据本公开实施例的第二方面，提出一种三维点云地图构建系统，所述三维点云地图构建系统包括：处理装置和搭载于可移动平台上的激光雷达系统；所述激光雷达系统用于采集多帧初始点云数据；所述处理器用于确定至少一个回环约束条件；根 据所述至少一个回环约束条件对所述激光雷达系统采集多帧所述初始点云数据时的位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据；根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图；所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定。
根据本公开实施例的第三方面，提出一种三维点云地图构建设备，所述噪点过滤设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现本公开任一实施例所述的方法。
根据本公开实施例的第四方面，提出一种计算机可读存储介质，所述可读存储介质上存储有若干计算机指令，所述计算机指令被执行时实现任一实施例所述方法的步骤。
应用本说明书实施例方案，获取多帧初始点云数据以及采集多帧所述初始点云数据时的位姿数据，建立回环约束条件以对所述位姿数据进行优化，得到每帧所述初始点云数据的优化位姿数据，再根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，能够对场景进行厘米级的三维重建，获取到精准的三维点云地图，提高了地图构建的精确度。
附图说明
为了更清楚地说明本公开实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本公开的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1A是一些实施例的传统地图的示意图。
图1B是一些实施例的高精度地图的示意图。
图2A是一些实施例的传统地图的导航线的示意图。
图2B是一些实施例的高精度地图的导航线的示意图。
图3是根据本公开的三维点云地图构建方法的流程图。
图4是根据本公开的实施例示出的回环约束条件的调整逻辑的示意图。
图5是根据本公开的另一实施例示出的三维点云地图构建系统的结构示意图。
图6是根据本公开的实施例示出的三维点云地图构建设备的框图。
具体实施方式
这里将详细地对示例性实施例进行说明，其示例表示在附图中。下面的描述涉及附图时，除非另有表示，不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本说明书相一致的所有实施方式。相反，它们仅是与如所附权利要求书中所详述的、本说明书的一些方面相一致的装置和方法的例子。
在本说明书使用的术语是仅仅出于描述特定实施例的目的，而非旨在限制本说明书。在本说明书和所附权利要求书中所使用的单数形式的“一种”、“所述”和“该”也旨在包括多数形式，除非上下文清楚地表示其他含义。还应当理解，本文中使用的术语“和/或”是指并包含一个或多个相关联的列出项目的任何或所有可能组合。
应当理解，尽管在本说明书可能采用术语第一、第二、第三等来描述各种信息，但这些信息不应限于这些术语。这些术语仅用来将同一类型的信息彼此区分开。例如，在不脱离本说明书范围的情况下，第一信息也可以被称为第二信息，类似地，第二信息也可以被称为第一信息。取决于语境，如在此所使用的词语“如果”可以被解释成为“在……时”或“当……时”或“响应于确定”。
图1A和图1B分别是一些实施例的传统地图和高精度地图的示意图。高精度电子地图也称为高分辨率地图(HD Map，High Definition Map)，是一种非常适于为无人驾驶服务的地图。与传统导航地图不同的是，高精度地图除了能提供的道路(Road)级别的导航信息外，还能够提供车道(Lane)级别的导航信息。图2A和图2B分别是一些实施例的传统地图和高精度地图的导航线的示意图。如图2A所示，传统地图的导航信息中仅有导航道路(图中R1至R4所示)的信息以及道路节点(图中N1和N2所示)的信息。如图2B所示，高精度地图的导航信息中不仅包括导航道路的信息，还包括各道路上的车道信息(图中Lane1至Lane10所示)。可以看出，无论是在信息的丰富度还是信息的精度方面，高精度地图都远高于传统导航地图。高精度地图可以提供传统地图提供不了的精确数据，从而为无人驾驶车辆提供更为准确的导航指引信息。
地图构建作为高精度地图模块的先行步骤，其构建质量直接影响到下游的高精度地图标注和高精度地图定位的精度。一种常见的地图构建方法可以是：车辆上搭载 有激光雷达，工作员驾驶车辆在需要构建高精度地图的区域内行驶，期间车辆上的激光雷达对周围环境进行扫描；行驶完成之后采集激光雷达的扫面的点云数据，并进行处理之后得到构建的地图。这种方法在构建高精度地图时，由于后期需要对激光雷达采集的点云数据进行融合叠加以形成点云密度足够的高精度地图，因此依赖于车辆上导航系统提供的位姿数据才能进行处理，但导航系统的精确度与稳定性又依赖于定位信号，在空旷的场地通常有较高的精确度，而在一些较为复杂的场景下，例如，高楼林立的城市环境，或是信号被遮挡的隧道中，定位精确度较低，以至于导航系统提供的位姿数据有时不能直接用于构建高精度地图。
同时，每帧地图数据中多达数万个三维点，每秒扫描得到的地图数据多大几十帧，数据量巨大，如果直接处理耗时很长，仅一段数公里的路段，构建高精度地图就需要耗时1周左右。
综上所述，目前构建高精度地图时存在地图构建精确度低，以及地图构建效率低的问题。基于此，本公开实施例提供一种三维点云地图构建方法、系统和设备。在本公开实施例中。可移动平台可以是车辆、无人机、可移动机器人等具有移动能力的设备。目标区域即需要构建三维点云地图的区域，该区域可以是一条道路、一个小区，或者是一块室内空间等，初始点云数据可以通过可移动平台上搭载的激光雷达系统在目标区域中进行采集。回环(loop closure)约束条件是指激光雷达系统不同时刻在同一地点采集点云数据时的位姿约束条件。下面以可移动平台为车辆的情况为例进行说明。
如图3所示，本公开实施例的三维点云地图构建方法可包括：
步骤301：获取搭载于可移动平台上的激光雷达系统采集的多帧初始点云数据，并获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据；
步骤302：确定至少一个回环约束条件，所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定；
步骤303：根据所述至少一个回环约束条件对多帧所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据；
步骤304：根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。
在步骤301中，首先需要获取目标区域的多帧初始点云数据。可以将车辆行驶 至所述目标区域，并由搭载于车辆上的激光雷达系统采集所述目标区域中的多帧初始点云数据。对于同一目标区域，可以执行两次或者两次以上采集初始点云数据的操作，即，将车辆在该目标区域下反复行驶多次，每行驶一次，就由搭载在车辆上的激光雷达系统采集一次初始点云数据。
进一步地，每一次可以在不同的时段采集初始点云数据。例如，第一次采集时，在8:00-9:00在目标区域A下采集初始点云数据，第二次次采集时，在13:00-14:00在目标区域A下采集初始点云数据，第三次采集时，在17:00-18:00在目标区域A下采集初始点云数据，等等。由于目标区域中可能存在移动物体(例如，车辆或者行人)，需要通过多次采集才能保证对目标区域进行较为全面的覆盖；另外，有些目标区域存在最低限速，车辆无法行驶得太慢，一次采集过程中采集到的点云数据的数量可能不足以构建三维点云地图，需要通过多次采集来保证细节，保证单位面积内点数足够多。而在不同时段采集初始点云数据，不仅能够增加点云密度，还能够减少移动物体的干扰、不同光线的干扰以及遮挡干扰等。
所述激光雷达系统采集多帧所述初始点云数据时的位姿数据可以通过搭载所述激光雷达系统的车辆的位姿数据来确定。在一些实施例中，车辆上还搭载有导航系统，所述位姿数据基于所述导航系统的导航数据获取。所述导航系统可以是GNSS(Global Navigation Satellite System，全球卫星导航系统)、惯性导航系统(INS，Inertial Navigation System)、视觉惯性导航系统(VINS，Visual-Inertial Navigation System)、实时动态(Real-Time Kinematic，RTK)载波相位差分定位系统(RTK系统)等，其中GNSS可以包括GPS(Global Positioning System，全球定位系统)导航系统、伽利略卫星定位系统、北斗卫星定位系统(BeiDou Navigation Satellite System，BDS)或者GLONASS等导航系统。所述位姿数据可以包括位置数据和姿态数据，所述位置数据即车辆的三维坐标位置(可以采用经纬高坐标)，所述姿态数据包括车辆的翻滚角、航向角和俯仰角。
在一些实施例中，所述导航系统包括第一导航系统和第二导航系统；所述位姿数据基于所述第一导航系统输出的第一导航数据和所述第二导航系统输出的第二导航数据而获取。通过包括第一导航系统和第二导航系统的组合导航系统来获取所述位姿数据，能够提高定位精确度，构建出来的三维点云地图的精确度更高。其中，第一导航系统和第二导航系统均可以是任意类型的导航系统，所述第一导航系统和第二导航系统的类型可以相同，也可以不同。例如，所述第一导航系统可以是RTK系统，所述 第二导航系统可以是惯性导航系统，所述位姿数据基于RTK系统数据及惯性导航系统数据融合生成。
在一些实施例中，可以分别为所述第一导航系统和所述第二导航系统设置不同的权重，所述位姿数据可以由所述第一导航数据和所述第二导航数据的加权平均值得到。其中，所述权重可以根据所述第一导航系统和所述第二导航系统各自的定位信号的信号强度动态设置。当两个导航系统的定位信号的信号强度差异较大时，为信号强度较强的导航系统设置较大的权重，并为信号强度较弱的导航系统设置较小的权重；当两个导航系统的定位信号的信号强度接近时，为两个导航系统设置相同或相近的权重。
由于初始点云数据和位姿数据是两套不同系统采集的数据，在采集到初始点云数据和位姿数据之后，还需要对所述位姿数据与所述初始点云数据进行对齐，确保每帧初始点云数据以及采集该帧初始点云数据时的位姿数据一一对应、匹配。具体来说，可以基于所述导航系统的全局时间戳和所述激光雷达系统的全局时间戳对所述位姿数据与所述初始点云数据进行对齐，使每一帧初始点云数据均与一帧位姿数据相对应。在本发明的一些实施例中，激光雷达系统的频率可以在数十赫兹，例如5～20Hz，即每秒钟采集5至20帧初始点云数据；导航系统的频率可以高于激光雷达系统的频率，例如使用惯性导航系统和/或RTK系统时可以在数百赫兹。因此，可以基于全局时间戳为每一帧初始点云数据在时间上对齐得到一帧位姿数据。在本发明的另一些实施例中，激光雷达系统的频率也可以高于导航系统的频率，此时由于位姿数据要少于初始点云数据，因此可以选择去掉特定帧的初始点云数据，也可以采用插值的方法提高位姿数据的帧数以实现初始点云数据与位姿数据在时间上对齐。
由于激光雷达系统采集到的初始点云数据中可能包括车辆、行人等移动物体，而构建的三维点云地图上一般是不保留移动物体的，因此，从所述初始点云数据中滤除移动物体对应的点云数据，在去除移动物体对应的点云数据之后，再将剩余的点云数据用于三维点云地图构建。例如，在一条道路上采集到的初始点云数据既可能包括行人、车辆等移动物体对应的初始点云数据，又可能包括道路两边的建筑物、道路周围的交通指示牌、道路护栏等对应的初始点云数据，则需要将行人、车辆等移动物体对应的初始点云数据过滤掉，仅保留建筑物、交通指示牌、道路护栏等对应的初始点云数据。
在步骤302中，多帧所述初始点云数据可包括所述可移动平台至少两次通过同 一地点时，所述激光雷达系统采集到的初始点云数据。所述可移动平台两次通过同一地点时采集到的两帧初始点云数据可用于建立一个回环约束条件；当所述车辆三次或者更多地通过同一地点时，则可以建立多个回环约束条件。具体的，在所述车辆搭载激光雷达系统进行地图数据采集时，可以两次或者多次地通过同一地点以提高该地点最后构建高精度地图的点云的密集程度，也可以同时提供更多的回环约束条件。
除了采用所述可移动平台两次通过同一地点时采集到的两帧初始点云数据来建立回环约束条件之外，所述回环约束条件还可以基于两帧空间位姿之差在第二预设范围内、且朝向相反的所述位姿数据而确定。空间位姿之差在第二预设范围内、且朝向相反的所述位姿数据，一般是指车辆在同一地点对应的正反双向道路上的位姿数据，例如，车辆在双向道路1上由南向北行驶，经过某位置时的位姿数据，以及车辆在所述双向道路1上由北向南行驶，再次经过该位置时的位姿数据。在高速公路等一些场景下，道路中间有绿化带等物体隔开，并不能直接确定是否双向道路，因此，需要根据采集初始点云数据时的位置和朝向可以判断当前采集初始点云数据的地点与之前采集初始点云数据的地点是否正反双向道路。可以通过路牌、外围楼房，隧道口等双向车道均可检测到的目标对象来进行确定所述两帧空间位姿之差在第二预设范围内、且朝向相反的所述位姿数据。
建立回环约束条件，能够减少位姿计算过程中的累积误差，提高位姿数据的优化精确度，从而提高三维点云地图的精确度。
在一些实施例中，所述回环约束条件的数量可动态调整。例如，可以根据所述位姿数据的置信度调整所述回环约束条件的数量；又例如，可以根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量。
在根据所述位姿数据的置信度调整所述回环约束条件的数量的实施例中，所述位姿数据的置信度用于表征获取到的位姿数据的可信程度。所述位姿数据的置信度与用于对车辆进行定位的定位信号的信号强度，以及三维点云地图的精确度相关。因此，可以根据对车辆进行定位的定位信号的强度，以及所述三维点云地图的精确度中的任意一者确定所述位姿数据的置信度。
一方面，如果定位信号的信号强度较强，可认为所述位姿数据的置信度较高；反之，如果定位信号的信号强度较弱，可认为所述位姿数据的置信度较低。例如，在较为空旷的区域，GPS信号的信号强度常常较强，定位精确度较高，获取到的位姿数据的可信程度也较高。而在高楼林立的城市中，或者是在高架桥、隧道等区域，GPS 信号受到建筑物的遮挡，信号强度常常较弱，以至于定位误差较大，获取到的位姿数据的可信程度较低。
另一方面，如果获取到的位姿数据精确度较高，那么，通常根据该位姿数据建立的三维点云地图的精确度也会较高，因此，可以根据三维点云地图的精确度来确定位姿数据的置信度。其中，三维点云地图的精确度可以通过三维点云地图中的点的三维坐标与该点的实际坐标之间的偏差来确定。偏差越小，精确度越高；反之，偏差越大，精确度越低。如果三维点云地图的精确度较高，可认为所述位姿数据的置信度较高；反之，如果三维点云地图的精确度较低，可认为所述位姿数据的置信度较低。
其中，所述三维点云地图的精确度可以根据如下方式来确定：在所述可移动平台的移动轨迹上每隔一段距离抽样一定范围内的三维点云地图；计算抽样的三维点云地图中各个地面点的高度差；根据所述高度差确定所述三维点云地图的精确度。进一步地，若所述高度差大于预设的高度阈值，判定所述三维点云地图的精确度低于预设的精确度阈值。若所述高度差小于或等于所述高度阈值，判定所述三维点云地图的精确度不低于预设的精确度阈值。
在实际应用场景中，根据车辆的移动轨迹，可以每隔200米抽样三维点云地图中1米*1米的正方形范围内的点，计算该正方形范围内的各个地面点云的高度差d，可认为这个正方形范围在一个平面上，理论上这个正方形范围内的各个地面点应该在同一水平面上，高度差为0，但如果存在高度差d大于一定阈值的地面点，则认为该正方形范围内构建的三维点云地图精度较低，在这个正方形范围内的三维点云数据对应的位姿数据都需要增加回环匹配次数(即，增加回环约束条件的数量)。
在确定位姿数据的置信度之后，可以根据置信度调整回环约束条件的数量。如果所述置信度小于预设的置信度阈值，可以增加所述回环约束条件的数量。例如，将回环约束条件的数量由3个增加为5个。如果所述置信度大于或等于预设的置信度阈值，可以减少所述回环约束条件的数量。例如，在数据采集时，可能在同一目标区域中重复采集了五次初始点云数据，但是，仅采用其中三次采集的初始点云数据来建立回环约束条件。
在根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量的实施例中，目标对象可以是目标区域中的标志性物体，比如交通标志、红绿灯、灯柱等，根据国标可以知道这些目标对象的参考三维尺寸信息(长度，宽度，高度)。可以计算所述三维尺寸信息与所述目标对象的参考三维尺寸信息之间 的差值；根据所述差值调整所述回环约束条件的数量。如果构建的三维点云地图中的目标对象的三维尺寸信息与参考三维尺寸信息的差值大于预设的尺寸阈值，可以认为构建的三维点云地图精度较低，需要增加回环约束条件的数量。例如，根据国标要求，三角形交通标志牌的一种标准尺寸是边长为70厘米。如果三维点云地图中某个三角形交通标志牌的尺寸与上述标准尺寸相差较大，则需要增加回环约束条件的数量。
图4是根据本公开的实施例示出的回环约束条件的调整逻辑的示意图。综上，本公开实施例的回环约束条件的数量调整策略是，在精确度较差的区域下花费更多的计算资源以及更多的优化次数，同时进行双向道路检测，减少累计误差。
在步骤303中，可以根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据。
在一些实施例中，可以采用图优化(Graph-based optimization)的方式来求解优化位姿数据。图由节点和边构成，在图优化SLAM(Graph-based Simultaneous Localization And Mapping，同步定位与建图)中，位姿是一个节点(node)或顶点(vertex)，位姿之间的关系构成边(edge)。具体而言，比如t+1时刻和t时刻之间的几何关系构成边，或者由视觉计算出来的位姿转换矩阵也可以构成边。一旦图构建完成了，就要调整位姿去尽量满足这些边构成的约束。
因此，所述根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据，包括：以所述位姿数据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数(比如，残差平方和最小)，以获取所述优化位姿数据。进一步地，还可以将回环约束条件与相邻帧位姿数据之间的相对位姿关系共同作为图的边来求解图优化。
图优化的具体步骤如下：
首先，将导航系统获取的位置(粗略的位姿数据)作为图的顶点。
然后，按照时序对所有三维点云数据进行遍历，做时序上的帧间配准，帧间配准可以使用NSS(normal space sampling，正态空间采样)，然后进行ICP处理，将获取前后两帧位姿数据的相对位姿关系，即帧间约束，作为图的边。
假设在第一位姿下采集到前一帧初始点云数据P＝{p 1,p 2,…,p n}，在第二位姿下采集到的后一帧初始点云数据Q＝{q 1,q 2,…,q n}，P和Q的坐标分别对应第一位姿下的坐标系和第二位姿下的坐标系。假设P中的第i个点和Q中的第i个点在三维空间中对 应同一个点，例如，p 99和q 99在三维空间中对应同一个点。在没有误差的情况下，从P坐标系转换到Q坐标系的公式为：
q i＝Rp i+t   (1)；
其中，R和t分别为第一位姿到第二位姿的旋转矩阵和平移矩阵。由于噪声及错误匹配(p i和q i其实不对应空间中的同一点，但特征匹配算法错误地认为二者是同一点)的存在，公式(1)不总是成立，因此，需要最小化的目标函数为：
Figure PCTCN2019129307-appb-000001
其中，n为每帧点云数据的数量。通过公式(1)和公式(2)即可获取用于表征前后两帧位姿数据的相对位姿关系的参数R和t。
再次遍历初始点云数据，将空间位置临近，但是时序上不连续的多帧所述初始点云数据(即第二次或是第三次开过的同样的位置)使用ICP估计这些帧的位姿数据之间的相对位姿关系，并加入图优化作为回环约束(同样也是作为图的边)。
最后，根据上述步骤中建立的图顶点和图的边来求解图的优化目标函数，获取所述优化位姿数据。
若在步骤302中根据任一条件判定为需要增加回环约束条件的数量，则步骤303在优化位姿数据求解时增加新的回环约束条件，并根据增加的回环约束条件，再次执行步骤303，进行优化位姿数据的求解，以此得到更高精度的优化位姿数据。重复上述步骤若干次，直到优化位姿数据的精确度达标。
在一般情况下，通过回环检测能够获得比较好的性能。但是，在高速路段等地图特征较少的区域，回环检测的误差可能较大，使得构建出来的三维点云地图的误差较大。为了解决该问题，可以将连续的多帧初始点云数据作为一帧初始点云数据进行处理，即，所述初始点云数据包括经过配准的连续多帧点云数据。由于采集每帧初始点云数据时的位姿可能不同，因此需要先对这多帧初始点云数据进行配准。配准过程可根据所述优化位姿数据来执行。
可以根据优化后的位姿数据累计一定距离范围(比如100m)的初始点云数据，再进行点集的回环配准，以此提高回环配准的准确性。具体来说，可以将所有的三维点云数据，按照一定的距离使用优化位姿数据，进行时序的累计，将累计的三维点云数据作为配准的输入，按照上述方式执行图优化，进一步提高位姿数据准确性。
所述优化位姿数据可以从一定范围内的位姿数据中查找得到；其中，所述优化位姿数据的查找范围根据对所述可移动平台进行定位的定位信号的信号强度和/或所述优化位姿数据的置信度而设置。在求解图优化的过程中，动态调节优化位姿数据的查找范围，比如在定位信号的信号强度差的时候，扩大优化位姿数据的查找范围，反之可以减小优化位姿数据的查找范围。在置信度低的时候，扩大优化位姿数据的查找范围，反之可以减小优化位姿数据的查找范围。实际操作时，可以以优化前的位姿数据为中心，在该位姿数据的邻域内查找优化位姿数据。在定位信号的信号强度差时，扩大邻域半径；在定位信号的信号强度强时，减小邻域半径。
在步骤304中，在构建三维点云地图之后，可以校验优化位姿数据，并根据校验成功的优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。车辆采集初始点云数据的轨迹应当具有平滑性，因此，可以根据所述优化位姿数据的平滑性来对所述优化位姿数据进行校验。
具体而言，可以计算所述优化位姿数据与所述优化位姿数据的前一帧或者前若干帧优化位姿数据，以及所述优化位姿数据的后一帧或者后若干帧优化位姿数据的平均值，如果所述优化位姿数据与所述平均值之间的差值大于预设值，则说明所述优化位姿数据较相邻优化位姿数据发生了比较大的变化(即，不平滑)，从而判定所述优化位姿数据校验失败。对于第k帧优化位姿数据，可以取该第k帧优化位姿数据与第k-1帧优化数据、第k+1帧优化位姿数据的平均值，并将第k帧优化位姿数据与该平均值进行比较。如果第k帧优化位姿数据与该平均值之间的差值大于预设值，则判定所述优化位姿数据校验失败。
上述实施例可以采用优化位姿数据中的航向角进行优化位姿数据的校验。例如，对第k帧所述优化位姿数据，计算第k帧所述优化位姿数据的航向角与前一帧(第k-1帧)优化位姿数据的航向角和后一帧(第k+1帧)优化位姿数据的航向角的平均值，如果第k帧所述优化位姿数据的航向角与所述航向角的平均值之差大于一定阈值(比如1°)，则认为所述优化位姿数据校验失败。在这种情况下，可以增加回环约束条件的数量。
一般在获取第一次优化的优化位姿数据之后，进行优化位姿数据的校验，校验失败的优化位姿数据即置信度较低，则需要扩大优化位姿数据的查找范围，和/或增加回环约束条件的数量。
本公开实施例的方法能够对目标区域进行厘米级的三维重建，获取到精准的地 图信息。在传统的导航定位方式受限的区域(如隧道里，高架桥下)，通过调整回环约束条件的数量，也能顺利地完成地图的构建。并且，可以使用GPU对多帧初始点云数据进行并行处理，具有良好的并行性，构建地图的所需时间较传统方法更少。
在一些实施例中，如图5所示，本公开实施例还提供一种三维点云地图构建系统，所述三维点云地图构建系统包括：
处理装置501和搭载于可移动平台上的激光雷达系统502；
所述激光雷达系统502用于采集多帧初始点云数据；
所述处理器501用于确定至少一个回环约束条件；根据所述至少一个回环约束条件对多帧所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据，得到每帧所述初始点云数据的优化位姿数据；根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图；所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定。
在一些实施例中，所述三维点云地图构建系统还包括：搭载于所述可移动平台上的导航系统503；所述导航系统503用于获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据。
在一些实施例中，所述导航系统503包括：第一导航系统和第二导航系统；所述第一导航系统用于获取所述激光雷达系统采集多帧所述初始点云数据时的第一位姿数据；所述第二导航系统用于获取所述激光雷达系统采集多帧所述初始点云数据时的第二位姿数据；所述处理装置还用于：基于所述第一导航数据和所述第二导航数据获取所述位姿数据。
在一些实施例中，所述处理装置501还用于：对所述位姿数据与所述初始点云数据进行对齐。
在一些实施例中，所述处理装置501对所述位姿数据与所述初始点云数据进行对齐，包括：基于所述导航系统的全局时间戳和所述激光雷达系统的全局时间戳对所述位姿数据与所述初始点云数据进行对齐。
在一些实施例中，多帧所述初始点云数据包括所述可移动平台至少两次通过同一地点时，所述激光雷达系统采集到的初始点云数据。
在一些实施例中，所述处理装置501还用于：对所述回环约束条件的数量进行 调整。
在一些实施例中，所述处理装置501对所述回环约束条件的数量进行调整，包括：根据所述位姿数据的置信度调整所述回环约束条件的数量；和/或根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量。
在一些实施例中，所述位姿数据的置信度根据对所述可移动平台进行定位的定位信号的强度和/或所述三维点云地图的精确度确定。
在一些实施例中，所述处理装置501还用于：在所述可移动平台的移动轨迹上每隔一段距离抽样一定范围内的三维点云地图；计算抽样的三维点云地图中各个地面点的高度差；根据所述高度差确定所述三维点云地图的精确度。
在一些实施例中，所述处理装置501根据所述高度差确定所述三维点云地图的精确度，包括：若所述高度差大于预设的高度阈值，判定所述三维点云地图的精确度低于预设的精确度阈值；和/或若所述高度差小于或等于所述高度阈值，判定所述三维点云地图的精确度不低于预设的精确度阈值。
在一些实施例中，所述处理装置501根据所述位姿数据的置信度调整所述回环约束条件的数量，包括：若所述置信度小于预设的置信度阈值，增加所述回环约束条件的数量；和/或若所述置信度大于或等于预设的置信度阈值，减少所述回环约束条件的数量。
在一些实施例中，所述处理装置501根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量，包括：计算所述三维尺寸信息与所述目标对象的参考三维尺寸信息之间的差值；根据所述差值调整所述回环约束条件的数量。
在一些实施例中，所述回环约束条件还基于两帧空间位姿之差在第二预设范围内、且朝向相反的所述位姿数据而确定。
在一些实施例中，所述初始点云数据包括经过配准的连续多帧点云数据。
在一些实施例中，所述处理装置501还用于：从所述初始点云数据中滤除移动物体对应的点云数据。
在一些实施例中，所述处理装置501根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据，包括：以所述位姿数 据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数，以获取所述优化位姿数据。
在一些实施例中，所述处理装置501以所述位姿数据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数，以获取所述优化位姿数据，包括：以所述位姿数据作为图顶点，并以所述回环约束条件以及相邻帧位姿数据之间的相对位姿关系作为图的边，求解图的优化目标函数，以获取所述优化位姿数据。
在一些实施例中，所述优化位姿数据从一定范围内的位姿数据中查找得到；其中，所述优化位姿数据的查找范围根据对所述可移动平台进行定位的定位信号的强度和/或所述优化位姿数据的置信度而设置。
在一些实施例中，所述处理装置501根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，包括：对所述优化位姿数据进行校验；根据校验成功的优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。
在一些实施例中，所述处理装置501对所述优化位姿数据进行校验，包括：计算所述优化位姿数据与相邻若干帧的优化位姿数据的平均值；计算所述优化位姿数据与所述平均值之间的差值；根据所述差值对所述优化位姿数据进行校验。
在一些实施例中，所述处理装置501根据所述差值对所述优化位姿数据进行校验，包括：若所述差值大于预设值，判定所述优化位姿数据校验成功，和/或若所述差值小于或等于所述预设值，判定所述优化位姿数据校验失败。
在一些实施例中，所述处理装置501根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，包括：根据所述优化位姿数据对多帧所述初始点云数据进行配准；根据配准后的多帧所述初始点云数据构建所述三维点云地图。
以上三维点云地图构建系统中处理装置501所执行的方法的具体实施例详见本公开三维点云地图构建方法的实施例，此处不再赘述。
在一些实施例中，本公开还提供一种三维点云地图构建设备，所述三维点云地图构建设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现任一实施例所述方法的步骤。
图6示出了本公开实施例所提供的一种更为具体的三维点云地图构建设备的硬件结构示意图，该设备可以包括：处理器601、存储器602、输入/输出接口603、通信接口604和总线605。其中处理器601、存储器602、输入/输出接口603和通信接口 604通过总线605实现彼此之间在设备内部的通信连接。
处理器601可以采用通用的CPU(Central Processing Unit，中央处理器)、微处理器、应用专用集成电路(Application Specific Integrated Circuit，ASIC)、或者一个或多个集成电路等方式实现，用于执行相关程序，以实现本说明书实施例所提供的技术方案。
存储器602可以采用ROM(Read Only Memory，只读存储器)、RAM(Random Access Memory，随机存取存储器)、静态存储设备，动态存储设备等形式实现。存储器602可以存储操作系统和其他应用程序，在通过软件或者固件来实现本说明书实施例所提供的技术方案时，相关的程序代码保存在存储器602中，并由处理器602来调用执行。
输入/输出接口603用于连接输入/输出模块，以实现信息输入及输出。输入输出/模块可以作为组件配置在设备中(图中未示出)，也可以外接于设备以提供相应功能。其中输入设备可以包括键盘、鼠标、触摸屏、麦克风、各类传感器等，输出设备可以包括显示器、扬声器、振动器、指示灯等。
通信接口604用于连接通信模块(图中未示出)，以实现本设备与其他设备的通信交互。其中通信模块可以通过有线方式(例如USB、网线等)实现通信，也可以通过无线方式(例如移动网络、WIFI、蓝牙等)实现通信。
总线605包括一通路，在设备的各个组件(例如处理器601、存储器602、输入/输出接口603和通信接口604)之间传输信息。
需要说明的是，尽管上述设备仅示出了处理器601、存储器602、输入/输出接口603、通信接口604以及总线605，但是在具体实施过程中，该设备还可以包括实现正常运行所必需的其他组件。此外，本领域的技术人员可以理解的是，上述设备中也可以仅包含实现本说明书实施例方案所必需的组件，而不必包含图中所示的全部组件。
在一些实施例中，本公开还提供一种计算机可读存储介质，所述可读存储介质上存储有若干计算机指令，所述计算机指令被执行时实现任一实施例所述方法的步骤。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器 (RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
通过以上的实施方式的描述可知，本领域的技术人员可以清楚地了解到本说明书实施例可借助软件加必需的通用硬件平台的方式来实现。基于这样的理解，本说明书实施例的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在存储介质中，如ROM/RAM、磁碟、光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本说明书实施例各个实施例或者实施例的某些部分所述的方法。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于装置实施例而言，由于其基本相似于方法实施例，所以描述得比较简单，相关之处参见方法实施例的部分说明即可。以上所描述的装置实施例仅仅是示意性的，其中所述作为分离部件说明的模块可以是或者也可以不是物理上分开的，在实施本说明书实施例方案时可以把各模块的功能在同一个或多个软件和/或硬件中实现。也可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性劳动的情况下，即可以理解并实施。
以上实施例中的各种技术特征可以任意进行组合，只要特征之间的组合不存在冲突或矛盾，但是限于篇幅，未进行一一描述，因此上述实施方式中的各种技术特征的任意进行组合也属于本说明书公开的范围。
本领域技术人员在考虑说明书及实践这里公开的说明书后，将容易想到本公开的其它实施方案。本公开旨在涵盖本公开的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本公开的一般性原理并包括本公开未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本公开的真正范围和精神由下面的权利要求指出。
应当理解的是，本公开并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。本公开的范围仅由所附的权利要求来限制。
以上所述仅为本公开的较佳实施例而已，并不用以限制本公开，凡在本公开的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本公开保护的范围之内。

Claims (48)

  1. 一种三维点云地图构建方法，其特征在于，所述方法包括：
    获取搭载于可移动平台上的激光雷达系统采集的多帧初始点云数据，并获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据；
    确定至少一个回环约束条件，所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定；
    根据所述至少一个回环约束条件对多帧所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据；
    根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。
  2. 根据权利要求1所述的方法，其特征在于，所述可移动平台上还搭载有导航系统，所述位姿数据基于所述导航系统的导航数据获取。
  3. 根据权利要求2所述的方法，其特征在于，所述导航系统包括第一导航系统和第二导航系统；
    所述获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据，包括：
    获取所述第一导航系统输出的第一导航数据和所述第二导航系统输出的第二导航数据；
    基于所述第一导航数据和所述第二导航数据获取所述位姿数据。
  4. 根据权利要求2所述的方法，其特征在于，所述方法还包括：
    对所述位姿数据与所述初始点云数据进行对齐。
  5. 根据权利要求4所述的方法，其特征在于，所述对所述位姿数据与所述初始点云数据进行对齐，包括：
    基于所述导航系统的全局时间戳和所述激光雷达系统的全局时间戳对所述位姿数据与所述初始点云数据进行对齐。
  6. 根据权利要求1所述的方法，其特征在于，多帧所述初始点云数据包括所述可移动平台至少两次通过同一地点时，所述激光雷达系统采集到的初始点云数据。
  7. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    对所述回环约束条件的数量进行调整。
  8. 根据权利要求7所述的方法，其特征在于，所述对所述回环约束条件的数量进行调整，包括：
    根据所述位姿数据的置信度调整所述回环约束条件的数量；和/或
    根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条 件的数量。
  9. 根据权利要求8所述的方法，其特征在于，所述位姿数据的置信度根据对所述可移动平台进行定位的定位信号的强度和/或所述三维点云地图的精确度确定。
  10. 根据权利要求9所述的方法，其特征在于，所述方法还包括：
    在所述可移动平台的移动轨迹上每隔一段距离抽样一定范围内的三维点云地图；
    计算抽样的三维点云地图中各个地面点的高度差；
    根据所述高度差确定所述三维点云地图的精确度。
  11. 根据权利要求10所述的方法，其特征在于，所述根据所述高度差确定所述三维点云地图的精确度，包括：
    若所述高度差大于预设的高度阈值，判定所述三维点云地图的精确度低于预设的精确度阈值；和/或
    若所述高度差小于或等于所述高度阈值，判定所述三维点云地图的精确度不低于预设的精确度阈值。
  12. 根据权利要求8所述的方法，其特征在于，所述根据所述位姿数据的置信度调整所述回环约束条件的数量，包括：
    若所述置信度小于预设的置信度阈值，增加所述回环约束条件的数量；和/或
    若所述置信度大于或等于预设的置信度阈值，减少所述回环约束条件的数量。
  13. 根据权利要求8所述的方法，其特征在于，所述根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量，包括：
    计算所述三维尺寸信息与所述目标对象的参考三维尺寸信息之间的差值；
    根据所述差值调整所述回环约束条件的数量。
  14. 根据权利要求1所述的方法，其特征在于，所述回环约束条件还基于两帧空间位姿之差在第二预设范围内、且朝向相反的所述位姿数据而确定。
  15. 根据权利要求1所述的方法，其特征在于，所述初始点云数据包括经过配准的连续多帧点云数据。
  16. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    从所述初始点云数据中滤除移动物体对应的点云数据。
  17. 根据权利要求1所述的方法，其特征在于，所述根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据，包括：
    以所述位姿数据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数，以获取所述优化位姿数据。
  18. 根据权利要求17所述的方法，其特征在于，所述以所述位姿数据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数，以获取所述优化位姿数据，包括：
    以所述位姿数据作为图顶点，并以所述回环约束条件以及相邻帧位姿数据之间的相对位姿关系作为图的边，求解图的优化目标函数，以获取所述优化位姿数据。
  19. 根据权利要求17所述的方法，其特征在于，所述优化位姿数据从一定范围内的位姿数据中查找得到；其中，所述优化位姿数据的查找范围根据对所述可移动平台进行定位的定位信号的强度和/或所述优化位姿数据的置信度而设置。
  20. 根据权利要求1所述的方法，其特征在于，所述根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，包括：
    对所述优化位姿数据进行校验；
    根据校验成功的优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。
  21. 根据权利要求20所述的方法，其特征在于，所述对所述优化位姿数据进行校验，包括：
    计算所述优化位姿数据与相邻若干帧的优化位姿数据的平均值；
    计算所述优化位姿数据与所述平均值之间的差值；
    根据所述差值对所述优化位姿数据进行校验。
  22. 根据权利要求21所述的方法，其特征在于，所述根据所述差值对所述优化位姿数据进行校验，包括：
    若所述差值大于预设值，判定所述优化位姿数据校验成功，和/或
    若所述差值小于或等于所述预设值，判定所述优化位姿数据校验失败。
  23. 根据权利要求1所述的方法，其特征在于，所述根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，包括：
    根据所述优化位姿数据对多帧所述初始点云数据进行配准；
    根据配准后的多帧所述初始点云数据构建所述三维点云地图。
  24. 一种三维点云地图构建系统，其特征在于，所述三维点云地图构建系统包括：
    处理装置和搭载于可移动平台上的激光雷达系统；
    所述激光雷达系统用于采集多帧初始点云数据；
    所述处理器用于确定至少一个回环约束条件；根据所述至少一个回环约束条件对多帧所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据，得到每帧所述初始点云数据的优化位姿数据；根据所述优化位姿数据，基于多帧所述初始点云 数据构建三维点云地图；所述回环约束条件基于两帧空间位姿之差在第一预设范围内、且时序上不连续的所述位姿数据而确定。
  25. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述三维点云地图构建系统还包括：
    搭载于所述可移动平台上的导航系统；
    所述导航系统用于获取所述激光雷达系统采集多帧所述初始点云数据时的位姿数据。
  26. 根据权利要求25所述的三维点云地图构建系统，其特征在于，所述导航系统包括：
    第一导航系统和第二导航系统；
    所述第一导航系统用于获取所述激光雷达系统采集多帧所述初始点云数据时的第一位姿数据；
    所述第二导航系统用于获取所述激光雷达系统采集多帧所述初始点云数据时的第二位姿数据；
    所述处理装置还用于：
    基于所述第一导航数据和所述第二导航数据获取所述位姿数据。
  27. 根据权利要求25所述的三维点云地图构建系统，其特征在于，所述处理装置还用于：
    对所述位姿数据与所述初始点云数据进行对齐。
  28. 根据权利要求27所述的三维点云地图构建系统，其特征在于，所述处理装置对所述位姿数据与所述初始点云数据进行对齐，包括：
    基于所述导航系统的全局时间戳和所述激光雷达系统的全局时间戳对所述位姿数据与所述初始点云数据进行对齐。
  29. 根据权利要求24所述的三维点云地图构建系统，其特征在于，多帧所述初始点云数据包括所述可移动平台至少两次通过同一地点时，所述激光雷达系统采集到的初始点云数据。
  30. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述处理装置还用于：
    对所述回环约束条件的数量进行调整。
  31. 根据权利要求30所述的三维点云地图构建系统，其特征在于，所述处理装置对所述回环约束条件的数量进行调整，包括：
    根据所述位姿数据的置信度调整所述回环约束条件的数量；和/或
    根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量。
  32. 根据权利要求31所述的三维点云地图构建系统，其特征在于，所述位姿数据的置信度根据对所述可移动平台进行定位的定位信号的强度和/或所述三维点云地图的精确度确定。
  33. 根据权利要求32所述的三维点云地图构建系统，其特征在于，所述处理装置还用于：
    在所述可移动平台的移动轨迹上每隔一段距离抽样一定范围内的三维点云地图；
    计算抽样的三维点云地图中各个地面点的高度差；
    根据所述高度差确定所述三维点云地图的精确度。
  34. 根据权利要求33所述的三维点云地图构建系统，其特征在于，所述处理装置根据所述高度差确定所述三维点云地图的精确度，包括：
    若所述高度差大于预设的高度阈值，判定所述三维点云地图的精确度低于预设的精确度阈值；和/或
    若所述高度差小于或等于所述高度阈值，判定所述三维点云地图的精确度不低于预设的精确度阈值。
  35. 根据权利要求31所述的三维点云地图构建系统，其特征在于，所述处理装置根据所述位姿数据的置信度调整所述回环约束条件的数量，包括：
    若所述置信度小于预设的置信度阈值，增加所述回环约束条件的数量；和/或
    若所述置信度大于或等于预设的置信度阈值，减少所述回环约束条件的数量。
  36. 根据权利要求31所述的三维点云地图构建系统，其特征在于，所述处理装置根据所述三维点云地图中至少一个目标对象的三维尺寸信息调整所述回环约束条件的数量，包括：
    计算所述三维尺寸信息与所述目标对象的参考三维尺寸信息之间的差值；
    根据所述差值调整所述回环约束条件的数量。
  37. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述回环约束条件还基于两帧空间位姿之差在第二预设范围内、且朝向相反的所述位姿数据而确定。
  38. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述初始点云数据包括经过配准的连续多帧点云数据。
  39. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述处理装置 还用于：
    从所述初始点云数据中滤除移动物体对应的点云数据。
  40. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述处理装置根据所述至少一个回环约束条件对所述位姿数据进行处理，得到每帧所述初始点云数据的优化位姿数据，包括：
    以所述位姿数据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数，以获取所述优化位姿数据。
  41. 根据权利要求40所述的三维点云地图构建系统，其特征在于，所述以所述位姿数据作为图顶点，并以所述回环约束条件作为图的边，求解图的优化目标函数，以获取所述优化位姿数据，包括：
    以所述位姿数据作为图顶点，并以所述回环约束条件以及相邻帧位姿数据之间的相对位姿关系作为图的边，求解图的优化目标函数，以获取所述优化位姿数据。
  42. 根据权利要求40所述的三维点云地图构建系统，其特征在于，所述优化位姿数据从一定范围内的位姿数据中查找得到；其中，所述优化位姿数据的查找范围根据对所述可移动平台进行定位的定位信号的强度和/或所述优化位姿数据的置信度而设置。
  43. 根据权利要求24所述的系统，其特征在于，所述处理装置根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，包括：
    对所述优化位姿数据进行校验；
    根据校验成功的优化位姿数据，基于多帧所述初始点云数据构建三维点云地图。
  44. 根据权利要求42所述的系统，其特征在于，所述处理装置对所述优化位姿数据进行校验，包括：
    计算所述优化位姿数据与相邻若干帧的优化位姿数据的平均值；
    计算所述优化位姿数据与所述平均值之间的差值；
    根据所述差值对所述优化位姿数据进行校验。
  45. 根据权利要求44所述的系统，其特征在于，所述处理装置根据所述差值对所述优化位姿数据进行校验，包括：
    若所述差值大于预设值，判定所述优化位姿数据校验成功，和/或
    若所述差值小于或等于所述预设值，判定所述优化位姿数据校验失败。
  46. 根据权利要求24所述的三维点云地图构建系统，其特征在于，所述处理装置根据所述优化位姿数据，基于多帧所述初始点云数据构建三维点云地图，包括：
    根据所述优化位姿数据对多帧所述初始点云数据进行配准；
    根据配准后的多帧所述初始点云数据构建所述三维点云地图。
  47. 一种三维点云地图构建设备，其特征在于，所述三维点云地图构建设备包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现权利要求1至23任意一项所述的方法。
  48. 一种计算机可读存储介质，所述可读存储介质上存储有若干计算机指令，其特征在于，所述计算机指令被执行时实现权利要求1至23任意一项所述方法的步骤。
PCT/CN2019/129307 2019-12-27 2019-12-27 三维点云地图构建方法、系统和设备 WO2021128297A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201980091861.6A CN113424232B (zh) 2019-12-27 2019-12-27 三维点云地图构建方法、系统和设备
PCT/CN2019/129307 WO2021128297A1 (zh) 2019-12-27 2019-12-27 三维点云地图构建方法、系统和设备

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/129307 WO2021128297A1 (zh) 2019-12-27 2019-12-27 三维点云地图构建方法、系统和设备

Publications (1)

Publication Number Publication Date
WO2021128297A1 true WO2021128297A1 (zh) 2021-07-01

Family

ID=76573535

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/129307 WO2021128297A1 (zh) 2019-12-27 2019-12-27 三维点云地图构建方法、系统和设备

Country Status (2)

Country Link
CN (1) CN113424232B (zh)
WO (1) WO2021128297A1 (zh)

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113534097A (zh) * 2021-08-27 2021-10-22 北京工业大学 一种适用于旋轴激光雷达的优化方法
CN113639745A (zh) * 2021-08-03 2021-11-12 北京航空航天大学 一种点云地图的构建方法、装置及存储介质
CN113671530A (zh) * 2021-08-06 2021-11-19 北京京东乾石科技有限公司 位姿确定方法、装置及存储介质和电子设备
CN113671527A (zh) * 2021-07-23 2021-11-19 国电南瑞科技股份有限公司 一种提高配网带电作业机器人的精准作业方法及装置
CN113689497A (zh) * 2021-08-11 2021-11-23 影石创新科技股份有限公司 位姿优化方法、装置、设备和存储介质
CN113706676A (zh) * 2021-08-26 2021-11-26 京东鲲鹏(江苏)科技有限公司 用于点云数据的模型自监督训练方法和装置
CN113744236A (zh) * 2021-08-30 2021-12-03 阿里巴巴达摩院(杭州)科技有限公司 回环检测方法、装置、存储介质及计算机程序产品
CN113768419A (zh) * 2021-09-17 2021-12-10 安克创新科技股份有限公司 确定扫地机清扫方向的方法、装置及扫地机
CN113776515A (zh) * 2021-08-31 2021-12-10 南昌工学院 一种机器人导航方法、装置、计算机设备和存储介质
CN113984065A (zh) * 2021-10-27 2022-01-28 山东亚历山大智能科技有限公司 一种用于室内机器人的反光板地图生成方法及系统
CN114137953A (zh) * 2021-10-12 2022-03-04 杭州电子科技大学 基于三维激光雷达的电力巡检机器人系统及建图方法
CN114234976A (zh) * 2021-11-29 2022-03-25 山东恒创智控科技有限公司 一种改进型slam与多传感器融合的机器人定位方法及系统
CN114322987A (zh) * 2021-12-27 2022-04-12 北京三快在线科技有限公司 一种构建高精地图的方法及装置
CN114372981A (zh) * 2022-03-21 2022-04-19 季华实验室 T型工件焊缝识别方法、装置、电子设备及存储介质
CN114463512A (zh) * 2021-12-24 2022-05-10 广州极飞科技股份有限公司 点云数据的处理方法、矢量化方法及装置
CN114659513A (zh) * 2022-03-11 2022-06-24 北京航空航天大学 一种面向非结构化道路的点云地图构建与维护方法
CN114723672A (zh) * 2022-03-09 2022-07-08 杭州易现先进科技有限公司 一种三维重建数据采集校验的方法、系统、装置和介质
CN115937383A (zh) * 2022-09-21 2023-04-07 北京字跳网络技术有限公司 渲染图像的方法、装置、电子设备及存储介质
CN116908818A (zh) * 2023-07-13 2023-10-20 广东喜讯智能科技有限公司 基于rtk无人机的激光雷达的标定方法、装置及存储介质
CN117761717A (zh) * 2024-02-21 2024-03-26 天津大学四川创新研究院 一种自动回环三维重建系统及运行方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115952139B (zh) * 2023-03-14 2023-06-30 武汉芯云道数据科技有限公司 一种移动设备多帧三维图像处理方法及系统
CN117351167A (zh) * 2023-12-05 2024-01-05 深圳市其域创新科技有限公司 一种基于gpu的三维地图构建方法、装置、设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180088234A1 (en) * 2016-09-27 2018-03-29 Carnegie Mellon University Robust Localization and Localizability Prediction Using a Rotating Laser Scanner
CN108267141A (zh) * 2016-12-30 2018-07-10 乐视汽车（北京）有限公司 道路点云数据处理系统
CN108280866A (zh) * 2016-12-30 2018-07-13 乐视汽车（北京）有限公司 道路点云数据处理方法及系统
CN108550318A (zh) * 2018-03-12 2018-09-18 浙江大华技术股份有限公司 一种构建地图的方法及装置
CN109507677A (zh) * 2018-11-05 2019-03-22 浙江工业大学 一种结合gps和雷达里程计的slam方法

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111417871A (zh) * 2017-11-17 2020-07-14 迪普迈普有限公司 基于激光雷达利用高清晰度地图的集成运动估计的迭代最近点处理
WO2019127445A1 (zh) * 2017-12-29 2019-07-04 深圳前海达闼云端智能科技有限公司 三维建图方法、装置、系统、云端平台、电子设备和计算机程序产品
CN109064506B (zh) * 2018-07-04 2020-03-13 百度在线网络技术（北京）有限公司 高精度地图生成方法、装置及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180088234A1 (en) * 2016-09-27 2018-03-29 Carnegie Mellon University Robust Localization and Localizability Prediction Using a Rotating Laser Scanner
CN108267141A (zh) * 2016-12-30 2018-07-10 乐视汽车（北京）有限公司 道路点云数据处理系统
CN108280866A (zh) * 2016-12-30 2018-07-13 乐视汽车（北京）有限公司 道路点云数据处理方法及系统
CN108550318A (zh) * 2018-03-12 2018-09-18 浙江大华技术股份有限公司 一种构建地图的方法及装置
CN109507677A (zh) * 2018-11-05 2019-03-22 浙江工业大学 一种结合gps和雷达里程计的slam方法

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113671527A (zh) * 2021-07-23 2021-11-19 国电南瑞科技股份有限公司 一种提高配网带电作业机器人的精准作业方法及装置
CN113639745B (zh) * 2021-08-03 2023-10-20 北京航空航天大学 一种点云地图的构建方法、装置及存储介质
CN113639745A (zh) * 2021-08-03 2021-11-12 北京航空航天大学 一种点云地图的构建方法、装置及存储介质
CN113671530A (zh) * 2021-08-06 2021-11-19 北京京东乾石科技有限公司 位姿确定方法、装置及存储介质和电子设备
CN113671530B (zh) * 2021-08-06 2024-01-12 北京京东乾石科技有限公司 位姿确定方法、装置及存储介质和电子设备
CN113689497A (zh) * 2021-08-11 2021-11-23 影石创新科技股份有限公司 位姿优化方法、装置、设备和存储介质
CN113706676B (zh) * 2021-08-26 2024-01-16 京东鲲鹏(江苏)科技有限公司 用于点云数据的模型自监督训练方法和装置
CN113706676A (zh) * 2021-08-26 2021-11-26 京东鲲鹏(江苏)科技有限公司 用于点云数据的模型自监督训练方法和装置
CN113534097A (zh) * 2021-08-27 2021-10-22 北京工业大学 一种适用于旋轴激光雷达的优化方法
CN113534097B (zh) * 2021-08-27 2023-11-24 北京工业大学 一种适用于旋轴激光雷达的优化方法
CN113744236A (zh) * 2021-08-30 2021-12-03 阿里巴巴达摩院(杭州)科技有限公司 回环检测方法、装置、存储介质及计算机程序产品
CN113776515A (zh) * 2021-08-31 2021-12-10 南昌工学院 一种机器人导航方法、装置、计算机设备和存储介质
CN113776515B (zh) * 2021-08-31 2022-06-10 南昌工学院 一种机器人导航方法、装置、计算机设备和存储介质
CN113768419A (zh) * 2021-09-17 2021-12-10 安克创新科技股份有限公司 确定扫地机清扫方向的方法、装置及扫地机
CN114137953A (zh) * 2021-10-12 2022-03-04 杭州电子科技大学 基于三维激光雷达的电力巡检机器人系统及建图方法
CN113984065A (zh) * 2021-10-27 2022-01-28 山东亚历山大智能科技有限公司 一种用于室内机器人的反光板地图生成方法及系统
CN114234976A (zh) * 2021-11-29 2022-03-25 山东恒创智控科技有限公司 一种改进型slam与多传感器融合的机器人定位方法及系统
CN114463512A (zh) * 2021-12-24 2022-05-10 广州极飞科技股份有限公司 点云数据的处理方法、矢量化方法及装置
CN114322987A (zh) * 2021-12-27 2022-04-12 北京三快在线科技有限公司 一种构建高精地图的方法及装置
CN114322987B (zh) * 2021-12-27 2024-02-23 北京三快在线科技有限公司 一种构建高精地图的方法及装置
CN114723672A (zh) * 2022-03-09 2022-07-08 杭州易现先进科技有限公司 一种三维重建数据采集校验的方法、系统、装置和介质
CN114659513A (zh) * 2022-03-11 2022-06-24 北京航空航天大学 一种面向非结构化道路的点云地图构建与维护方法
CN114659513B (zh) * 2022-03-11 2024-04-09 北京航空航天大学 一种面向非结构化道路的点云地图构建与维护方法
CN114372981B (zh) * 2022-03-21 2022-06-17 季华实验室 T型工件焊缝识别方法、装置、电子设备及存储介质
CN114372981A (zh) * 2022-03-21 2022-04-19 季华实验室 T型工件焊缝识别方法、装置、电子设备及存储介质
CN115937383A (zh) * 2022-09-21 2023-04-07 北京字跳网络技术有限公司 渲染图像的方法、装置、电子设备及存储介质
CN115937383B (zh) * 2022-09-21 2023-10-10 北京字跳网络技术有限公司 渲染图像的方法、装置、电子设备及存储介质
CN116908818A (zh) * 2023-07-13 2023-10-20 广东喜讯智能科技有限公司 基于rtk无人机的激光雷达的标定方法、装置及存储介质
CN117761717A (zh) * 2024-02-21 2024-03-26 天津大学四川创新研究院 一种自动回环三维重建系统及运行方法

Also Published As

Publication number Publication date
CN113424232A (zh) 2021-09-21
CN113424232B (zh) 2024-03-15

Similar Documents

Publication Publication Date Title
WO2021128297A1 (zh) 三维点云地图构建方法、系统和设备
US11482008B2 (en) Directing board repositioning during sensor calibration for autonomous vehicles
US10552689B2 (en) Automatic occlusion detection in road network data
US20240005167A1 (en) Annotating high definition map data with semantic labels
CN111912417B (zh) 地图构建方法、装置、设备及存储介质
KR20200121274A (ko) 전자 지도를 업데이트하기 위한 방법, 장치 및 컴퓨터 판독 가능한 저장 매체
US11590989B2 (en) Training data generation for dynamic objects using high definition map data
CN112380312B (zh) 基于栅格检测的激光地图更新方法、终端及计算机设备
WO2020224305A1 (zh) 用于设备定位的方法、装置及设备
US20210278549A1 (en) Lane-level navigation system for ground vehicles with lidar and cellular signals
US11024054B2 (en) Method, apparatus, and system for estimating the quality of camera pose data using ground control points of known quality
JP2022542289A (ja) 地図作成方法、地図作成装置、電子機器、記憶媒体及びコンピュータプログラム製品
US11232582B2 (en) Visual localization using a three-dimensional model and image segmentation
CN110850457A (zh) 一种用于室内煤场的无人机定位导航方法
Tschopp et al. Hough $^ 2$ Map–Iterative Event-Based Hough Transform for High-Speed Railway Mapping
CN113989451A (zh) 高精地图构建方法、装置及电子设备
CN110223223A (zh) 街道扫描方法、装置及扫描仪
Chen et al. Multi-level scene modeling and matching for smartphone-based indoor localization
TWM600873U (zh) 用於檢測道路破損之檢測系統
Antoniou et al. Localization and driving behavior classification with smartphone sensors in direct absence of global navigation satellite systems
CN114120631A (zh) 构建动态高精度地图的方法、装置及交通云控平台
Rae et al. Reducing multipath effects in vehicle localization by fusing GPS with machine vision
Ernst et al. Large-scale 3D Roadside Modelling with Road Geometry Analysis: Digital Roads New Zealand
Chawla et al. Monocular vision based crowdsourced 3d traffic sign positioning with unknown camera intrinsics and distortion coefficients
Kreibich et al. Lane-level matching algorithm based on GNSS, IMU and map data

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19957485

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19957485

Country of ref document: EP

Kind code of ref document: A1