WO2024051458A1 - 续扫重定位方法、装置、设备、存储介质及三维续扫方法 - Google Patents

续扫重定位方法、装置、设备、存储介质及三维续扫方法 Download PDF

Info

Publication number
WO2024051458A1
WO2024051458A1 PCT/CN2023/113314 CN2023113314W WO2024051458A1 WO 2024051458 A1 WO2024051458 A1 WO 2024051458A1 CN 2023113314 W CN2023113314 W CN 2023113314W WO 2024051458 A1 WO2024051458 A1 WO 2024051458A1
Authority
WO
WIPO (PCT)
Prior art keywords
current
dimensional
semantic map
spatial
semantic
Prior art date
Application number
PCT/CN2023/113314
Other languages
English (en)
French (fr)
Inventor
赵开勇
Original Assignee
深圳市其域创新科技有限公司
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 深圳市其域创新科技有限公司 filed Critical 深圳市其域创新科技有限公司
Publication of WO2024051458A1 publication Critical patent/WO2024051458A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00Three dimensional [3D] modelling, e.g. data description of 3D objects
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T19/00Manipulating 3D models or images for computer graphics
    • G06T19/20Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/26Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74Image or video pattern matching; Proximity measures in feature spaces

Definitions

  • Embodiments of the present invention relate to the technical field of three-dimensional scanning, and specifically relate to a continued scanning relocation method, device, equipment, storage medium and three-dimensional continuing scanning method.
  • embodiments of the present invention provide a method, device, computing device, storage medium and three-dimensional continued scanning method to solve the problems existing in the prior art.
  • a method for continued scanning and relocation includes:
  • the current spatial semantic map based on multiple frames of current sensor information acquired by the scanning device in the three-dimensional scene.
  • the multiple frames of current sensor information are obtained from different sensors, and the current spatial semantic map has current spatial location information;
  • the historical spatial semantic map is determined based on multiple frames of historical sensor information.
  • the multiple frames of historical sensor information are determined by the different sensors before the scanning device interrupts scanning. It is obtained that the historical spatial semantic map has historical spatial location information;
  • the scanning device is successfully repositioned in the three-dimensional scene.
  • the current sensor information of multiple frames includes current three-dimensional sensor information of multiple frames
  • the current spatial semantic map includes a current three-dimensional semantic map, which is determined based on multiple frames of current sensor information obtained by the scanning device in the three-dimensional scene.
  • the current spatial semantic map further includes:
  • the current three-dimensional sensor information of multiple frames includes current visual image information of multiple frames
  • the current three-dimensional semantic map includes the current visual semantic map
  • the current sensor information of multiple frames obtained by the scanning device in the three-dimensional scene is Determine the current spatial semantic map, further including:
  • performing three-dimensional visual semantic segmentation on multiple frames of the current visual image information based on the current three-dimensional visual point cloud and generating the current visual semantic map further includes:
  • the current visual semantic map is generated according to the characteristic information of the three-dimensional visual semantic object, where the characteristic information includes the current spatial position information.
  • the current three-dimensional sensor information of multiple frames includes current laser information of multiple frames
  • the current three-dimensional semantic map includes the current laser semantic map, which is determined based on the current sensor information of multiple frames acquired by the scanning device in the three-dimensional scene.
  • the current spatial semantic map further includes:
  • the current sensor information of multiple frames also includes current IMU information of multiple frames, and determining the current spatial semantic map based on the current sensor information of multiple frames acquired by the scanning device in the three-dimensional scene further includes:
  • the current IMU space trajectory point cloud and the current three-dimensional point cloud are matched according to the spatial coordinate position to obtain a second value of the current IMU space trajectory point cloud and the current three-dimensional point cloud.
  • Matching degree further includes:
  • the current spatial semantic map includes k current spatial semantic objects, so
  • the historical spatial semantic map includes h historical spatial semantic objects.
  • the current spatial semantic map and the historical spatial semantic map are matched through a subgraph search method to obtain the current spatial semantic map and the historical spatial semantic map.
  • the first matching degree of the map further includes:
  • a three-dimensional continuous scanning method includes:
  • the three-dimensional model reconstruction is completed based on the continued scanning sensor information of the scanning device.
  • a device for continuous scanning and repositioning includes:
  • the first determination module is used to determine the current spatial semantic map based on multiple frames of current sensor information obtained by the scanning device in the three-dimensional scene.
  • the multiple frames of current sensor information are obtained from different sensors, and the current spatial semantic map has current spatial location information.
  • the first acquisition module is used to acquire the historical spatial semantic map of the scanning device for the three-dimensional scene.
  • the historical spatial semantic map is determined based on multiple frames of historical sensor information.
  • the multiple frames of historical sensor information are interrupted by the scanning device. Obtained from different sensors before scanning, the historical spatial semantic map has historical spatial location information;
  • the first matching module is used to match the current spatial semantic map and the historical spatial semantic map through sub-graph search, and obtain the first matching degree between the current spatial semantic map and the historical spatial semantic map;
  • a first judgment module used to judge whether the first matching degree is greater than or equal to the first matching threshold
  • a second determination module configured to determine that the scanning device is successfully repositioned in the three-dimensional scene if the first matching degree is greater than or equal to the first matching threshold.
  • a computing device including: a processor, a memory, a communication interface, and a communication bus.
  • the processor, the memory, and the communication interface complete each other through the communication bus. communication between;
  • the memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform the operation of the scanning relocation method described in any one of the above.
  • a computer-readable storage medium in which at least one executable instruction is stored, and the executable instruction executes any one of the above-mentioned instructions when running. Continue the operation of the relocation method.
  • the embodiment of the present invention can effectively reduce the amount of data, reduce the processing load of the processor, and reduce the storage pressure of the storage medium.
  • the subgraph search method can quickly match the current spatial semantic map and the historical spatial semantic map.
  • the historical spatial semantic map improves matching efficiency and thereby increases relocation speed.
  • the processor determines the position of the scanning device through relocation, so that the processor can automatically continue the interrupted scanning part of the 3D model during the continued scanning and modeling process until the scanning is completed. Generate a 3D model of a complete 3D scene. There is no need to manually splice the partial 3D model after the continued scan to the historical 3D model before the continued scan, which facilitates operation and improves scanning efficiency.
  • the spatial position of the scanning device can be calculated from different dimensions, making the spatial position of the scanning device more accurate and improving relocation accuracy, which is beneficial to the processor. According to the position of the scanning device, the interruption position of the unfinished three-dimensional model before the interruption of scanning is more accurately determined.
  • Figure 1 shows a schematic flow chart of a method for continuous scanning and relocation provided by an embodiment of the present invention
  • Figure 2 shows a schematic flow chart of a three-dimensional continuous scanning method provided by some embodiments of the present invention
  • Figure 3 shows a schematic structural diagram of a continuous scanning and repositioning device provided by an embodiment of the present invention
  • Figure 4 shows a schematic structural diagram of a three-dimensional continuous scanning device provided by an embodiment of the present invention
  • Figure 5 shows a schematic structural diagram of a computing device provided by some embodiments of the present invention.
  • the inventor provides a method for continuous scanning and relocation.
  • multiple frames of different current sensor information are obtained, and the current spatial semantic map of the scanning device is calculated based on the current sensor information of multiple frames, and the scanning device is interrupted.
  • the historical spatial semantic map previously determined by multiple frames of historical sensor information is matched with the current spatial semantic map to determine the current location of the scanning device. If the relocation is successful, the scanning device can continue scanning based on the relocation location, and in During the re-scanning process, the historical 3D reconstruction model before the scan is automatically continued based on the relocation position, and the 3D information after the scan is automatically spliced to the historical 3D reconstruction model until the entire 3D model is completely formed.
  • the processor determines the position of the scanning device through relocation, and the processor automatically continues the reconstruction of the 3D model during the continued scanning modeling process. There is no need to manually splice the partial 3D model after the continued scanning to the historical 3D model before the continued scanning. Medium-sized, easy to operate and improve scanning efficiency. In addition, relocation through subgraph search of the semantic map can quickly locate and improve relocation efficiency. Coordinating positioning with different sensors can help improve positioning accuracy.
  • FIG. 1 shows a flow chart of a method for continued scanning and relocation provided by an embodiment of the present invention.
  • the method is executed by a computing device.
  • the computing device may be a computing device including one or more processors.
  • the processor may be a central processing unit.
  • the processor CPU may be an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention, which are not limited here.
  • the one or more processors included in the computing device may be the same type of processor, such as one or more CPUs; or they may be different types of processors, such as one or more CPUs and one or more ASICs, where No restrictions.
  • the method includes the following steps:
  • Step 110 Determine the current spatial semantic map based on the multi-frame current sensor information obtained by the scanning device in the three-dimensional scene.
  • the multi-frame current sensor information is obtained from different sensors, and the current spatial semantic map has current spatial location information.
  • Step 120 Obtain the historical spatial semantic map of the scanning device for the three-dimensional scene.
  • the historical spatial semantic map is determined based on multi-frame historical sensor information.
  • the multi-frame historical sensor information is obtained from different sensors before the scanning device interrupts scanning.
  • the historical spatial semantic map has historical space. location information.
  • Step 130 Match the current spatial semantic map and the historical spatial semantic map through subgraph search to obtain the first matching degree between the current spatial semantic map and the historical spatial semantic map.
  • Step 140 Determine whether the first matching degree is greater than or equal to the first matching threshold.
  • Step 150 If the first matching degree is greater than or equal to the first matching threshold, determine that the scanning device is successfully repositioned in the three-dimensional scene.
  • the scanning device performs three-dimensional scanning of the three-dimensional scene through the visual sensor to perform three-dimensional modeling of the three-dimensional scene based on the multi-frame image information of the visual sensor.
  • the scanning device interrupts scanning, it needs to perform a repositioning scan to confirm the starting point of the scanning device after rescanning, and to provide a basis for the scanning device to automatically stitch the 3D model.
  • the processor calculates the current spatial position information, semantic information, size and other characteristic information of the three-dimensional scene where the scanning device is located based on the current sensor information of multiple frames of the three-dimensional scene scanned by the sensor, thereby determining the current space of the three-dimensional scene. Semantic map.
  • the current spatial semantic map is used to represent various parts of the three-dimensional scene.
  • the three-dimensional scene is a bedroom, corresponding to various entities such as beds, chairs, tables, etc.
  • the processor can obtain the corresponding sensor information according to the sensor
  • the information recognizes the semantic information of the corresponding bed, chair, table, spatial position information and other characteristic information and generates the corresponding current spatial semantic map.
  • the current spatial semantic map includes multiple current spatial semantic objects, such as the current spatial semantic object representing the bed.
  • the corresponding current spatial semantic object includes the semantic information of the bed, the size information of the bed, the spatial position information of the bed, the three-dimensional point cloud information of the bed, texture information and other characteristic information, as well as one of multiple current spatial semantic objects.
  • the relationship between them can also be reflected in the current spatial semantic map, such as the spatial distance between different current spatial semantic objects.
  • the semantic information can be automatically obtained through machine learning or manually predefined. It is not limited here and can be set as needed.
  • V represents the various parameter sets that describe the subgraph, such as
  • the description information of the object includes information such as shape, appearance, material, text description, etc.
  • F represents a subgraph, for example, a chair, which can form its own graph structure and be expressed as a small subgraph.
  • E stands for edge, which represents the relationship between nodes between F.
  • the processor can be installed in the scanning device, or can be separated from the scanning device.
  • the processor and the scanning device can be connected by wire or wirelessly. There is no limit here, and they can be set as needed. .
  • the sensor is a sensor that can obtain the spatial position information of the scanning device, such as a visual sensor, radar, IMU, ultrasonic sensor, tof sensor, etc. Different sensors can be used to calculate the spatial position of the scanning device from different dimensions, making the spatial position of the scanning device more accurate. .
  • step 120 before the scanning device interrupts scanning, different sensors have scanned parts of the three-dimensional scene and formed multi-frame historical sensor information.
  • the processor accordingly performs calculations based on the multi-frame historical sensor information to determine that the three-dimensional scene has been scanned by the scanning device.
  • Part of the historical spatial location information, semantic information, size and other characteristic information is generated, and the corresponding historical spatial semantic map is generated and stored in the storage medium.
  • the generation method of the historical spatial semantic map is the same as the generation method of the current spatial semantic map, and will not be described in detail here.
  • the current spatial semantic map and the historical spatial semantic map need to be matched through sub-graph search to obtain the first matching degree between the current spatial semantic map and the historical spatial semantic map.
  • each object node of the current spatial semantic map can be quickly matched to improve the matching efficiency.
  • the subgraph search can use the random walk algorithm or Ullmann algorithm, or other algorithms. There is no limit here and can be set as needed.
  • the first matching degree is used to measure the matching between the current spatial semantic map and the historical spatial semantic map.
  • the current spatial semantic map includes k semantic objects such as chairs and tables.
  • the relevant information of the current spatial semantic map and the historical spatial semantic map constructed still have errors even if they are the same semantic object. Therefore, in some cases, when matching semantic objects, if the current spatial semantic map If the semantic object in the map has a certain degree of similarity with the semantic object of the historical spatial semantic map, the two semantic objects can be considered to be the same. For example, if the similarity is above 80%, the two semantic objects can be considered to be the same. Or, in order to improve the positioning accuracy, the similarity can be set above 90% before two semantic objects are considered to be the same.
  • the setting of similarity can be set as needed and is not limited here.
  • the first matching degree can be calculated through a simple ratio relationship; or the first matching degree can also be calculated through weighting to improve the matching accuracy; or the hash value of the local location is constructed according to the global characteristics and the local sensitive hash to form Fuzzy matching hash value, then fuzzy matching hash value, and then partial exact matching to obtain the first matching degree; or other calculation methods, which are not limited here and can be determined as needed.
  • the first matching degree is calculated by weighting
  • a semantic object with a similarity of 90% has a greater weight
  • a semantic object with a similarity of 80% has a smaller weight.
  • the object is the same as the historical space semantic object.
  • steps 140 and 150 after calculating the first matching degree, it is necessary to determine whether the first matching degree is greater than the first matching threshold to determine whether the scanning device is successfully relocated.
  • the first matching threshold is set as needed. Usually the first matching threshold is set to more than 70% to ensure that the current spatial semantic map and the historical spatial semantic map have a greater and more accurate match, which is beneficial to the scanning device after the relocation is successful. Accuracy of 3D reconstruction model.
  • the processor determines that the relocation of the scanning device in the three-dimensional scene is successful.
  • the processor determines the location of the unfinished three-dimensional model before interrupting the scanning based on the location of the scanning device. For the interrupted part, during the continued scanning process, the interrupted part will continue to be connected according to the scanning information of the scanning device and the corresponding three-dimensional model will be generated until the scanning is completed, and a three-dimensional model of the complete three-dimensional scene will be generated.
  • the amount of data can be effectively reduced, the processing load of the processor can be reduced, and the storage pressure of the storage medium can be reduced.
  • the subgraph search method can quickly match the current spatial semantic map and the historical spatial semantic map. , improve matching efficiency, thereby increasing relocation speed.
  • the processor determines the position of the scanning device through relocation, so that the processor can automatically continue the 3D model at the interrupted scanning part during the continued scanning modeling process, until the scanning is completed, and generate a 3D model of the complete 3D scene, without the need to manually continue the scanning. Part of the 3D model is spliced into the historical 3D model before continued scanning, which facilitates operation and improves scanning efficiency.
  • the spatial position of the scanning device can be calculated from different dimensions, making the spatial position of the scanning device more accurate and improving relocation accuracy, which is beneficial to the processor. According to the position of the scanning device, the interruption position of the unfinished three-dimensional model before the interruption of scanning is more accurately determined.
  • the multiple frames of current sensor information include multiple frames of current three-dimensional sensor information
  • the current spatial semantic map includes the current three-dimensional semantic map.
  • Step 110 further includes:
  • Step a01 Generate the current three-dimensional point cloud of the world coordinate system based on the current three-dimensional sensor information of multiple frames.
  • Step a02 Perform three-dimensional semantic segmentation on multiple frames of current three-dimensional sensor information based on the current three-dimensional point cloud, generate the current three-dimensional semantic map, and use the current three-dimensional semantic map as the current spatial semantic map.
  • the three-dimensional sensor refers to a sensor that can scan the three-dimensional information of the three-dimensional scene, such as a visual sensor, a laser radar, an ultrasonic sensor, etc.
  • the three-dimensional sensor can be one sensor or two or more different sensors. , there is no limit here, set it as needed.
  • the processor generates the current three-dimensional point cloud of the world coordinate system based on multiple frames of current three-dimensional sensor information to provide spatial position information. Then, the processor performs three-dimensional semantic segmentation calculations based on the current three-dimensional point cloud to generate the current three-dimensional semantic map.
  • Each three-dimensional semantic object in the current three-dimensional semantic map contains corresponding three-dimensional point cloud information.
  • the scanning equipment can also use three-dimensional sensors to perform three-dimensional scanning and three-dimensional model reconstruction at the same time.
  • the three-dimensional sensor is a visual sensor
  • the visual sensor scans the three-dimensional scene around the scanning device to obtain multiple frames of visual images from different perspectives.
  • the processor performs calculations based on the multiple frames of visual images, which can be obtained through beam adjustment or other algorithms.
  • the corresponding current three-dimensional visual point cloud is the current three-dimensional point cloud.
  • each frame of visual image is subjected to visual semantic recognition and segmentation through an image recognition algorithm.
  • the image recognition algorithm can be obtained through FCN, U-net, SegNet, DeepLab v1, RefineNet, PSPnet, Deeplab v2, v3 and other algorithms. Then the visual semantics of multiple frames of visual images are fused to obtain a three-dimensional visual semantic object.
  • three-dimensional semantic segmentation calculations can also be performed through PointNet, PointNet++, PointSIFT, SPG, 3P-RNN, pointwize and other algorithms to obtain corresponding three-dimensional visual semantic objects.
  • Each segmented three-dimensional visual semantic object has a corresponding three-dimensional visual point cloud, from which the current three-dimensional visual semantic map can be constructed.
  • each three-dimensional visual semantic object is a sub-map, and each sub-map Graphs have corresponding semantic names, spatial positions, shapes, sizes and other characteristic information.
  • the three-dimensional sensor can be a visual sensor that the scanning device itself has, or, in some embodiments, The three-dimensional sensor is an additional visual sensor, which is not limited here and can be set as needed.
  • the lidar scans the three-dimensional scene around the scanning device to obtain a multi-frame laser point cloud.
  • the processor calculates the corresponding current three-dimensional laser point cloud based on the multi-frame laser point cloud.
  • the current three-dimensional laser point cloud is
  • the laser point cloud is the current three-dimensional point cloud.
  • the current three-dimensional laser point cloud is semantically recognized and segmented through a three-dimensional semantic segmentation algorithm to obtain a three-dimensional laser semantic object.
  • Each segmented three-dimensional laser semantic object has a corresponding three-dimensional laser point cloud, which can be constructed
  • the current three-dimensional laser semantic map In the current three-dimensional laser semantic map, each three-dimensional laser semantic object is a sub-image, and each sub-image has corresponding semantic name, spatial position, shape, size and other characteristic information.
  • the ultrasonic sensor scans the three-dimensional scene around the scanning device to obtain multiple frames of ultrasonic images.
  • the processor calculates based on the multi-frame ultrasonic images to obtain the corresponding current three-dimensional ultrasonic point cloud.
  • the current three-dimensional ultrasonic point cloud The cloud is the current three-dimensional point cloud.
  • each frame of ultrasound image is semantically recognized and segmented through an image recognition algorithm.
  • the image recognition algorithm can be obtained through FCN, U-net, SegNet, DeepLab v1, RefineNet, PSPnet, Deeplab v2, v3 and other algorithms, and then The semantic features of multiple frames of ultrasound images are fused to obtain a three-dimensional ultrasound semantic object.
  • three-dimensional semantic segmentation calculations can also be performed through PointNet, PointNet++, PointSIFT, SPG, 3P-RNN, pointwize and other algorithms to obtain corresponding three-dimensional visual semantic objects.
  • Each segmented three-dimensional ultrasonic semantic object has a corresponding three-dimensional ultrasonic point cloud, from which the current three-dimensional ultrasonic semantic map can be constructed.
  • each three-dimensional ultrasonic semantic object is a sub-map, and each sub-map Graphs have corresponding semantic names, spatial positions, shapes, sizes and other characteristic information.
  • the current sensor information obtained by each sensor scanning the three-dimensional scene is calculated by the processor to obtain the corresponding three-dimensional semantic object and three-dimensional point cloud, and then the three-dimensional semantic objects obtained by the different sensors are Objects and 3D point clouds undergo feature information and Error integration is performed to fuse, the fused three-dimensional semantic objects and three-dimensional point clouds are obtained, and the corresponding current three-dimensional semantic map is constructed.
  • three-dimensional sensors include visual sensors and lidar.
  • the processor generates corresponding current three-dimensional visual point clouds based on multiple frames of current visual image information and performs semantic segmentation to generate the current visual semantic map.
  • the current visual semantic map includes multiple three-dimensional visual semantic objects. Multiple sub-images are formed, each sub-image has corresponding semantic name, spatial position, shape, size and other characteristic information; and the processor generates the corresponding current three-dimensional laser point cloud based on the current laser information of multiple frames and performs semantic segmentation to generate the current Laser semantic map.
  • the current laser semantic map includes multiple sub-images formed by multiple three-dimensional laser semantic objects. Each sub-image has corresponding semantic name, spatial position, shape, size and other characteristic information.
  • the current visual semantic map and the current laser semantic map are matched and fused based on the semantic name, spatial position, shape, size and other characteristic information, such as semantic name matching, spatial position matching, shape matching, size matching, where spatial position , shape, size and other parts involved in point cloud calculations require error-compatible calculations during fusion to fuse the three-dimensional point clouds generated by different sensors.
  • the processor By setting up a three-dimensional sensor, the processor directly calculates a more accurate three-dimensional point cloud of the three-dimensional scene, and constructs a more accurate current three-dimensional semantic map, which is conducive to more accurate relocation of the scanning equipment.
  • the three-dimensional sensor includes two or more different sensors, different sensors obtain the three-dimensional information of the three-dimensional scene from different dimensions, so that the processor can obtain a more accurate current three-dimensional point of the three-dimensional scene by fusing the information of different sensors. Cloud to further improve the accuracy of scanning device relocation.
  • the current three-dimensional sensor information of multiple frames includes current visual image information of multiple frames
  • the current three-dimensional semantic map includes the current visual semantic map
  • the current spatial semantic map is determined based on the current sensor information of multiple frames.
  • Step b01 Generate the current three-dimensional visual point cloud of the world coordinate system based on the current visual image information of multiple frames;
  • Step b02 Perform three-dimensional visual semantic segmentation on multiple frames of current visual image information based on the current three-dimensional visual point cloud, and generate a current visual semantic map.
  • step b01 for the case where the three-dimensional sensor includes a visual sensor, the visual sensor scans the three-dimensional scene around the scanning device to obtain multiple frames of visual images from different viewing angles.
  • the processor performs calculations based on the multiple frames of visual images and can use beam adjustment. method or other algorithms to obtain the corresponding current three-dimensional visual point cloud.
  • the current three-dimensional visual semantic map obtained through step b02 can effectively and intuitively represent the information of each object in the three-dimensional scene, such as the semantic name, spatial position, shape, size and other characteristic information of the corresponding object.
  • the visual sensor is used as a three-dimensional sensor to adapt to the visual sensor of the scanning device used for three-dimensional modeling scanning.
  • the visual sensor of the scanning device can be used as a three-dimensional sensor, thereby reducing the number and type of sensor settings and reducing the cost. Cost; in some embodiments, if additional visual sensors are provided, the number of visual sensors will be larger accordingly, and the accuracy of the three-dimensional modeling of the scanning device can be improved by combining data from multiple visual sensors, as well as the accuracy of the scanning device. Relocation accuracy.
  • step b02 further includes:
  • Step b021 Perform three-dimensional visual semantic segmentation on multiple frames of current visual image information through an image recognition algorithm to obtain a three-dimensional visual semantic object of multiple frames of current visual image information;
  • Step b022 Calculate the spatial position of the three-dimensional visual semantic object based on the current three-dimensional visual point cloud, and obtain the current spatial position information of the three-dimensional visual semantic object;
  • Step b023 Generate the current visual semantic map according to the characteristic information of the three-dimensional visual semantic object.
  • the characteristic information includes the current spatial position information.
  • step b021 multiple frames of current visual image information are subjected to three-dimensional visual semantic recognition and segmentation through an image recognition algorithm to obtain a three-dimensional semantic object.
  • the semantic name of the three-dimensional visual semantic object is set as needed, and is not limited here.
  • each frame of ultrasound image is semantically recognized and segmented through an image recognition algorithm.
  • the image recognition algorithm can be obtained through FCN, U-net, SegNet, DeepLab v1, RefineNet, PSPnet, Deeplab v2, v3 and other algorithms, and then The semantic features of multiple frames of ultrasound images are fused to obtain a three-dimensional ultrasound semantic object.
  • three-dimensional semantic segmentation calculations can also be performed through PointNet, PointNet++, PointSIFT, SPG, 3P-RNN, pointwize and other algorithms to obtain corresponding three-dimensional visual semantic objects.
  • step b022 the current three-dimensional visual point cloud and the three-dimensional visual semantic object obtained based on the multi-frame visual image are converted to the world coordinate system for calculation, and the current spatial position information of the corresponding three-dimensional visual semantic object is obtained.
  • the current spatial position information and Current three-dimensional visual point clouds can calculate the spatial position, shape, size and other characteristic information of three-dimensional visual semantic objects.
  • a semantic subgraph corresponding to a three-dimensional visual semantic object is generated based on characteristic information such as semantic name, spatial position, shape, size, etc., and there are multiple three-dimensional visual semantic objects in the three-dimensional scene, corresponding to multiple semantic subgraphs, so that The corresponding current visual semantic map can be generated.
  • the current three-dimensional sensor information of multiple frames includes current laser information of multiple frames
  • the current three-dimensional semantic map includes the current laser semantic map.
  • Step 110 further includes:
  • Step c01 Generate the current three-dimensional laser point cloud of the world coordinate system based on multiple frames of current laser information
  • Step c02 Perform three-dimensional laser semantic segmentation on multiple frames of current laser information based on the current three-dimensional laser point cloud, and generate a current laser semantic map.
  • step c01 for the case where the three-dimensional sensor includes a lidar, the lidar scans the three-dimensional scene around the scanning device to obtain a multi-frame laser point cloud, and the processor calculates based on the multi-frame laser point cloud to obtain the corresponding current three-dimensional laser point cloud.
  • step c02 when performing semantic segmentation, the current three-dimensional laser point cloud is semantically recognized and segmented through a three-dimensional semantic segmentation algorithm to obtain a three-dimensional laser semantic object.
  • the semantic name of the three-dimensional laser semantic object is set as needed and is not limited here.
  • three-dimensional laser semantic segmentation processing can be performed through algorithms such as Introduction, Motivation, and Overview to generate corresponding three-dimensional laser semantic objects.
  • the current three-dimensional laser point cloud and the three-dimensional laser semantic object are converted into the world coordinate system for calculation, and the current spatial position information of the corresponding three-dimensional laser semantic object is obtained. Based on the current spatial position information and the current three-dimensional laser point cloud, the three-dimensional object can be calculated.
  • the spatial position, shape, size and other characteristic information of the laser semantic object are then generated based on the semantic name, spatial position, shape, size and other characteristic information to generate a semantic sub-graph corresponding to the three-dimensional laser semantic object.
  • lidar is used as the three-dimensional sensor to match the laser used by the scanning device for three-dimensional modeling scanning. Radar compatible.
  • the lidar of the scanning device can be used as a three-dimensional sensor, thereby reducing the number and type of sensors and reducing costs; in some embodiments, if additional lidars are installed, the number of lidars will increase accordingly. More, the accuracy of the three-dimensional modeling of the scanning device and the accuracy of the relocation of the scanning device can be improved through the combination of multiple lidar data.
  • the multi-frame current sensor information also includes multi-frame current IMU information
  • step 110 further includes:
  • Step d01 Generate the current IMU space trajectory point cloud of the world coordinate system based on the current IMU information of multiple frames;
  • Step d02 Match the current IMU space trajectory point cloud and the current three-dimensional point cloud according to the spatial coordinate position, and obtain the second matching degree of the current IMU space trajectory point cloud and the current three-dimensional point cloud;
  • Step d03 Determine whether the second matching degree is greater than or equal to the second matching threshold
  • Step d04 If the second matching degree is greater than or equal to the second matching threshold, determine the current fusion point cloud;
  • Step d05 Perform three-dimensional fusion semantic segmentation on multiple frames of current three-dimensional sensor information and multiple frames of current IMU information based on the current fusion point cloud, generate the current fusion semantic map, and use the current fusion semantic map as the current spatial semantic map.
  • step d01 the IMU sensor is set on the scanning device to obtain the motion trajectory of the scanning device.
  • the processor obtains the current attitude angle and acceleration of the IMU sensor with respect to the scanning device, calculates and converts it into a world coordinate system through integration
  • the current IMU spatial trajectory point cloud where the current IMU spatial trajectory point cloud represents the current spatial motion trajectory of the scanning device.
  • the scanning device can perform a special trajectory movement before interrupting scanning. During the relocation process, the scanning device also performs a special trajectory movement to improve the accuracy of relocation.
  • special trajectories such as figure-8 shapes, or M-shapes or other trajectories with multiple turns are not limited here and can be set as needed.
  • step d02 since the current IMU space trajectory point cloud has spatial position information of different point features in the motion trajectory, and the current three-dimensional point cloud also has spatial position information of multiple point features, therefore, the current IMU space can be moved through the spatial coordinate position.
  • the trajectory point cloud and the current three-dimensional point cloud are fused. Among them, it is necessary to determine whether the current IMU space trajectory point cloud and the current three-dimensional point cloud match. Therefore, it is necessary to calculate the second matching degree of the current IMU space trajectory point cloud and the current three-dimensional point cloud. to determine if fusion is possible.
  • the calculation of the second matching degree is similar to the calculation of the first matching degree. It can be calculated through a simple ratio relationship, or the first matching degree can also be calculated through weighting to improve the matching accuracy, or other calculation methods will not be done here. Limit, determined as needed.
  • the processor determines whether the second matching degree is greater than or equal to the second matching threshold; if the processor determines that the second matching degree is greater than or equal to the second matching threshold, it means that the current IMU space trajectory point cloud and the current three-dimensional point Cloud matching can be used for fusion, in which fusion can be performed through neighborhood search to obtain the current fused point cloud.
  • the current fused point cloud is mainly based on the current three-dimensional point cloud, and the current IMU space trajectory point cloud is fused to the current three-dimensional point cloud. middle. Among them, the second matching threshold is set as needed.
  • the second matching threshold is set to more than 70% to ensure that the second matching degree of the current IMU space trajectory point cloud and the current three-dimensional point cloud is larger and more accurate, which is beneficial to the current situation. Fusing the accuracy of point clouds to more accurately position scanning devices.
  • the processor determines that the second matching degree is less than the second matching threshold, point cloud fusion will not be performed. At this time, the current three-dimensional point cloud will be retained, and the original current three-dimensional semantic map and historical spatial semantics will be used accordingly.
  • the map performs subgraph search matching to relocate the scanning device.
  • the processor can perform 3D semantic segmentation accordingly based on the multi-frame 3D sensor information to obtain the corresponding 3D semantic objects in the 3D scene, and then the processor calculates the current fused point cloud and 3D semantic objects in the world coordinate system. Obtain the corresponding three-dimensional fusion semantic object and the current spatial position information and object point cloud based on the three-dimensional fusion semantic object. Based on the current spatial position information and object point cloud of the three-dimensional fusion semantic object, the spatial position, shape, and shape of the three-dimensional fusion semantic object can be calculated. Characteristic information such as size, etc.
  • semantic subgraphs corresponding to three-dimensional fusion semantic objects based on characteristic information such as semantic names, spatial positions, shapes, sizes, etc., and there are multiple three-dimensional fusion semantic objects in a three-dimensional scene, corresponding to multiple semantic subgraphs. , so that the corresponding current three-dimensional fusion semantic map can be generated.
  • the three-dimensional sensor is used to sense the three-dimensional scene around the scanning device, and determine the position of the scanning device through three-dimensional recognition and matching of the three-dimensional scene, while the IMU sensor is used to sense the movement of the scanning device itself.
  • the trajectory determines the pose of the scanning device.
  • step d02 further includes:
  • Step d021 Determine m matching points whose distances between c trajectory points in the current IMU space trajectory point cloud and at least one candidate point in the current three-dimensional point cloud are less than or equal to the preset distance, m ⁇ c;
  • the matching of the current IMU space trajectory point cloud and the current three-dimensional point cloud is mainly the matching of all trajectory points of the current IMU space trajectory point cloud.
  • the current IMU space trajectory point cloud and the current three-dimensional point cloud are matched and calculated in the same world coordinate system. If at least one candidate point in the current three-dimensional point cloud is searched and one of the c trajectory points of the current IMU space trajectory point cloud is searched, If the distance between the points is less than or equal to the preset distance, it is considered that one of the trajectory points in the current IMU space trajectory point cloud is successfully matched.
  • the distance is a Euclidean distance.
  • the preset distance is set to 10 centimeters to reduce matching errors and improve matching accuracy.
  • the preset distance can also be set to 5 centimeters, 8 centimeters, 15 centimeters, or other distance values, which are not limited here and can be set as needed. Among them, the smaller the preset distance setting, the higher the matching accuracy.
  • the current spatial semantic map includes k current spatial semantic objects
  • the historical spatial semantic map includes h historical spatial semantic objects.
  • Step e01 Determine n current spatial semantic objects matching k current spatial semantic objects and h historical spatial semantic objects, n ⁇ k.
  • the current spatial semantic map includes k current spatial semantic objects
  • the historical spatial semantic map includes h historical spatial semantic objects
  • it is mainly to match k Matching of semantic objects in the current space. If n current spatial semantic objects match n historical spatial semantic objects, correspondingly, the first matching degree s1 n/k is determined through the ratio relationship.
  • FIG. 2 shows a flow chart of a three-dimensional continuous scanning method of a scanning device provided by an embodiment of the present invention.
  • the method is executed by a computing device.
  • the method includes the following steps:
  • Step 210 Determine whether the scanning device needs to continue scanning the three-dimensional scene.
  • Step 220 If it is necessary to continue scanning the three-dimensional scene, determine whether the scanning device is successfully relocated according to the above-mentioned continued scanning relocation method.
  • Step 230 If the relocation is successful, continue scanning the three-dimensional scene until the scanning is completed.
  • the scanning device is prone to interrupt the scanning process due to factors such as processor computing power, battery power, etc.
  • the processor determines that the scanning device cannot It is necessary to continue scanning the three-dimensional scene. If the scanning device has not completed scanning and the three-dimensional model has not been reconstructed, the processor determines that the scanning device needs to continue scanning the three-dimensional scene. At this time, the scanning device needs to be repositioned to determine the location of the scanning device.
  • the processor uses the continued scan to The relocation method performs relocation calculation on the scanning device to determine whether the scanning device is successfully relocated.
  • step 230 if the relocation is successful, correspondingly, the position of the scanning device can be determined, thereby determining the interruption boundary of the historical three-dimensional reconstruction model when the scanning device is interrupted.
  • the processor continues to complete based on the interruption boundary based on the continued scanning sensor information of the scanning device.
  • the 3D model reconstruction automatically continues the interrupted boundaries of the historical 3D reconstruction model before the scan, and automatically splices the 3D information generated after the scan to the historical 3D reconstruction model until the entire 3D model is reconstructed, which facilitates operation and improves scanning efficiency.
  • the processor calculates and generates the corresponding 3D point cloud of the continuation scan based on the sensor information of the continuation scan after the scanning device continues to scan, and matches the 3D point cloud of the continuation scan with the historical 3D point cloud of the historical 3D reconstruction model through point cloud registration.
  • the interruption boundary of the historical 3D reconstruction model is continued, so that the continuous scan reconstruction of the 3D model is automatically continued from the interruption boundary until the 3D model reconstruction is completed.
  • the ICP algorithm can be used to match the continued scanned 3D point cloud with the historical 3D point cloud of the historical 3D reconstruction model, so that the continued scanned 3D model and the historical 3D model are automatically spliced and aligned.
  • the scanning device is repositioned to determine the location of the scanning device that needs to continue scanning, thereby determining the interruption boundary of the historical three-dimensional reconstruction model when the scanning device is interrupted, and the processor is based on the continued scanning of the scanning device.
  • the sensor information continues to complete the 3D model reconstruction based on the interruption boundary to automatically continue the interruption boundary of the historical 3D reconstruction model before the scan, and automatically splice the 3D information generated after the continuation of the scan to the historical 3D reconstruction model until the entire 3D model reconstruction is completed, which is convenient operation to improve scanning efficiency.
  • Figure 3 shows a schematic structural diagram of a continuous scan relocation device provided by an embodiment of the present invention.
  • the device 300 includes:
  • the first determination module 310 is used to determine the current spatial semantic map based on multiple frames of current sensor information obtained by the scanning device in the three-dimensional scene.
  • the multiple frames of current sensor information are obtained from different sensors, and the current spatial semantic map has current spatial location information;
  • the first acquisition module 320 is used to obtain the historical spatial semantic map of the scanning device for the three-dimensional scene.
  • the historical spatial semantic map is determined based on multi-frame historical sensor information.
  • the multi-frame historical sensor information is determined by scanning.
  • the historical spatial semantic map has historical spatial location information obtained from different sensors before the device interrupts scanning;
  • the first matching module 330 is used to match the current spatial semantic map and the historical spatial semantic map through sub-graph search to obtain the first matching degree between the current spatial semantic map and the historical spatial semantic map;
  • the first judgment module 340 is used to judge whether the first matching degree is greater than or equal to the first matching threshold
  • the second determination module 350 is configured to determine that the scanning device is successfully repositioned in the three-dimensional scene if the first matching degree is greater than or equal to the first matching threshold.
  • the multiple frames of current sensor information include multiple frames of current three-dimensional sensor information
  • the current spatial semantic map includes the current three-dimensional semantic map
  • the first determination module 310 further includes:
  • the first generation unit is used to generate the current three-dimensional point cloud of the world coordinate system based on multiple frames of current three-dimensional sensor information;
  • the second generation unit is used to perform three-dimensional semantic segmentation on multiple frames of current three-dimensional sensor information based on the current three-dimensional point cloud, generate the current three-dimensional semantic map, and use the current three-dimensional semantic map as the current spatial semantic map.
  • the multiple frames of current three-dimensional sensor information include multiple frames of current visual image information
  • the current three-dimensional semantic map includes the current visual semantic map
  • the first determination module 310 further includes:
  • the third generation unit is used to generate the current three-dimensional visual point cloud of the world coordinate system based on the current visual image information of multiple frames;
  • the fourth generation unit is used to perform three-dimensional visual semantic segmentation on multiple frames of current visual image information based on the current three-dimensional visual point cloud, and generate the current visual semantic map.
  • three-dimensional visual semantic segmentation is performed on multiple frames of current visual image information based on the current three-dimensional visual point cloud to generate the current visual semantic map.
  • the fourth generation unit further includes:
  • the first acquisition unit is used to first perform three-dimensional visual semantic segmentation on multiple frames of current visual image information through an image recognition algorithm, and obtain a three-dimensional visual semantic object of multiple frames of current visual image information;
  • the first computing unit is used to calculate the spatial position of the three-dimensional visual semantic object based on the current three-dimensional visual point cloud, and obtain the current spatial position information of the three-dimensional visual semantic object;
  • the fifth generation unit is used to generate the current visual semantic map based on the characteristic information of the three-dimensional visual semantic object.
  • the characteristic information includes the current spatial position information.
  • the multiple frames of current three-dimensional sensor information include multiple frames of current laser information
  • the current three-dimensional semantic map includes the current laser semantic map
  • the first determination module 310 further includes:
  • the sixth generation unit is used to generate the current three-dimensional laser point cloud of the world coordinate system based on multiple frames of current laser information
  • the seventh generation unit is used to perform three-dimensional laser semantic segmentation on multiple frames of current laser information based on the current three-dimensional laser point cloud, and generate a current laser semantic map.
  • the multi-frame current sensor information also includes multi-frame current IMU information
  • the first determining module 310 further includes:
  • the eighth generation unit is used to generate the current IMU space trajectory point cloud of the world coordinate system based on the current IMU information of multiple frames;
  • the second acquisition unit is used to match the current IMU space trajectory point cloud and the current three-dimensional point cloud according to the spatial coordinate position, and obtain the second matching degree of the current IMU space trajectory point cloud and the current three-dimensional point cloud;
  • the first discriminating unit is used to determine whether the second matching degree is greater than or equal to the second matching threshold
  • the first determination unit is used to determine the current fusion point cloud if the second matching degree is greater than or equal to the second matching threshold;
  • the first fusion unit is used to perform three-dimensional fusion semantic segmentation on multiple frames of current three-dimensional sensor information and multiple frames of current IMU information based on the current fusion point cloud, generate the current fusion semantic map, and use the current fusion semantic map as the current spatial semantic map.
  • matching the current IMU space trajectory point cloud and the current three-dimensional point cloud according to the spatial coordinate position to obtain the second matching degree of the current IMU space trajectory point cloud and the current three-dimensional point cloud further includes:
  • the second calculation unit is used to determine m matching points whose distances between c trajectory points in the current IMU space trajectory point cloud and at least one candidate point in the current three-dimensional point cloud are less than or equal to the preset distance, m ⁇ c;
  • the current spatial semantic map includes k current spatial semantic objects
  • the historical spatial semantic map includes h historical spatial semantic objects
  • the first matching module 330 further includes:
  • the fourth computing unit is used to determine n current spatial semantic objects matching k current spatial semantic objects and h historical spatial semantic objects, n ⁇ k;
  • Figure 4 shows a schematic structural diagram of a three-dimensional continuous scanning device provided by an embodiment of the present invention.
  • the device 400 includes:
  • the second judgment module 410 is used to judge whether the scanning device needs to continue scanning the three-dimensional scene
  • the third judgment module 420 is used to judge whether the scanning device is successfully repositioned according to the above-mentioned continued scanning relocation method if it is necessary to continue scanning the three-dimensional scene;
  • the execution module 420 is used to complete the three-dimensional model reconstruction based on the continued scanning sensor information of the scanning device if the relocation is successful.
  • Figure 5 shows a schematic structural diagram of a computing device provided by an embodiment of the present invention. Specific embodiments of the present invention do not limit the specific implementation of the computing device.
  • the computing device may include: a processor (processor) 502, a communications interface (Communications Interface) 504, a memory (memory) 506, and a communication bus 508.
  • processor processor
  • communications interface Communication Interface
  • memory memory
  • the processor 502, the communication interface 504, and the memory 506 complete communication with each other through the communication bus 508.
  • the communication interface 504 is used to communicate with network elements of other devices such as clients or other servers.
  • the processor 502 is configured to execute the program 510. Specifically, the processor 502 may execute the relevant steps described above in the embodiment of the method for continuing scanning and relocation.
  • program 510 may include program code including computer-executable instructions.
  • the processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.
  • the one or more processors included in the computing device may be the same type of processor, such as one or more CPUs; or they may be different types of processors, such as one or more CPUs and one or more ASICs.
  • Memory 506 is used to store programs 510.
  • Memory 506 may include high-speed RAM memory, It may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
  • Embodiments of the present invention also provide a computer-readable storage medium, in which at least one executable instruction is stored, and the executable instruction executes any of the above-mentioned operations of the continuation scan and relocation method when running.
  • modules in the devices in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment.
  • the modules or units or components in the embodiments may be combined into one module or unit or component, and they may be divided into multiple sub-modules or sub-units or sub-components.
  • All features disclosed in this specification including accompanying claims, abstract and drawings) and any method so disclosed may be employed in any combination, except that at least some of such features and/or processes or units are mutually exclusive. All processes or units of the equipment are combined.
  • Each feature disclosed in this specification may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Graphics (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Health & Medical Sciences (AREA)
  • Geometry (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Architecture (AREA)
  • Length Measuring Devices By Optical Means (AREA)

Abstract

本发明实施例涉及三维扫描技术领域,具体涉及一种续扫重定位方法、装置、设备、存储介质及三维续扫方法,该续扫重定位方法包括:根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,获取扫描设备针对三维场景的历史空间语义图谱,将当前空间语义图谱与历史空间语义图谱通过子图搜索方式进行匹配,得到当前空间语义图谱与历史空间语义图谱的第一匹配度;若第一匹配度大于或等于第一匹配阈值,确定扫描设备在三维场景中的重定位成功。本发明实施例通过子图搜索方式,提高匹配效率,提高重定位速度,处理器通过重定位方式确定扫描设备的位置,使得处理器可以自动接续三维模型中断扫描部位,方便操作,提高扫描效率。

Description

续扫重定位方法、装置、设备、存储介质及三维续扫方法 技术领域
本发明实施例涉及三维扫描技术领域,具体涉及一种续扫重定位方法、装置、设备、存储介质及三维续扫方法。
背景技术
现有的手持式、背包式等移动激光扫描设备正越来越广泛的应用在三维重建领域,然而受限于处理器运算能力、电池电量等因素,在某些复杂的场景下这些移动扫描设备容易中断扫描进程,移动扫描设备重新启动扫描后,往往需要多次扫描,再通过人工合成等方式才能最终生成三维模型,操作麻烦,且效率较低。
因此,有必要提供一种续扫重定位方法、装置、计算设备、存储介质及三维续扫方法,以克服上述问题。
发明内容
鉴于上述问题,本发明实施例提供了一种续扫重定位方法、装置、计算设备、存储介质及三维续扫方法,用于解决现有技术中存在的问题。
根据本发明实施例的第一方面,提供了一种续扫重定位方法,所述方法包括:
根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,多帧所述当前传感器信息由不同传感器得到,所述当前空间语义图谱具有当前空间位置信息;
获取所述扫描设备针对所述三维场景的历史空间语义图谱,所述历史空间语义图谱根据多帧历史传感器信息确定,多帧所述历史传感器信息由所述扫描设备中断扫描前的不同所述传感器得到,所述历史空间语义图谱具有历史空间位置信息;
将所述当前空间语义图谱与所述历史空间语义图谱通过子图搜索方式进行匹配,得到所述当前空间语义图谱与所述历史空间语义图谱的第一匹配度;
判断所述第一匹配度是否大于或等于第一匹配阈值;
若所述第一匹配度大于或等于所述第一匹配阈值,确定所述扫描设备在所述三维场景中的重定位成功。
在一些实施例中,多帧所述当前传感器信息包括多帧当前三维传感器信息,所述当前空间语义图谱包括当前三维语义图谱,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
根据多帧所述当前三维传感器信息生成世界坐标系的当前三维点云;
根据所述当前三维点云对多帧所述当前三维传感器信息进行三维语义分割,生成所述当前三维语义图谱,将所述当前三维语义图谱作为所述当前空间语义 图谱。
在一些实施例中,多帧所述当前三维传感器信息包括多帧当前视觉图像信息,所述当前三维语义图谱包括当前视觉语义图谱,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
根据多帧所述当前视觉图像信息生成所述世界坐标系的当前三维视觉点云;
根据所述当前三维视觉点云对多帧所述当前视觉图像信息进行三维视觉语义分割,生成所述当前视觉语义图谱。
在一些实施例中,所述根据所述当前三维视觉点云对多帧所述当前视觉图像信息进行三维视觉语义分割,生成所述当前视觉语义图谱,进一步包括:
通过图像识别算法对多帧所述当前视觉图像信息进行三维视觉语义分割,得到多帧所述当前视觉图像信息的三维视觉语义对象;
根据所述当前三维视觉点云对所述三维视觉语义对象进行空间位置计算,得到所述三维视觉语义对象的所述当前空间位置信息;
根据所述三维视觉语义对象的特性信息,生成所述当前视觉语义图谱,所述特性信息包括所述当前空间位置信息。
在一些实施例中,多帧所述当前三维传感器信息包括多帧当前激光信息,所述当前三维语义图谱包括当前激光语义图谱,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
根据多帧所述当前激光信息生成所述世界坐标系的当前三维激光点云;
根据所述当前三维激光点云对多帧所述当前激光信息进行三维激光语义分割,生成所述当前激光语义图谱。
在一些实施例中,多帧所述当前传感器信息还包括多帧当前IMU信息,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
根据多帧所述当前IMU信息生成所述世界坐标系的当前IMU空间轨迹点云;
将所述当前IMU空间轨迹点云和所述当前三维点云根据空间坐标位置进行匹配,得到所述当前IMU空间轨迹点云和所述当前三维点云的第二匹配度;
判断所述第二匹配度是否大于或等于第二匹配阈值;
若所述第二匹配度大于或等于所述第二匹配阈值,确定当前融合点云;
根据所述当前融合点云对多帧所述当前三维传感器信息和多帧所述当前IMU信息进行三维融合语义分割,生成当前融合语义图谱,将所述当前融合语义图谱作为所述当前空间语义图谱。
在一些实施例中,所述将所述当前IMU空间轨迹点云和所述当前三维点云根据空间坐标位置进行匹配,得到所述当前IMU空间轨迹点云和所述当前三维点云的第二匹配度,进一步包括:
确定所述当前IMU空间轨迹点云中c个轨迹点与所述当前三维点云中至少一候选点的距离小于或等于预设距离的m个匹配点,m≤c;
计算所述第二匹配度s2=m/c。
在一些实施例中,所述当前空间语义图谱包括k个当前空间语义对象,所 述历史空间语义图谱包括h个历史空间语义对象,所述将所述当前空间语义图谱与所述历史空间语义图谱通过子图搜索方式进行匹配,得到所述当前空间语义图谱与所述历史空间语义图谱的第一匹配度,进一步包括:
确定k个所述当前空间语义对象与h个所述历史空间语义对象匹配的n个所述当前空间语义对象,n≤k;
计算所述第一匹配度s1=n/k。
根据本发明实施例的第二方面,提供了一种三维续扫方法,所述方法包括:
判断扫描设备是否需要对三维场景进行续扫;
若需要对所述三维场景进行续扫,根据上述任一项所述的续扫重定位方法判断所述扫描设备是否重定位成功;
若重定位成功,根据所述扫描设备的续扫传感器信息完成三维模型重建。
根据本发明实施例的第三方面,提供了一种续扫重定位装置,所述装置包括:
第一确定模块,用于根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,多帧所述当前传感器信息由不同传感器得到,所述当前空间语义图谱具有当前空间位置信息;
第一获取模块,用于获取所述扫描设备针对所述三维场景的历史空间语义图谱,所述历史空间语义图谱根据多帧历史传感器信息确定,多帧所述历史传感器信息由所述扫描设备中断扫描前的不同所述传感器得到,所述历史空间语义图谱具有历史空间位置信息;
第一匹配模块,用于将所述当前空间语义图谱与所述历史空间语义图谱通过子图搜索方式进行匹配,得到所述当前空间语义图谱与所述历史空间语义图谱的第一匹配度;
第一判断模块,用于判断所述第一匹配度是否大于或等于第一匹配阈值;
第二确定模块,用于若所述第一匹配度大于或等于所述第一匹配阈值,确定所述扫描设备在所述三维场景中的重定位成功。
根据本发明实施例的第四方面,提供了一种计算设备,包括:处理器、存储器、通信接口和通信总线,所述处理器、所述存储器和所述通信接口通过所述通信总线完成相互间的通信;
所述存储器用于存放至少一可执行指令,所述可执行指令使所述处理器执行如上述任一项所述的续扫重定位方法的操作。
根据本发明实施例的第五方面,提供了一种计算机可读存储介质,所述存储介质中存储有至少一可执行指令,所述可执行指令在运行时执行如上述任一项所述的续扫重定位方法的操作。
本发明实施例通过确定当前空间语义图谱和历史空间语义图谱,可以有效减少数据量,降低处理器的处理量,并且可以减轻存储介质的存储压力,子图搜索方式可以快速匹配当前空间语义图谱和历史空间语义图谱,提高匹配效率,从而提高重定位速度。处理器通过重定位方式确定扫描设备的位置,使得处理器可以在续扫建模过程中自动接续三维模型中断扫描部位,直至扫描完成后, 生成完整三维场景的三维模型,无需手动将续扫后的部分三维模型拼接至续扫前的历史三维模型中,方便操作,提高扫描效率。
此外,通过不同传感器获取多帧当前传感器信息以及多帧历史传感器信息,可以从不同维度对扫描设备的空间位置计算,使得扫描设备的空间位置更加准确,提高重定位准确性,从而有利于处理器根据扫描设备所处位置更加准确确定中断扫描前未完成的三维模型的中断部位。
上述说明仅是本发明实施例技术方案的概述,为了能够更清楚了解本发明实施例的技术手段,而可依照说明书的内容予以实施,并且为了让本发明实施例的上述和其它目的、特征和优点能够更明显易懂,以下特举本发明的具体实施方式。
附图说明
附图仅用于示出实施方式,而并不认为是对本发明的限制。而且在整个附图中,用相同的参考符号表示相同的部件。在附图中:
图1示出了本发明实施例提供的续扫重定位方法的流程示意图;
图2示出了本发明一些实施例提供的三维续扫方法的流程示意图;
图3示出了本发明实施例提供的续扫重定位装置的结构示意图;
图4示出了本发明实施例提供的三维续扫装置的结构示意图;
图5示出了本发明一些实施例提供的计算设备的结构示意图。
具体实施方式
下面将参照附图更详细地描述本发明的示例性实施例。虽然附图中显示了本发明的示例性实施例,然而应当理解,可以以各种形式实现本发明而不应被这里阐述的实施例所限制。
针对现有移动扫描设备续扫麻烦以及效率低的问题,发明人发现,现有移动扫描设备容易由于处理器运算能力、电池电量等因素而中断扫描进程,在移动扫描设备重启扫描后,移动扫描设备不能重定位,无法确认重新扫描后移动扫描设备的位置,无法通过移动扫描设备自动将重新扫描的场景生成三维模型,需要手动选择位置作为重新扫描的起点,经过多次扫描,以及人工合成等方式生成三维模型,操作麻烦,且效率较低。
发明人提供了一种续扫重定位方法,通过在扫描设备上设置不同的传感器,获取多帧不同的当前传感器信息,通过多帧当前传感器信息计算扫描设备的当前空间语义图谱,将扫描设备中断前由多帧历史传感器信息确定的历史空间语义图谱与当前空间语义图谱进行匹配,以判断当前扫描设备所处位置,若重定位成功,则扫描设备可以根据重定位所在位置,继续扫描,并且在重新扫描的过程中,根据重定位位置,自动接续续扫前的历史三维重建模型,将续扫后的三维信息自动拼接至历史三维重建模型直至整个三维模型完整成型。处理器通过重定位方式确定扫描设备的位置,以及处理器在续扫建模过程中自动接续三维模型的重建,无需手动将续扫后的部分三维模型拼接至续扫前的历史三维模 型中,方便操作,提高扫描效率。此外通过语义图谱的子图搜索进行重定位的方式,能够快速定位,提高重定位效率,通过不同传感器配合定位,有利于提高定位准确性。
图1示出了本发明实施例提供的续扫重定位方法的流程图,该方法由计算设备执行,该计算设备可以是包括一个或多个处理器的计算设备,该处理器可能是中央处理器CPU,或者是特定集成电路ASIC(Application Specific Integrated Circuit),或者是被配置成实施本发明实施例的一个或多个集成电路,在此不做限定。计算设备包括的一个或多个处理器,可以是同一类型的处理器,如一个或多个CPU;也可以是不同类型的处理器,如一个或多个CPU以及一个或多个ASIC,在此不做限定。
如图1所示,该方法包括以下步骤:
步骤110:根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,多帧当前传感器信息由不同传感器得到,当前空间语义图谱具有当前空间位置信息。
步骤120:获取扫描设备针对三维场景的历史空间语义图谱,历史空间语义图谱根据多帧历史传感器信息确定,多帧历史传感器信息由扫描设备中断扫描前的不同传感器得到,历史空间语义图谱具有历史空间位置信息。
步骤130:将当前空间语义图谱与历史空间语义图谱通过子图搜索方式进行匹配,得到当前空间语义图谱与历史空间语义图谱的第一匹配度。
步骤140:判断第一匹配度是否大于或等于第一匹配阈值。
步骤150:若第一匹配度大于或等于第一匹配阈值,确定扫描设备在三维场景中的重定位成功。
其中,步骤110中,其中,扫描设备通过视觉传感器对三维场景进行三维扫描,以根据视觉传感器的多帧图像信息对三维场景进行三维建模。扫描设备在中断扫描后,需要进行重定位扫描,以确认重新扫描后扫描设备的起点,以及提供扫描设备进行自动拼接三维模型的基础。进行重定位的过程中,处理器根据传感器扫描的三维场景的多帧当前传感器信息,计算扫描设备所处三维场景的当前空间位置信息、语义信息、尺寸等特性信息,从而确定三维场景的当前空间语义图谱。该当前空间语义图谱用于表示三维场景各个部分,例如三维场景为卧室,对应的具有床、椅子、桌子等各个实体,扫描设备的传感器扫描后,可以获取到对应的传感器信息,处理器根据传感器信息识别到对应的床、椅子、桌子的语义信息以及空间位置信息等特性信息并生成对应的当前空间语义图谱,该当前空间语义图谱包括多个当前空间语义对象,例如代表床的当前空间语义对象,此时,对应的当前空间语义对象包括表示为床的语义信息、床的尺寸信息、床的空间位置信息、床的三维点云信息、纹理信息等特性信息,以及多个当前空间语义对象之间的关系也可以在当前空间语义图谱中体现,例如不同当前空间语义对象之间的空间距离。其中,语义信息可以通过机器学习方式自动获取或者人工预定义方式获取,在此不做限定,根据需要设置。
当前空间语义图谱,其中,V表示对子图进行描述的各种参数集合,例如 物体的描述信息,包括形状、外观、材质、文字描述等信息;F表示一个子图,例如,一个椅子,有可以形成自己的一个图结构,表示成一个小的子图。E即edge,表示F之间节点之间的关系。
其中,处理器可以设于扫描设备中,也可以与扫描设备分离,当处理器与扫描设备分离设置时,处理器与扫描设备可以有线连接也可以无线连接,在此不做限定,根据需要设置。
传感器是能够获取扫描设备空间位置信息的传感器,例如视觉传感器、雷达、IMU、超声波传感器、tof传感器等,采用不同的传感器可以从不同维度对扫描设备空间位置计算,使得扫描设备的空间位置更加准确。
步骤120中,在扫描设备中断扫描前,不同传感器已扫描到三维场景的部分场景并形成多帧历史传感器信息,处理器相应根据多帧历史传感器信息进行计算,确定三维场景中已被扫描设备扫描部分的历史空间位置信息、语义信息、尺寸等特性信息,并生成对应的历史空间语义图谱存储于存储介质中。历史空间语义图谱的生成方式与当前空间语义图谱的生成方式相同,在此不做赘述。
步骤130中,为确定扫描设备所处三维场景的位置,需要将当前空间语义图谱与历史空间图谱通过子图搜索方式进行匹配,得到当前空间语义图谱与历史空间语义图谱的第一匹配度。通过子图搜索方式可以快速匹配当前空间语义图谱的各个对象节点,提高匹配效率,子图搜索可以采用随机游走算法或者Ullmann algorithm算法,或者其他算法,在此不做限定,根据需要设置。第一匹配度用于衡量当前空间语义图谱与历史空间语义图谱的匹配情况,例如,当前空间语义图谱包括椅子、桌子等k个语义对象,每一语义对象具有对应的语义信息、空间位置信息、尺寸信息、三维点云信息等特性信息,假设在历史空间语义图谱中匹配到n个对应的语义对象,则通过比值关系计算对应得到第一匹配度为s1=n/k。
由于扫描误差和计算误差,所构建的当前空间语义图谱和历史空间语义图谱的相关信息即使是同一语义对象也还是有误差,因此,在一些情况下,在进行语义对象匹配时,若当前空间语义图谱中的语义对象与历史空间语义图谱的语义对象具有一定的相似度即可认为两个语义对象时相同的,例如该相似度在80%以上,则可以认为两个语义对象相同。或者,为提高定位准确度,可以将相似度设在90%以上,才认为两个语义对象相同。相似度的设置根据需要设置,在此不做限定。
在一些情况下,第一匹配度可以通过简单的比值关系计算;或者第一匹配度也可以通过加权计算,以提高匹配准确度;或者根据全局特征按照局部敏感hash构建局部位置的hash值,形成模糊匹配的hash值,然后模糊匹配hash值,然后再局部进行精确匹配,得到第一匹配度;或者其他计算方式,在此不做限定,根据需要确定。
针对通过加权计算第一匹配度的情况,例如,相似度在90%的语义对象的权重更大,相似度在80%的语义对象的权重较小。作为举例,当前空间语义图谱中共有k个当前空间语义对象,历史空间语义图谱中共有h个历史空间语义 对象,其中n1个当前空间语义对象与n1个历史空间语义对象具有90%以上的相似度,n2个当前空间语义对象与n2个历史空间语义对象具有80%-90%的相似度,此时可计算的第一匹配度s1=w1*n1/k+w2*n2/k,w1>w2,且w1+w2=1;在一些情况,若认为n3个70%-80%相似度的当前空间语义对象和历史空间语义对象相同,相应的,第一匹配度s1=w1*n1/k+w2*n2/k+w3*n3/k,w1>w2>w3,且w1+w2+w3=1。
步骤140和步骤150中,在计算第一匹配度后,需要判断第一匹配度是否大于第一匹配阈值,以判断扫描设备是否重定位成功。第一匹配阈值的设置根据需要设置,通常第一匹配阈值设置为70%以上,以保证当前空间语义图谱与历史空间语义图谱的匹配度较大以及较准确,有利于扫描设备在重定位成功后三维重建模型的准确度。其中,若第一匹配度大于或等于第一匹配阈值,处理器确定扫描设备在三维场景中的重定位成功,相应的,处理器根据扫描设备所处位置确定中断扫描前未完成的三维模型的中断部位,在续扫过程中,根据扫描设备的扫描信息继续将中断部位接续并生成对应的三维模型,直至扫描完成后,生成完整三维场景的三维模型。
通过确定当前空间语义图谱和历史空间语义图谱,可以有效减少数据量,降低处理器的处理量,并且可以减轻存储介质的存储压力,子图搜索方式可以快速匹配当前空间语义图谱和历史空间语义图谱,提高匹配效率,从而提高重定位速度。处理器通过重定位方式确定扫描设备的位置,使得处理器可以在续扫建模过程中自动接续三维模型中断扫描部位,直至扫描完成后,生成完整三维场景的三维模型,无需手动将续扫后的部分三维模型拼接至续扫前的历史三维模型中,方便操作,提高扫描效率。
此外,通过不同传感器获取多帧当前传感器信息以及多帧历史传感器信息,可以从不同维度对扫描设备的空间位置计算,使得扫描设备的空间位置更加准确,提高重定位准确性,从而有利于处理器根据扫描设备所处位置更加准确确定中断扫描前未完成的三维模型的中断部位。
在一些实施例中,多帧当前传感器信息包括多帧当前三维传感器信息,当前空间语义图谱包括当前三维语义图谱,步骤110进一步包括:
步骤a01:根据多帧当前三维传感器信息生成世界坐标系的当前三维点云。
步骤a02:根据当前三维点云对多帧当前三维传感器信息进行三维语义分割,生成当前三维语义图谱,将当前三维语义图谱作为当前空间语义图谱。
其中,该步骤a01至步骤a02中,三维传感器表示能够扫描三维场景的三维信息的传感器,例如视觉传感器、激光雷达、超声波传感器等,该三维传感器可以是一个传感器,也可以是两个以上不同传感器,在此不做限定,根据需要设置。其中,处理器根据多帧当前三维传感器信息生成世界坐标系的当前三维点云,以提供空间位置信息。而后,处理器根据当前三维点云进行三维语义分割计算,生成当前三维语义图谱,该当前三维语义图谱中的各个三维语义对象包含对应的三维点云信息。需要说明的是,扫描设备还可以同时利用三维传感器进行三维扫描以及三维模型重建。
在三维传感器为视觉传感器的情况下,视觉传感器对扫描设备周围的三维场景进行扫描得到多帧不同视角的视觉图像,处理器根据多帧视觉图像进行计算,可以通过光束平差法或者其他算法得到对应的当前三维视觉点云,该当前三维视觉点云即是当前三维点云。在进行语义分割时,通过图像识别算法对每一帧视觉图像进行视觉语义识别分割,图像识别算法可以通过FCN、U-net、SegNet、DeepLab v1、RefineNet、PSPnet、Deeplab v2、v3等算法得到,而后融合多帧视觉图像的视觉语义得到三维视觉语义对象。或者在一些实施例中,也可以通过PointNet、PointNet++、PointSIFT、SPG、3P-RNN、pointwize等算法进行三维语义分割计算,得到对应的三维视觉语义对象。其中每一分割得到的三维视觉语义对象具有对应的三维视觉点云,由此可以构建当前三维视觉语义图谱,该当前三维视觉语义图谱中,每一三维视觉语义对象为一个子图,每一子图有对应的语义名称、空间位置、形状、尺寸等特性信息。其中,在一些实施例中,若扫描设备本身具有视觉传感器以扫描三维场景进行三维模型建模,在此情况下,三维传感器可以是扫描设备本身具有的视觉传感器,或者,在一些实施例中,三维传感器是另外设置的视觉传感器,在此不做限定,根据需要设置。
在三维传感器为激光雷达的情况下,激光雷达对扫描设备周围的三维场景进行扫描得到多帧激光点云,处理器根据多帧激光点云进行计算得到对应的当前三维激光点云,该当前三维激光点云即是当前三维点云。在进行语义分割时,通过三维语义分割算法对当前三维激光点云进行语义识别分割,得到三维激光语义对象,其中每一分割得到的三维激光语义对象具有对应的三维激光点云,由此可以构建当前三维激光语义图谱,该当前三维激光语义图谱中,每一三维激光语义对象为一个子图,每一子图有对应的语义名称、空间位置、形状、尺寸等特性信息。
在三维传感器为超声波传感器的情况下,超声波传感器对扫描设备周围的三维场景进行扫描得到多帧超声波图像,处理器根据多帧超声波图像进行计算得到对应的当前三维超声波点云,该当前三维超声波点云即是当前三维点云。在进行语义分割时,通过图像识别算法对每一帧超声波图像进行语义识别分割,图像识别算法可以通过FCN、U-net、SegNet、DeepLab v1、RefineNet、PSPnet、Deeplab v2、v3等算法得到,而后融合多帧超声波图像的语义特征得到三维超声波语义对象。或者在一些实施例中,也可以通过PointNet、PointNet++、PointSIFT、SPG、3P-RNN、pointwize等算法进行三维语义分割计算,得到对应的三维视觉语义对象。其中每一分割得到的三维超声波语义对象具有对应的三维超声波点云,由此可以构建当前三维超声波语义图谱,该当前三维超声波语义图谱中,每一三维超声波语义对象为一个子图,每一子图有对应的语义名称、空间位置、形状、尺寸等特性信息。
在三维传感器包括两个以上不同传感器的情况下,每一传感器扫描三维场景所得到的当前传感器信息经过处理器计算后得到对应的三维语义对象和三维点云,而后将不同的传感器得到的三维语义对象和三维点云经过特性信息以及 误差整合进行融合,得到融合后的三维语义对象和三维点云,构建对应的当前三维语义图谱。
例如,三维传感器包括视觉传感器和激光雷达,处理器根据多帧当前视觉图像信息生成对应的当前三维视觉点云以及进行语义分割,生成当前视觉语义图谱,当前视觉语义图谱包括多个三维视觉语义对象形成的多个子图,每一子图有对应的语义名称、空间位置、形状、尺寸等特性信息;以及处理器根据多帧当前激光信息生成对应的当前三维激光点云以及进行语义分割,生成当前激光语义图谱,当前激光语义图谱包括多个三维激光语义对象形成的多个子图,每一子图有对应的语义名称、空间位置、形状、尺寸等特性信息。在进行融合时,根据语义名称、空间位置、形状、尺寸等特性信息将当前视觉语义图谱和当前激光语义图谱匹配融合,如语义名称匹配,空间位置匹配、形状匹配、尺寸匹配,其中,空间位置、形状、尺寸等涉及点云计算的部分在融合时需要进行误差兼容计算,以将不同传感器生成的三维点云融合。
通过设置三维传感器,处理器直接计算得到三维场景较为准确的三维点云,构建较为准确的当前三维语义图谱,有利于扫描设备重定位更为准确。此外,在三维传感器包括两个以上不同传感器的情况下,不同的传感器从不同的维度获得三维场景的三维信息,从而使得处理器融合不同传感器信息得到的更为准确的针对三维场景的当前三维点云,进一步提高扫描设备重定位准确度。
在一些实施例中,多帧当前三维传感器信息包括多帧当前视觉图像信息,当前三维语义图谱包括当前视觉语义图谱,根据多帧当前传感器信息确定当前空间语义图谱,步骤110进一步包括:
步骤b01:根据多帧当前视觉图像信息生成世界坐标系的当前三维视觉点云;
步骤b02:根据当前三维视觉点云对多帧当前视觉图像信息进行三维视觉语义分割,生成当前视觉语义图谱。
步骤b01中,针对三维传感器包括视觉传感器的情况,其中,视觉传感器对扫描设备周围的三维场景进行扫描得到多帧不同视角的视觉图像,处理器根据多帧视觉图像进行计算,可以通过光束平差法或者其他算法得到对应的当前三维视觉点云。通过步骤b02得到的当前三维视觉语义图谱,可以有效直观表示三维场景中各个物体的信息,例如对应物体的语义名称、空间位置、形状、尺寸等特性信息。
采用视觉传感器作为三维传感器,以与扫描设备用于三维建模扫描的视觉传感器相适应,在一些实施例中,可以将扫描设备的视觉传感器作为三维传感器,从而减少传感器数量以及种类的设置,降低成本;在一些实施例中,若额外设置视觉传感器,相应的,视觉传感器的数量则会较多,可以通过多个视觉传感器的数据结合,提高扫描设备的三维建模的准确性,以及扫描设备重定位准确性。
在一些实施例中,步骤b02进一步包括:
步骤b021:通过图像识别算法对多帧当前视觉图像信息进行三维视觉语义分割,得到多帧当前视觉图像信息的三维视觉语义对象;
步骤b022:根据当前三维视觉点云对三维视觉语义对象进行空间位置计算,得到三维视觉语义对象的当前空间位置信息;
步骤b023:根据三维视觉语义对象的特性信息,生成当前视觉语义图谱,特性信息包括当前空间位置信息。
步骤b021中,通过图像识别算法对多帧当前视觉图像信息进行三维视觉语义识别分割,得到三维语义对象,在此过程中,该三维视觉语义对象的语义名称根据需要设置,在此不做限定。在进行语义分割时,通过图像识别算法对每一帧超声波图像进行语义识别分割,图像识别算法可以通过FCN、U-net、SegNet、DeepLab v1、RefineNet、PSPnet、Deeplab v2、v3等算法得到,而后融合多帧超声波图像的语义特征得到三维超声波语义对象。或者在一些实施例中,也可以通过PointNet、PointNet++、PointSIFT、SPG、3P-RNN、pointwize等算法进行三维语义分割计算,得到对应的三维视觉语义对象。
步骤b022中,基于多帧视觉图像得到的当前三维视觉点云以及三维视觉语义对象均转换到世界坐标系中计算,得到对应的三维视觉语义对象的当前空间位置信息,根据该当前空间位置信息以及当前三维视觉点云可以计算得到三维视觉语义对象的空间位置、形状、尺寸等特性信息。
步骤b023中,根据语义名称、空间位置、形状、尺寸等特性信息生成对应三维视觉语义对象的语义子图,而三维场景中具有多个三维视觉语义对象,对应的具有多个语义子图,从而可以生成对应的当前视觉语义图谱。
在一些实施例中,多帧当前三维传感器信息包括多帧当前激光信息,当前三维语义图谱包括当前激光语义图谱,步骤110进一步包括:
步骤c01:根据多帧当前激光信息生成世界坐标系的当前三维激光点云;
步骤c02:根据当前三维激光点云对多帧当前激光信息进行三维激光语义分割,生成当前激光语义图谱。
步骤c01中,针对三维传感器包括激光雷达的情况,激光雷达对扫描设备周围的三维场景进行扫描得到多帧激光点云,处理器根据多帧激光点云进行计算得到对应的当前三维激光点云。
步骤c02中,在进行语义分割时,通过三维语义分割算法对当前三维激光点云进行语义识别分割,得到三维激光语义对象,该三维激光语义对象的语义名称根据需要设置,在此不做限定。在一些实施例中,可以通过Introduction、Motivation、Overview等算法进行三维激光语义分割处理,生成对应的三维激光语义对象。而后将当前三维激光点云以及三维激光语义对象均转换到世界坐标系中计算,得到对应的三维激光语义对象的当前空间位置信息,根据该当前空间位置信息以及当前三维激光点云可以计算得到三维激光语义对象的空间位置、形状、尺寸等特性信息,而后根据语义名称、空间位置、形状、尺寸等特性信息生成对应三维激光语义对象的语义子图,而三维场景中具有多个三维激光语义对象,对应的具有多个语义子图,从而可以生成对应的当前激光语义图谱。
在一些实施例中,若扫描设备用于三维建模扫描的传感器为激光雷达,对应的,采用激光雷达作为三维传感器,以与扫描设备用于三维建模扫描的激光 雷达相适应。在一些实施例中,可以将扫描设备的激光雷达作为三维传感器,从而减少传感器数量以及种类的设置,降低成本;在一些实施例中,若额外设置激光雷达,相应的,激光雷达的数量则会较多,可以通过多个激光雷达的数据结合,提高扫描设备的三维建模的准确性,以及扫描设备重定位准确性。
在一些实施例中,多帧当前传感器信息还包括多帧当前IMU信息,步骤110进一步包括:
步骤d01:根据多帧当前IMU信息生成世界坐标系的当前IMU空间轨迹点云;
步骤d02:将当前IMU空间轨迹点云和当前三维点云根据空间坐标位置进行匹配,得到当前IMU空间轨迹点云和当前三维点云的第二匹配度;
步骤d03:判断第二匹配度是否大于或等于第二匹配阈值;
步骤d04:若第二匹配度大于或等于第二匹配阈值,确定当前融合点云;
步骤d05:根据当前融合点云对多帧当前三维传感器信息和多帧当前IMU信息进行三维融合语义分割,生成当前融合语义图谱,将当前融合语义图谱作为当前空间语义图谱。
步骤d01中,IMU传感器设置于扫描设备上,以获取扫描设备的运动轨迹,在重定位过程中,处理器获取IMU传感器关于扫描设备的当前姿态角和加速度,通过积分计算并转换为世界坐标系的当前IMU空间轨迹点云,其中该当前IMU空间轨迹点云表示扫描设备当前的空间运动轨迹。
在一些实施例中,可以在扫描设备中断扫描前进行特殊轨迹运动,在重定位过程中,扫描设备同样进行特殊轨迹运动,以提高重定位的准确度。其中,特殊轨迹如8字形,或者M形或者其他具有多个转折的轨迹,在此不做限定,根据需要设置。
步骤d02中,由于当前IMU空间轨迹点云具有运动轨迹中不同点特征的空间位置信息,而当前三维点云同样具有多个点特征的空间位置信息,因此,可以通过空间坐标位置将当前IMU空间轨迹点云和当前三维点云进行融合,其中,需要判断当前IMU空间轨迹点云和当前三维点云是否匹配,因此,需要计算当前IMU空间轨迹点云和当前三维点云的第二匹配度,以确定是否可以融合。第二匹配度的计算与第一匹配度的计算方式类似,可以通过简单的比值关系计算,或者第一匹配度也可以通过加权计算,以提高匹配准确度,或者其他计算方式,在此不做限定,根据需要确定。
步骤d03和d04中,处理器判断第二匹配度是否大于或等于第二匹配阈值;若处理器判断第二匹配度大于或等于第二匹配阈值,则说明当前IMU空间轨迹点云和当前三维点云匹配,可以进行融合,其中,可以通过邻域搜索方式进行融合,得到当前融合点云,该当前融合点云以当前三维点云为主,将当前IMU空间轨迹点云融合到当前三维点云中。其中,第二匹配阈值的设置根据需要设置,通常第二匹配阈值设置为70%以上,以保证当前IMU空间轨迹点云和当前三维点云的第二匹配度较大以及较准确,有利于当前融合点云的准确度,从而更准确定位扫描设备。
需要说明的是,若处理器判断第二匹配度小于第二匹配阈值,相应的,不进行点云融合,此时保留当前三维点云,并相应以原有的当前三维语义图谱与历史空间语义图谱进行子图搜索匹配,以对扫描设备进行重定位。
步骤d05中,处理器根据多帧三维传感器信息能够相应的进行三维语义分割,得到三维场景中对应的三维语义对象,而后处理器在世界坐标系中对当前融合点云以及三维语义对象进行计算,得到对应的三维融合语义对象以及基于三维融合语义对象的当前空间位置信息和对象点云,根据三维融合语义对象的当前空间位置信息以及对象点云可以计算得到三维融合语义对象的空间位置、形状、尺寸等特性信息,而后根据语义名称、空间位置、形状、尺寸等特性信息生成对应三维融合语义对象的语义子图,而三维场景中具有多个三维融合语义对象,对应的具有多个语义子图,从而可以生成对应的当前三维融合语义图谱。
其中,三维传感器用于感测扫描设备周围的三维场景,通过对三维场景进行三维识别以及匹配,确定扫描设备的位置,而IMU传感器用于感测扫描设备自身的运动,通过扫描设备自身的运动轨迹确定扫描设备的位姿。通过设置IMU传感器与三维传感器共同确定扫描设备的位置,能够在不同的维度对扫描设备进行定位,有利于提高定位准确性。
在一些实施例中,步骤d02进一步包括:
步骤d021:确定当前IMU空间轨迹点云中c个轨迹点与当前三维点云中至少一候选点的距离小于或等于预设距离的m个匹配点,m≤c;
步骤d022:计算第二匹配度s2=m/c。
步骤d021中,当前IMU空间轨迹点云与当前三维点云的匹配,主要为对当前IMU空间轨迹点云的所有轨迹点的匹配。其中,将当前IMU空间轨迹点云与当前三维点云与同一世界坐标系中进行匹配计算,若搜索到当前三维点云中至少一候选点与当前IMU空间轨迹点云c个轨迹点中一个轨迹点的距离小于或等于预设距离,则认为当前IMU空间轨迹点云其中一个轨迹点匹配成功,若共有m个轨迹点能够匹配成功,则将m个轨迹点作为m个匹配点,由此计算对应的第二匹配度。其中,该距离为欧式距离,优选的,预设距离设为10厘米,以减少匹配误差,提高匹配精度。当然,在一些实施例中,预设距离也可以设置为5厘米或者8厘米或者15厘米或者其他距离数值在此不做限定,根据需要设置。其中,预设距离设置越小,匹配精度越高。
步骤d022中,通过比值关系确定第二匹配度,即第二匹配度s2=m/c。
在一些实施例中,当前空间语义图谱包括k个当前空间语义对象,历史空间语义图谱包括h个历史空间语义对象,步骤130进一步包括:
步骤e01:确定k个当前空间语义对象与h个历史空间语义对象匹配的n个当前空间语义对象,n≤k。
步骤e02:计算第一匹配度s1=n/k。
步骤e01和步骤e02中,由于当前空间语义图谱包括k个当前空间语义对象,历史空间语义图谱包括h个历史空间语义对象,在进行匹配时,主要为对k个 当前空间语义对象的匹配。若n个当前空间语义对象与n个历史空间语义对象匹配,则相应的,通过比值关系确定第一匹配度s1=n/k。
图2示出了本发明实施例提供的扫描设备的三维续扫方法的流程图,该方法由计算设备执行。该方法包括以下步骤:
步骤210:判断扫描设备是否需要对三维场景进行续扫。
步骤220:若需要对三维场景进行续扫,根据上述的续扫重定位方法判断扫描设备是否重定位成功。
步骤230:若重定位成功,继续扫描三维场景直至扫描完成。
步骤210和步骤220中,扫描设备容易由于处理器运算能力、电池电量等因素而中断扫描进程,待扫描设备重新启动继续工作后,若扫描设备已完成扫描,相应的,处理器判断扫描设备不需要对三维场景进行续扫。若扫描设备还未完成扫描,三维模型还未重建完成,处理器判断扫描设备需要对三维场景进行续扫,此时需要对扫描设备进行重定位以确定扫描设备的位置,处理器通过利用续扫重定位方法对扫描设备进行重定位计算,以判断扫描设备是否重定位成功。
步骤230中,若重定位成功,相应的,可以确定扫描设备的位置,从而确定扫描设备扫描中断时历史三维重建模型的中断边界,处理器根据扫描设备的续扫传感器信息继续根据该中断边界完成三维模型重建,以自动接续续扫前的历史三维重建模型的中断边界,将续扫后生成的三维信息自动拼接至历史三维重建模型直至整个三维模型重建完成,方便操作,提高扫描效率。
其中,处理器根据扫描设备续扫后的续扫传感器信息计算生成对应的续扫三维点云,通过点云配准方式将续扫三维点云与历史三维重建模型的历史三维点云进行匹配,接续历史三维重建模型的中断边界,从而自动从中断边界继续进行三维模型的续扫重建,直至三维模型重建完成。在一些实施例中,可以通过ICP的算法,将续扫三维点云与历史三维重建模型的历史三维点云进行匹配,使得续扫三维模型与历史三维模型自动拼接对齐。
通过步骤210至步骤230,以对扫描设备进行重定位,以确定需要进行续扫的扫描设备的位置,从而确定扫描设备扫描中断时历史三维重建模型的中断边界,处理器根据扫描设备的续扫传感器信息继续根据该中断边界完成三维模型重建,以自动接续续扫前的历史三维重建模型的中断边界,将续扫后生成的三维信息自动拼接至历史三维重建模型直至整个三维模型重建完成,方便操作,提高扫描效率。
图3示出了本发明实施例提供的续扫重定位装置的结构示意图,该装置300包括:
第一确定模块310,用于根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,多帧当前传感器信息由不同传感器得到,当前空间语义图谱具有当前空间位置信息;
第一获取模块320,用于获取扫描设备针对三维场景的历史空间语义图谱,历史空间语义图谱根据多帧历史传感器信息确定,多帧历史传感器信息由扫描 设备中断扫描前的不同传感器得到,历史空间语义图谱具有历史空间位置信息;
第一匹配模块330,用于将当前空间语义图谱与历史空间语义图谱通过子图搜索方式进行匹配,得到当前空间语义图谱与历史空间语义图谱的第一匹配度;
第一判断模块340,用于判断第一匹配度是否大于或等于第一匹配阈值;
第二确定模块350,用于若第一匹配度大于或等于第一匹配阈值,确定扫描设备在三维场景中的重定位成功。
在一些实施例中,多帧当前传感器信息包括多帧当前三维传感器信息,当前空间语义图谱包括当前三维语义图谱,第一确定模块310进一步包括:
第一生成单元,用于根据多帧当前三维传感器信息生成世界坐标系的当前三维点云;
第二生成单元,用于根据当前三维点云对多帧当前三维传感器信息进行三维语义分割,生成当前三维语义图谱,将当前三维语义图谱作为当前空间语义图谱。
在一些实施例中,多帧当前三维传感器信息包括多帧当前视觉图像信息,当前三维语义图谱包括当前视觉语义图谱,第一确定模块310进一步包括:
第三生成单元,用于根据多帧当前视觉图像信息生成世界坐标系的当前三维视觉点云;
第四生成单元,用于根据当前三维视觉点云对多帧当前视觉图像信息进行三维视觉语义分割,生成当前视觉语义图谱。
在一些实施例中,根据当前三维视觉点云对多帧当前视觉图像信息进行三维视觉语义分割,生成当前视觉语义图谱,第四生成单元进一步包括:
第一获得单元,用于第一通过图像识别算法对多帧当前视觉图像信息进行三维视觉语义分割,得到多帧当前视觉图像信息的三维视觉语义对象;
第一计算单元,用于根据当前三维视觉点云对三维视觉语义对象进行空间位置计算,得到三维视觉语义对象的当前空间位置信息;
第五生成单元,用于根据三维视觉语义对象的特性信息,生成当前视觉语义图谱,特性信息包括当前空间位置信息。
在一些实施例中,多帧当前三维传感器信息包括多帧当前激光信息,当前三维语义图谱包括当前激光语义图谱,第一确定模块310进一步包括:
第六生成单元,用于根据多帧当前激光信息生成世界坐标系的当前三维激光点云;
第七生成单元,用于根据当前三维激光点云对多帧当前激光信息进行三维激光语义分割,生成当前激光语义图谱。
在一些实施例中,多帧当前传感器信息还包括多帧当前IMU信息,第一确定模块310进一步包括:
第八生成单元,用于根据多帧当前IMU信息生成世界坐标系的当前IMU空间轨迹点云;
第二获得单元,用于将当前IMU空间轨迹点云和当前三维点云根据空间坐标位置进行匹配,得到当前IMU空间轨迹点云和当前三维点云的第二匹配度;
第一判别单元,用于判断第二匹配度是否大于或等于第二匹配阈值;
第一判定单元,用于若第二匹配度大于或等于第二匹配阈值,确定当前融合点云;
第一融合单元,用于根据当前融合点云对多帧当前三维传感器信息和多帧当前IMU信息进行三维融合语义分割,生成当前融合语义图谱,将当前融合语义图谱作为当前空间语义图谱。
在一些实施例中,将当前IMU空间轨迹点云和当前三维点云根据空间坐标位置进行匹配,得到当前IMU空间轨迹点云和当前三维点云的第二匹配度,进一步包括:
第二计算单元,用于确定当前IMU空间轨迹点云中c个轨迹点与当前三维点云中至少一候选点的距离小于或等于预设距离的m个匹配点,m≤c;
第三计算单元,用于计算第二匹配度s2=m/c。
在一些实施例中,当前空间语义图谱包括k个当前空间语义对象,历史空间语义图谱包括h个历史空间语义对象,第一匹配模块330进一步包括:
第四计算单元,用于确定k个当前空间语义对象与h个历史空间语义对象匹配的n个当前空间语义对象,n≤k;
第五计算单元,用于计算第一匹配度s1=n/k。
图4示出了本发明实施例提供的三维续扫装置的结构示意图,该装置400包括:
第二判断模块410,用于判断扫描设备是否需要对三维场景进行续扫;
第三判断模块420,用于若需要对三维场景进行续扫,根据上述的续扫重定位方法判断扫描设备是否重定位成功;
执行模块420,用于若重定位成功,根据扫描设备的续扫传感器信息完成三维模型重建。
图5示出了本发明实施例提供的计算设备的结构示意图,本发明具体实施例并不对计算设备的具体实现做限定。
如图5所示,该计算设备可以包括:处理器(processor)502、通信接口(Communications Interface)504、存储器(memory)506、以及通信总线508。
其中:处理器502、通信接口504、以及存储器506通过通信总线508完成相互间的通信。通信接口504,用于与其它设备比如客户端或其它服务器等的网元通信。处理器502,用于执行程序510,具体可以执行上述用于续扫重定位方法实施例中的相关步骤。
具体地,程序510可以包括程序代码,该程序代码包括计算机可执行指令。
处理器502可能是中央处理器CPU,或者是特定集成电路ASIC(Application Specific Integrated Circuit),或者是被配置成实施本发明实施例的一个或多个集成电路。计算设备包括的一个或多个处理器,可以是同一类型的处理器,如一个或多个CPU;也可以是不同类型的处理器,如一个或多个CPU以及一个或多个ASIC。
存储器506,用于存放程序510。存储器506可能包含高速RAM存储器, 也可能还包括非易失性存储器(non-volatile memory),例如至少一个磁盘存储器。
本发明实施例还提供一种种计算机可读存储介质,存储介质中存储有至少一可执行指令,可执行指令在运行时执行上述任一项的续扫重定位方法的操作。
在此提供的算法或显示不与任何特定计算机、虚拟系统或者其它设备固有相关。各种通用系统也可以与基于在此的示教一起使用。根据上面的描述,构造这类系统所要求的结构是显而易见的。此外,本发明实施例也不针对任何特定编程语言。应当明白,可以利用各种编程语言实现在此描述的本发明的内容,并且上面对特定语言所做的描述是为了披露本发明的最佳实施方式。
在此处所提供的说明书中,说明了大量具体细节。然而,能够理解,本发明的实施例可以在没有这些具体细节的情况下实践。在一些实例中,并未详细示出公知的方法、结构和技术,以便不模糊对本说明书的理解。
类似地,应当理解,为了精简本发明并帮助理解各个发明方面中的一个或多个,在上面对本发明的示例性实施例的描述中,本发明实施例的各个特征有时被一起分组到单个实施例、图、或者对其的描述中。然而,并不应将该公开的方法解释成反映如下意图:即所要求保护的本发明要求比在每个权利要求中所明确记载的特征更多的特征。
本领域技术人员可以理解,可以对实施例中的设备中的模块进行自适应性地改变并且把它们设置在与该实施例不同的一个或多个设备中。可以把实施例中的模块或单元或组件组合成一个模块或单元或组件,以及可以把它们分成多个子模块或子单元或子组件。除了这样的特征和/或过程或者单元中的至少一些是相互排斥之外,可以采用任何组合对本说明书(包括伴随的权利要求、摘要和附图)中公开的所有特征以及如此公开的任何方法或者设备的所有过程或单元进行组合。除非另外明确陈述,本说明书(包括伴随的权利要求、摘要和附图)中公开的每个特征可以由提供相同、等同或相似目的的替代特征来代替。
应该注意的是上述实施例对本发明进行说明而不是对本发明进行限制,并且本领域技术人员在不脱离所附权利要求的范围的情况下可设计出替换实施例。在权利要求中,不应将位于括号之间的任何参考符号构造成对权利要求的限制。单词“包含”不排除存在未列在权利要求中的元件或步骤。位于元件之前的单词“一”或“一个”不排除存在多个这样的元件。本发明可以借助于包括有若干不同元件的硬件以及借助于适当编程的计算机来实现。在列举了若干装置的单元权利要求中,这些装置中的若干个可以是通过同一个硬件项来具体体现。单词第一、第二、以及第三等的使用不表示任何顺序。可将这些单词解释为名称。上述实施例中的步骤,除有特殊说明外,不应理解为对执行顺序的限定。

Claims (12)

  1. 一种续扫重定位方法,其特征在于,所述方法包括:
    根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,多帧所述当前传感器信息由不同传感器得到,所述当前空间语义图谱具有当前空间位置信息;
    获取所述扫描设备针对所述三维场景的历史空间语义图谱,所述历史空间语义图谱根据多帧历史传感器信息确定,多帧所述历史传感器信息由所述扫描设备中断扫描前的不同所述传感器得到,所述历史空间语义图谱具有历史空间位置信息;
    将所述当前空间语义图谱与所述历史空间语义图谱通过子图搜索方式进行匹配,得到所述当前空间语义图谱与所述历史空间语义图谱的第一匹配度;
    判断所述第一匹配度是否大于或等于第一匹配阈值;
    若所述第一匹配度大于或等于所述第一匹配阈值,确定所述扫描设备在所述三维场景中的重定位成功。
  2. 根据权利要求1所述的续扫重定位方法,其特征在于,多帧所述当前传感器信息包括多帧当前三维传感器信息,所述当前空间语义图谱包括当前三维语义图谱,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
    根据多帧所述当前三维传感器信息生成世界坐标系的当前三维点云;
    根据所述当前三维点云对多帧所述当前三维传感器信息进行三维语义分割,生成所述当前三维语义图谱,将所述当前三维语义图谱作为所述当前空间语义图谱。
  3. 根据权利要求2所述的续扫重定位方法,其特征在于,多帧所述当前三维传感器信息包括多帧当前视觉图像信息,所述当前三维语义图谱包括当前视觉语义图谱,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
    根据多帧所述当前视觉图像信息生成所述世界坐标系的当前三维视觉点云;
    根据所述当前三维视觉点云对多帧所述当前视觉图像信息进行三维视觉语义分割,生成所述当前视觉语义图谱。
  4. 根据权利要求3所述的续扫重定位方法,其特征在于,所述根据所述当前三维视觉点云对多帧所述当前视觉图像信息进行三维视觉语义分割,生成所述当前视觉语义图谱,进一步包括:
    通过图像识别算法对多帧所述当前视觉图像信息进行三维视觉语义分割,得到多帧所述当前视觉图像信息的三维视觉语义对象;
    根据所述当前三维视觉点云对所述三维视觉语义对象进行空间位置计算,得到所述三维视觉语义对象的所述当前空间位置信息;
    根据所述三维视觉语义对象的特性信息,生成所述当前视觉语义图谱,所述特性信息包括所述当前空间位置信息。
  5. 根据权利要求2所述的续扫重定位方法,其特征在于,多帧所述当前三 维传感器信息包括多帧当前激光信息,所述当前三维语义图谱包括当前激光语义图谱,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
    根据多帧所述当前激光信息生成所述世界坐标系的当前三维激光点云;
    根据所述当前三维激光点云对多帧所述当前激光信息进行三维激光语义分割,生成所述当前激光语义图谱。
  6. 根据权利要求2-5任一项所述的续扫重定位方法,其特征在于,多帧所述当前传感器信息还包括多帧当前IMU信息,所述根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,进一步包括:
    根据多帧所述当前IMU信息生成所述世界坐标系的当前IMU空间轨迹点云;
    将所述当前IMU空间轨迹点云和所述当前三维点云根据空间坐标位置进行匹配,得到所述当前IMU空间轨迹点云和所述当前三维点云的第二匹配度;
    判断所述第二匹配度是否大于或等于第二匹配阈值;
    若所述第二匹配度大于或等于所述第二匹配阈值,确定当前融合点云;
    根据所述当前融合点云对多帧所述当前三维传感器信息和多帧所述当前IMU信息进行三维融合语义分割,生成当前融合语义图谱,将所述当前融合语义图谱作为所述当前空间语义图谱。
  7. 根据权利要求6所述的续扫重定位方法,其特征在于,所述将所述当前IMU空间轨迹点云和所述当前三维点云根据空间坐标位置进行匹配,得到所述当前IMU空间轨迹点云和所述当前三维点云的第二匹配度,进一步包括:
    确定所述当前IMU空间轨迹点云中c个轨迹点与所述当前三维点云中至少一候选点的距离小于或等于预设距离的m个匹配点,m≤c;
    计算所述第二匹配度s2=m/c。
  8. 根据权利要求1所述的续扫重定位方法,其特征在于,所述当前空间语义图谱包括k个当前空间语义对象,所述历史空间语义图谱包括h个历史空间语义对象,所述将所述当前空间语义图谱与所述历史空间语义图谱通过子图搜索方式进行匹配,得到所述当前空间语义图谱与所述历史空间语义图谱的第一匹配度,进一步包括:
    确定k个所述当前空间语义对象与h个所述历史空间语义对象匹配的n个所述当前空间语义对象,n≤k;
    计算所述第一匹配度s1=n/k。
  9. 一种三维续扫方法,其特征在于,所述方法包括:
    判断扫描设备是否需要对三维场景进行续扫;
    若需要对所述三维场景进行续扫,根据权利要求1-8任一项所述的续扫重定位方法判断所述扫描设备是否重定位成功;
    若重定位成功,根据所述扫描设备的续扫传感器信息完成三维模型重建。
  10. 一种续扫重定位装置,其特征在于,所述装置包括:
    第一确定模块,用于根据扫描设备在三维场景中获取的多帧当前传感器信息确定当前空间语义图谱,多帧所述当前传感器信息由不同传感器得到,所述 当前空间语义图谱具有当前空间位置信息;
    第一获取模块,用于获取所述扫描设备针对所述三维场景的历史空间语义图谱,所述历史空间语义图谱根据多帧历史传感器信息确定,多帧所述历史传感器信息由所述扫描设备中断扫描前的不同所述传感器得到,所述历史空间语义图谱具有历史空间位置信息;
    第一匹配模块,用于将所述当前空间语义图谱与所述历史空间语义图谱通过子图搜索方式进行匹配,得到所述当前空间语义图谱与所述历史空间语义图谱的第一匹配度;
    第一判断模块,用于判断所述第一匹配度是否大于或等于第一匹配阈值;
    第二确定模块,用于若所述第一匹配度大于或等于所述第一匹配阈值,确定所述扫描设备在所述三维场景中的重定位成功。
  11. 一种计算设备,其特征在于,包括:处理器、存储器、通信接口和通信总线,所述处理器、所述存储器和所述通信接口通过所述通信总线完成相互间的通信;
    所述存储器用于存放至少一可执行指令,所述可执行指令使所述处理器执行如权利要求1-8中任一项所述的续扫重定位方法的操作。
  12. 一种计算机可读存储介质,其特征在于,所述存储介质中存储有至少一可执行指令,所述可执行指令在运行时执行如权利要求1-8中任一项所述的续扫重定位方法的操作。
PCT/CN2023/113314 2022-09-06 2023-08-16 续扫重定位方法、装置、设备、存储介质及三维续扫方法 WO2024051458A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2022110831701 2022-09-06
CN202211083170.1A CN115205470B (zh) 2022-09-06 2022-09-06 续扫重定位方法、装置、设备、存储介质及三维续扫方法

Publications (1)

Publication Number Publication Date
WO2024051458A1 true WO2024051458A1 (zh) 2024-03-14

Family

ID=83572419

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/113314 WO2024051458A1 (zh) 2022-09-06 2023-08-16 续扫重定位方法、装置、设备、存储介质及三维续扫方法

Country Status (2)

Country Link
CN (1) CN115205470B (zh)
WO (1) WO2024051458A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115205470B (zh) * 2022-09-06 2023-02-21 深圳市其域创新科技有限公司 续扫重定位方法、装置、设备、存储介质及三维续扫方法
CN116299368B (zh) * 2023-05-19 2023-07-21 深圳市其域创新科技有限公司 激光扫描仪的精度测量方法、装置、扫描仪及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108021884A (zh) * 2017-12-04 2018-05-11 深圳市沃特沃德股份有限公司 基于视觉重定位的扫地机断电续扫方法、装置及扫地机
CN110956664A (zh) * 2019-12-17 2020-04-03 武汉易维晟医疗科技有限公司 一种手持式三维扫描系统的相机位置实时重定位方法
WO2021228703A1 (en) * 2020-05-12 2021-11-18 Koninklijke Philips N.V. Repositioning method
CN115205470A (zh) * 2022-09-06 2022-10-18 深圳市其域创新科技有限公司 续扫重定位方法、装置、设备、存储介质及三维续扫方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111258313B (zh) * 2020-01-20 2022-06-07 深圳市普渡科技有限公司 多传感器融合slam系统及机器人
CN113140031B (zh) * 2020-01-20 2024-04-19 苏州佳世达光电有限公司 三维影像建模系统、方法及应用其的口腔扫描设备
CN112927269A (zh) * 2021-03-26 2021-06-08 深圳市无限动力发展有限公司 基于环境语义的地图构建方法、装置和计算机设备
CN113902801A (zh) * 2021-09-24 2022-01-07 四川启睿克科技有限公司 移动机器人重定位方法、装置、设备及存储介质
CN114202701A (zh) * 2021-12-16 2022-03-18 浙江建德通用航空研究院 一种基于物体语义的无人机视觉重定位方法
CN114627365B (zh) * 2022-03-24 2023-01-31 北京易航远智科技有限公司 场景重识别方法、装置、电子设备及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108021884A (zh) * 2017-12-04 2018-05-11 深圳市沃特沃德股份有限公司 基于视觉重定位的扫地机断电续扫方法、装置及扫地机
CN110956664A (zh) * 2019-12-17 2020-04-03 武汉易维晟医疗科技有限公司 一种手持式三维扫描系统的相机位置实时重定位方法
WO2021228703A1 (en) * 2020-05-12 2021-11-18 Koninklijke Philips N.V. Repositioning method
CN115205470A (zh) * 2022-09-06 2022-10-18 深圳市其域创新科技有限公司 续扫重定位方法、装置、设备、存储介质及三维续扫方法

Also Published As

Publication number Publication date
CN115205470B (zh) 2023-02-21
CN115205470A (zh) 2022-10-18

Similar Documents

Publication Publication Date Title
WO2024051458A1 (zh) 续扫重定位方法、装置、设备、存储介质及三维续扫方法
US11341746B2 (en) Method and apparatus for identifying a damaged part of a vehicle, server, client and system
US20220383579A1 (en) Object modeling and movement method and apparatus, and device
Liang et al. Image based localization in indoor environments
WO2019192358A1 (zh) 一种全景视频合成方法、装置及电子设备
WO2022068225A1 (zh) 点云标注的方法、装置、电子设备、存储介质及程序产品
WO2022078467A1 (zh) 机器人自动回充方法、装置、机器人和存储介质
US11067669B2 (en) Method and apparatus for adjusting point cloud data acquisition trajectory, and computer readable medium
WO2019091310A1 (zh) 区域属性确定
WO2019062651A1 (zh) 一种定位建图的方法及系统
US9632678B2 (en) Image processing apparatus, image processing method, and program
WO2022000755A1 (zh) 机器人及其行动控制方法、装置和计算机可读存储介质
JP2015079490A (ja) フレームを選択する方法、装置、及びシステム
US11216919B2 (en) Image processing method, apparatus, and computer-readable recording medium
WO2022262160A1 (zh) 传感器标定方法及装置、电子设备和存储介质
CN111260789B (zh) 避障方法、虚拟现实头戴设备以及存储介质
JP6936866B2 (ja) 外観特徴の記述属性を認識する方法及び装置
WO2022088611A1 (zh) 障碍检测方法、装置、电子设备、存储介质及计算机程序
US10930069B1 (en) 3D scanning and modeling system
JP2022087153A (ja) 情報処理システム
WO2022062355A1 (zh) 一种融合定位方法及装置
CN112097742B (zh) 一种位姿确定方法及装置
WO2023088127A1 (zh) 室内导航方法、服务器、装置和终端
US20230367319A1 (en) Intelligent obstacle avoidance method and apparatus based on binocular vision, and non-transitory computer-readable storage medium
WO2022011560A1 (zh) 图像裁剪方法与装置、电子设备及存储介质

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: 23862155

Country of ref document: EP

Kind code of ref document: A1