WO2022183656A1 - 数据生成方法、装置、设备、存储介质及程序 - Google Patents

数据生成方法、装置、设备、存储介质及程序 Download PDF

Info

Publication number
WO2022183656A1
WO2022183656A1 PCT/CN2021/105485 CN2021105485W WO2022183656A1 WO 2022183656 A1 WO2022183656 A1 WO 2022183656A1 CN 2021105485 W CN2021105485 W CN 2021105485W WO 2022183656 A1 WO2022183656 A1 WO 2022183656A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
image
voxel
information
fusion
Prior art date
Application number
PCT/CN2021/105485
Other languages
English (en)
French (fr)
Inventor
段永利
孙佳明
周晓巍
Original Assignee
浙江商汤科技开发有限公司
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 浙江商汤科技开发有限公司 filed Critical 浙江商汤科技开发有限公司
Priority to KR1020227014409A priority Critical patent/KR20220125715A/ko
Publication of WO2022183656A1 publication Critical patent/WO2022183656A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation 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
    • G06V10/267—Segmentation 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 by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30244—Camera pose

Definitions

  • the present disclosure relates to the field of computer vision, and in particular, to a data generation method, apparatus, device, storage medium and program.
  • scene reconstruction With the rapid development of computer vision technology, scene reconstruction has become an increasingly important application in the field of computer vision.
  • scene reconstruction usually includes geometric information such as the shape and position of each target object in the scene.
  • the interaction effect can be achieved by obtaining the geometric information of the target object in the scene.
  • the embodiment of the present disclosure proposes a data generation solution.
  • An embodiment of the present disclosure provides a data generation method, the method is executed by an electronic device, and the method includes:
  • Semantic segmentation is performed on the target image to obtain semantic information of the target image
  • the map data of the target scene is obtained.
  • the semantic information obtained from the segmentation in the target image is fused through the voxel information set of the target image, so as to realize the continuous fusion of semantic information in the target scene, and obtain map data containing the continuously fused semantic information, which can effectively improve the quality of the obtained map data. Data completeness and quality.
  • the fusion of the semantic information into the voxel information set to obtain a fusion voxel information set includes: projecting at least one voxel in the voxel information set to the voxel information set The target image, according to the semantic information of at least one pixel in the target image, determine the semantic probability distribution information of the at least one voxel after projection, and obtain the fusion voxel information set.
  • the fusion voxel information set contains semantic information.
  • the probability distribution can continuously express the semantic information in the target scene with high quality, and make the map data of the target scene obtained based on the fusion voxel information set can be more widely used in many scenes that require semantic information. .
  • the fusion of the semantic information into the voxel information set to obtain the fusion voxel information set further includes: acquiring collection data obtained by performing data collection on the target scene , and fuse the collected data into the voxel information set to obtain the fused voxel information set.
  • the comprehensiveness of the data in the combination of the fusion voxel information can be increased, and the map data of the target scene obtained based on the fusion voxel information set can be more comprehensive and complete, and have higher quality.
  • the fusion of the collected data into the voxel information set to obtain a fusion voxel information set includes: according to the fusion weight of the collected data, and the voxel information According to the projection relationship between at least one voxel in the set and the collected data, information fusion is performed on at least one voxel in the voxel information set to obtain the fused voxel information set.
  • the collected data includes various forms of data
  • different information can be flexibly fused according to different forms of the collected data, which can not only improve the data integrity of the fusion voxel information set, but also improve the fusion efficiency.
  • the comprehensiveness of the obtained map data of the target scene and the efficiency of data generation can be improved.
  • the target image includes at least two images to be processed
  • obtaining map data of the target scene according to the fusion voxel information set corresponding to the target image includes:
  • the fusion voxel information sets corresponding to the at least two images to be processed are combined to obtain map data of the target scene.
  • the map data is obtained by merging the fusion voxel information set corresponding to at least one target image, which can effectively improve the data integrity and comprehensiveness of the map data, and in the case of including multiple continuous target images, the map data can be A continuous representation of the target scene.
  • the method further includes: performing loop closure detection on at least one frame of the target image, and using the loop closure detected target image as the loop closure image; determining a second pose of the loop closure image, wherein , the accuracy of the second pose is higher than the accuracy of the first pose of the target image corresponding to the loopback image; according to the second pose of the loopback image, the map data of the target scene is updated .
  • the obtained second pose has higher accuracy than the first pose determined by the target image corresponding to the loopback image, so that the map data updated based on the second pose has higher accuracy. precision.
  • the updating the map data of the target scene according to the second pose of the loopback image includes: acquiring a corresponding image of the loopback image in the first pose Fusing the voxel information set as the first target set; based on the first pose and the second pose corresponding to the loopback image, re-fuse the information fused in the first target set to obtain the second target set; According to the second target set, the map data of the target scene is updated.
  • the pose and various types of information in the map data can be optimized separately.
  • the amount of calculation in the map data update process can be effectively reduced. , and improve the update efficiency of map data to realize real-time update of map data.
  • the information fused in the first target set is re-fused based on the first pose and the second pose corresponding to the loop closure image to obtain a second target set, including : According to the first pose corresponding to the loopback image, de-fuse the information fused in the first target set to obtain a third target set; according to the second pose corresponding to the loopback image, defuse the loopback image The information of the image is fused into the third target set to obtain the second target set.
  • the information fused in the first target set includes various forms of data
  • different information can be flexibly de-fused according to different forms of the information fused in the first target set, which can improve the efficiency of de-fusion and flexibility, thereby improving the update efficiency of map data.
  • An embodiment of the present disclosure provides a data generating apparatus, the apparatus comprising:
  • the voxel information set establishment part is configured to determine the first pose of the target image in the target scene, and establish the voxel information set of the target image according to the first pose, wherein the target image includes At least one frame of image obtained by data collection of the target scene;
  • a semantic segmentation part configured to perform semantic segmentation on the target image to obtain semantic information of the target image
  • a fusion part configured to fuse the semantic information into the voxel information set to obtain a fusion voxel information set
  • the data generating part is configured to obtain map data of the target scene according to the fusion voxel information set corresponding to the target image.
  • the fusion part is further configured to: project at least one voxel in the voxel information set to the target image, and according to the semantic information of at least one pixel in the target image, Determining the projected semantic probability distribution information of the at least one voxel to obtain the fusion voxel information set.
  • the fusion part is further configured to: acquire collection data obtained by performing data collection on the target scene, and fuse the collection data into the voxel information set to obtain the collection data.
  • the fusion voxel information set is further configured to: acquire collection data obtained by performing data collection on the target scene, and fuse the collection data into the voxel information set to obtain the collection data.
  • the fusion voxel information set is further configured to: acquire collection data obtained by performing data collection on the target scene, and fuse the collection data into the voxel information set to obtain the collection data.
  • the fusion part is further configured to: according to the fusion weight of the collected data and the projection relationship between at least one voxel in the voxel information set and the collected data, Perform information fusion on at least one voxel in the voxel information set to obtain the fused voxel information set.
  • the target image includes at least two to-be-processed images
  • the data generating part is further configured to: perform the fused voxel information set corresponding to the at least two to-be-processed images. Combined to obtain the map data of the target scene.
  • the apparatus further includes: a detection part, configured to perform loop closure detection on at least one frame of the target image, and use the target image whose loop closure is detected as a loop closure image; a determination part, configured to determine the loop closure The second pose of the loopback image, wherein the accuracy of the second pose is higher than the accuracy of the first pose of the target image corresponding to the loopback image; the update part is configured to be based on the loopback image.
  • the second pose updates the map data of the target scene.
  • the updating part is further configured to: acquire a fusion voxel information set corresponding to the loopback image in the first pose as a first target set; based on the loopback image The corresponding first pose and the second pose are re-fused to the information fused in the first target set to obtain a second target set; according to the second target set, the map data of the target scene is processed. renew.
  • the updating part is further configured to: de-fuse the information fused in the first target set according to the first pose corresponding to the loopback image to obtain a third target set ; According to the second pose corresponding to the loopback image, fuse the information of the loopback image into the third target set to obtain the second target set.
  • Embodiments of the present disclosure also provide an electronic device, including:
  • processor configured to: execute the data generation method described in any of the foregoing embodiments.
  • Embodiments of the present disclosure further provide a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, implement the data generation method described in any of the foregoing embodiments.
  • Embodiments of the present disclosure further provide a computer program, where the computer program includes computer-readable codes, and when the computer-readable codes are executed in an electronic device, the processor of the electronic device executes any of the foregoing implementations The data generation method described in the example.
  • the embodiments of the present disclosure provide at least one data generation method, device, device, storage medium and program, by determining the first pose of the target image in the target scene, establishing the voxel information set of the target image according to the first pose, and The semantic information obtained by semantically segmenting the target image is fused into the voxel information set to obtain the fused voxel information set, and the map data of the target scene is obtained according to the fused voxel information set.
  • the semantic information obtained by segmentation in the target image can be fused by establishing a voxel information set, so as to realize the continuous fusion of semantic information in the target scene, and obtain map data containing the continuously fused semantic information, which can effectively improve the quality of the obtained map data. Data completeness and quality.
  • FIG. 1 shows a schematic flowchart of a data generation method according to an embodiment of the present disclosure
  • FIG. 2 shows a schematic diagram of a system architecture to which the data generation method according to an embodiment of the present disclosure is applied;
  • FIG. 3 shows a schematic frame diagram of a data generating apparatus according to an embodiment of the present disclosure
  • FIG. 4 shows a schematic flowchart of obtaining map data according to an application example of the present disclosure
  • FIG. 5 shows a schematic diagram of semantic effect fusion of a target scene in an application example of the present disclosure
  • FIG. 6 shows a schematic block diagram of the architecture of an electronic device 800 according to an embodiment of the present disclosure
  • FIG. 7 shows a schematic block diagram of the architecture of an electronic device 1900 according to an embodiment of the present disclosure.
  • Multiple or multiple in the embodiments of the present disclosure may refer to at least two or at least two, respectively.
  • FIG. 1 shows a schematic flowchart of a data generation method according to an embodiment of the present disclosure.
  • the method can be applied to a data generation apparatus, and the data generation apparatus can be a terminal device, a server, or other processing equipment.
  • the terminal device may be user equipment (User Equipment, UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (Personal Digital Assistant, PDA), handheld device, computing device, in-vehicle device and possible wearable devices, etc.
  • the data generation method can be applied to a cloud server or a local server, and the cloud server can be a public cloud server or a private cloud server, which can be flexibly selected according to the actual situation.
  • the data generation method may also be implemented by a processor invoking computer-readable instructions stored in a memory.
  • the data generation method may include:
  • Step S11 determining the first pose of the target image in the target scene, and establishing the voxel information set of the target image according to the first pose.
  • the target image includes at least one frame of image obtained by performing data collection on the target scene.
  • Step S12 perform semantic segmentation on the target image to obtain semantic information of the target image.
  • Step S13 fuse the semantic information into the voxel information set to obtain a fused voxel information set.
  • step S14 map data of the target scene is obtained according to the fusion voxel information set corresponding to the target image.
  • the target scene may be any scene with reconstruction requirements, and its location and scope can be flexibly selected according to actual conditions, which are not limited in the embodiments of the present disclosure, and are not limited to the following disclosed embodiments.
  • the target scene may be an indoor scene or an outdoor scene, and may also include an indoor scene and an outdoor scene at the same time.
  • the target image may be at least one frame of image obtained by performing data collection on the target scene.
  • the number of images included in the target image is not limited in the embodiments of the present disclosure, and can be flexibly determined according to the actual situation of data collection on the target scene.
  • each frame of image obtained by data collection of the target scene may be used as the target image, or a frame or frame may be selected from multiple frames of images obtained by data collection of the target scene. Multiple frames are used as the target image, and the selected method can be flexibly selected according to the actual situation.
  • it may be randomly selected, or a target image may be selected by sampling the collected images according to a certain frequency.
  • data collection may be performed on the target scene through an image collection device such as a video camera or a camera.
  • data acquisition can also be performed on the target scene through other devices including an image acquisition device.
  • an inertial measurement unit IMU
  • IMU Inertial Measurement Unit
  • the integrated device of the vision sensor of the image acquisition device performs data acquisition on the target scene. For example: a smartphone with a camera, etc.
  • the image collection device in the case where data collection is performed on the target scene through an image collection device or other apparatus including an image collection device, the image collection device may also have a function of collecting depth information, for example , in some embodiments of the present disclosure, the image acquisition device may include a time of flight (Time Of Flight, TOF) camera.
  • TOF Time Of Flight
  • the collected collection data can also be flexibly changed, and can be flexibly determined according to the actual situation, and is not limited to the following disclosed embodiments.
  • the acquired data may include at least one frame of the target image as described in the above disclosed embodiments; in some embodiments of the present disclosure, the acquired data may also include other data.
  • the collected data when data is collected on the target scene through an integrated device including an IMU, the collected data may further include IMU data.
  • the acquisition data when the image acquisition device includes a TOF camera, the acquisition data may further include depth information of the target image, and the like.
  • step S11 may determine the first pose of the target image based on the target image in the target scene.
  • the pose of the image may be the pose of the device that collects the image. Based on the pose of the image, the coordinate correspondence between the image and the world coordinate system can be determined, thereby determining the position of each pixel in the image in space, etc. .
  • the first pose may be the pose of the image capture device when the image capture device captures the target image.
  • the manner of determining the first pose is not limited in the embodiments of the present disclosure, and any method for determining the image pose can be used as the method for determining the first pose in step S11, and is not limited to the following disclosed embodiments.
  • the first pose of the target image may be determined by performing image pose estimation based only on the target image in the target scene.
  • SLAM Simultaneous Localization And Mapping
  • VIO Visual-Inertial Odometry
  • the method performs pose estimation on the target image to obtain the first pose with six degrees of freedom; in some embodiments of the present disclosure, in the case where the collected data includes depth information, the pose estimation can also be assisted based on the depth information to determine The first pose of the target image.
  • step S11 may further establish a set of voxel information of the target image according to the first pose of the target image.
  • Voxel is the abbreviation of Volume Pixel, which is the smallest unit of digital data in three-dimensional space division.
  • voxels may be used as a representation of each position in the target scene in the embodiments of the present disclosure.
  • the target image since the target image is an image in the target scene, it can represent the scene content in the target scene under a certain or certain field of view. Therefore, in some embodiments of the present disclosure, the target image may correspond to at least part of the position in the target scene, and therefore, the pixels in the target image correspond to at least part of the voxels representing the target scene. Meanwhile, in some embodiments of the present disclosure, a set of voxel information of the target image may be established according to the first pose of the target image, and the set of voxel information may include relevant voxel information of at least some voxels in the target scene, to A correspondence of the target image to at least a portion of the voxels representing the target scene is achieved.
  • the relevant voxel information contained in the voxel information set, the information content contained therein can be flexibly determined according to the actual situation, and is not limited to the following disclosed embodiments.
  • the voxel information set may include: the fusion weight W(v) of the voxel information set v in the fusion process, the image information of the target image such as the color value C(v), and the depth of the target image Information such as the truncation sign function D(v), etc.
  • the image information and depth information of the target image may also be represented by other functions; in some embodiments of the present disclosure, the voxel information set may also include correlations required by other map data.
  • the information can be flexibly expanded according to the actual situation, so I will not list them one by one.
  • the voxel information set in the embodiments of the present disclosure contains relevant information of voxels, and voxels as three-dimensional data need to determine the relevant information through depth, the following disclosures are implemented
  • the collected data and the voxel information set contain depth information, and in the process of fusing the voxel information set, the depth information needs to be fused as an example for description.
  • the manner of establishing the voxel information set of the target image according to the first pose is not limited in the embodiments of the present disclosure, and can be flexibly determined according to the actual situation, and is not limited to the following disclosed embodiments.
  • a voxel information set of each target image can be established by a voxel hashing method, and each voxel information set can be stored and searched by using a hash table.
  • the voxel information set may also be established, stored and searched through other data structures. Among them, which data structures are selected and how to establish each voxel information set based on these structures, the implementation form can be flexibly selected according to the actual situation, and will not be listed one by one here.
  • semantic segmentation of the target image may also be performed in step S12 to obtain semantic information of the target image.
  • the implementation order of step S11 and step S12 is not limited in this embodiment of the present disclosure, and step S11 and step S12 may be performed simultaneously, or may be performed sequentially according to a preset order, wherein the selection of the execution order can be flexibly based on the actual situation Decide.
  • the semantic segmentation of the target image may be to segment objects of different categories contained in the target image, and the semantic information of the target image may be category information of an object corresponding to at least one pixel in the target image.
  • the target image when the target scene is an indoor scene, the target image may be an image in an indoor scene, and the target image may include wall objects, ground objects, ceiling objects, table objects, chair objects, etc.
  • objects such as walls, floors, ceilings, tables, and chairs contained in the target image can be segmented to determine which pixels in the target image correspond to the category information of the wall, and which pixels correspond to the category information of the wall. Category information of the ground, etc.
  • the category corresponding to the semantic information of the target image may also change, and may also include other indoor object categories, such as cabinets or windows, etc., and may also include other categories.
  • the category of outdoor objects such as: sky, tree or road, etc.
  • the manner of performing semantic segmentation on the target image is not limited in the embodiments of the present disclosure. Any method that can segment the target image can be used as the implementation manner of semantic segmentation in the embodiments of the present disclosure, and is not limited to the following disclosed implementations. example.
  • the target image may be processed through a segmentation algorithm to obtain semantic information of the target image.
  • the target image may also be input into the image segmentation neural network to obtain semantic information output by the image segmentation neural network.
  • step S12 semantic segmentation is performed on the target image, which may be semantic segmentation of one target image at a time, or semantic segmentation of multiple target images at a time, and the number of target images corresponding to the semantic segmentation can be based on actual conditions.
  • the situation is determined flexibly, and is not limited in the embodiments of the present disclosure.
  • step S13 may be used to fuse the semantic information into the corresponding voxel information set to obtain a fusion voxel information set.
  • the information content that can be included in the voxel information set can be flexibly selected according to the actual situation. Therefore, in addition to the semantic information, other information can also be considered in the fusion voxel information set. Among them, which information to integrate can be flexibly selected according to the actual situation.
  • the voxel information set can be fused into the semantic information on the basis of the set information including the shape and position of each target object, so that the target scene can be processed with higher quality.
  • the manner of fusing the semantic information into the voxel information set in step S13 is not limited in the embodiments of the present disclosure, and can be flexibly selected according to the actual situation. For details, please refer to the following disclosed embodiments, which will not be expanded here.
  • the map data of the target scene can be obtained according to the fusion voxel information set corresponding to the target image.
  • the map data of the target scene may be data obtained by summarizing each frame of target images in the target scene, and the data content contained therein may be flexibly determined according to the actual situation. Therefore, the implementation of step S14 can be flexibly determined according to the actual data requirements of the map data.
  • step S14 please refer to the following disclosed embodiments for details, which is also not expanded here.
  • the first pose of the target image in the target scene is determined, the voxel information set of the target image is established according to the first pose, and the semantic information obtained by semantically segmenting the target image is fused into the volume
  • a fusion voxel information set is obtained, and the map data of the target scene is obtained according to the fusion voxel information set.
  • the semantic information obtained from the segmentation in the target image is fused by establishing a voxel information set.
  • the Fusion of semantic information can realize the continuous fusion of semantic information in the target scene, and obtain map data containing the continuously fused semantic information, which can effectively improve the data comprehensiveness and quality of the obtained map data.
  • FIG. 2 shows a schematic diagram of a system architecture to which the data generation method according to an embodiment of the present disclosure can be applied; as shown in FIG. 2 , the system architecture includes an acquisition terminal 201 , a network 202 and a data generation terminal 203 .
  • the acquisition terminal 201 and the data generation terminal 203 establish a communication connection through the network 202
  • the acquisition terminal 201 reports the target image in the target scene to the data generation terminal 203 through the network 202 .
  • the data generation terminal 203 In response to the target image in the target scene, the data generation terminal 203 firstly determines the first pose of the target image in the target scene, and establishes a voxel information set of the target image according to the first pose; and performs semantic segmentation on the target image to obtain The semantic information of the target image; secondly, the semantic information is fused into the voxel information set to obtain the fusion voxel information set; finally, the map data of the target scene is obtained according to the fusion voxel information set corresponding to the target image.
  • the data generation terminal 203 uploads the map data of the target scene to the network 202 , and sends the map data to the acquisition terminal 201 through the network 202 .
  • the acquisition terminal 201 may include an image acquisition device, and the data generation terminal 203 may include a visual processing device or a remote server with visual information processing capability.
  • Network 202 may employ wired or wireless connections.
  • the acquisition terminal 201 can communicate with the visual processing device through a wired connection, such as data communication through a bus; when the data generation terminal 203 is a remote server, the acquisition terminal 201 can Data exchange with remote server through wireless network.
  • the acquisition terminal 201 may be a vision processing device with a video capture module, or a host with a camera.
  • the data generation method of the embodiment of the present disclosure may be executed by the acquisition terminal 201 , and the above-mentioned system architecture may not include the network 202 and the data generation terminal 203 .
  • step S13 includes:
  • At least one voxel in the voxel information set is projected to the target image, and the projection method and angle can be flexibly selected according to the actual situation, which is not limited in the embodiments of the present disclosure.
  • Which voxels in the voxel information set are projected to the target image can also be flexibly selected according to the actual situation.
  • each voxel included in the voxel information set can be projected to the target image. ;
  • the selected voxels can also be projected to the target image after random selection or sampling selection of a certain proportion is performed on the voxels.
  • the semantic probability distribution information of the projected at least one voxel may be determined according to the semantic information of the at least one pixel in the target image.
  • the semantic probability distribution information may be the probability distribution of semantic information of multiple voxels. How to determine the semantic probability distribution information of at least one voxel after projection according to the semantic information of at least one pixel in the target image, its implementation form can be based on The actual situation is flexible.
  • At least one pixel in the target image may be in a one-to-one correspondence with the projected voxel, and after the semantic information is fused according to the corresponding situation, the at least one pixel is determined based on the fusion result of the semantic information of multiple voxels.
  • Semantic probability distribution information for a voxel can also be obtained according to the semantic information of at least one pixel in the target image, and fused with the semantic probability distribution information of the projected voxels Wait. Among them, how to realize can be flexibly selected according to the actual situation.
  • At least one voxel in the voxel information set is projected to the target image, and according to the semantic information of at least one pixel in the target image, the semantic probability distribution information of the projected at least one voxel is determined to obtain
  • the way of fusing the voxel information set can be expressed by the following formula (1):
  • I 1,...,k is the target image of each frame
  • I k is the current target image
  • I 1,...,k ) is the voxel belonging to the th
  • Z is the normalization factor
  • I 1,...,k-1 ) is the information before fusion of the semantic information of the current target image (that is, the fusion of I 1 to I k -1 frame of the semantic information of the target image) voxels belong to the i-th category of semantic probability distribution information
  • P(O u(v,k) l i
  • I k ) is the semantic information obtained by semantically segmenting the current target image probability distribution
  • O u(v,k) is the projection result of projecting voxel v to the current target image.
  • the semantic segmentation information P(O u(v,k) l i
  • I k ) obtained by semantically segmenting the current target image can be compared with k-1 before the current target image
  • the semantic probability distribution information of the voxels corresponding to the frame target image are multiplied and normalized, and the semantic information of the current target image is fused into the voxel information set to obtain the fusion voxel information set.
  • the fusion voxel information set is obtained by projecting at least one voxel in the voxel information set to the target image, and determining the semantic probability distribution information of the projected at least one voxel according to the semantic information of at least one pixel in the target image.
  • the voxels in the voxel information set can be projected to the target image, and the semantic information of the two-dimensional target image can be fused into the two-dimensional projection of the three-dimensional voxels, so that the fusion voxel information set contains semantic information
  • the probability distribution of the target scene can continuously express the semantic information in the target scene with high quality, and the map data of the target scene obtained based on the fusion voxel information set can be more widely used in many scenes that require semantic information. middle.
  • step S13 may further include:
  • Acquisition data obtained by performing data acquisition on the target scene is acquired, and the acquired data is fused into a voxel information set to obtain a fused voxel information set.
  • the data content included in the collected data may be flexible according to the data collection manner, for example, including depth information, IMU data, or other information.
  • the data collection manner for example, including depth information, IMU data, or other information.
  • all or part of the collected data may be fused into the voxel information set according to the actual situation of the collected data.
  • Which collected data is selected for fusion at the same time is not limited in the embodiments of the present disclosure, and can be flexibly selected according to the actual situation, and is not limited to the following disclosed embodiments.
  • the fusion method can also be flexibly changed. For details, please refer to the following disclosed embodiments, which will not be expanded here.
  • the collection data obtained by data collection of the target scene is obtained, and the collection data is fused into a voxel information set to obtain a fusion voxel information set.
  • the comprehensiveness of the data in the combination of the fusion voxel information can be increased, and the map data of the target scene obtained based on the fusion voxel information set can be more comprehensive and complete, and have higher quality.
  • the acquired data includes depth information of the target image, and in some embodiments of the present disclosure, the acquired data may also include color information, etc., and the acquired data is fused into a voxel information set to obtain a fusion body
  • a collection of prime information which can include:
  • information fusion is performed on at least one voxel in the voxel information set to obtain a fusion voxel information set.
  • the depth information may be information collected by a TOF camera as described in the above disclosed embodiments.
  • the depth information may be represented in the form of a truncated sign function D(v).
  • the color information may be RGB color information obtained from the captured target image itself.
  • the color information may be represented in the form of a color value C(v) or the like.
  • the way of fusion of collected data may change flexibly. It can be seen from the above disclosed embodiments that in the case where the collected data includes at least one of depth information and color information, the fusion weight of the collected data and the relationship between at least one voxel in the voxel information set and the collected data can be The projection relationship is used to fuse the collected data. At the same time, how to perform fusion based on the fusion weight and the projection relationship of voxels can be flexibly determined in the implementation process.
  • the fusion process of depth information can be represented by the following formulas (2) and (3):
  • D'(v) is the depth information of voxel v after fusion
  • D(v) is the depth information of voxel v before fusion
  • W(v) is the voxel determined according to the information in the voxel information set
  • the weight of v, w i (v) is the fusion weight
  • d i (v) is the distance between the voxel v's corresponding back-projection point in the depth information and the voxel v
  • W'(v) is the voxel v fusion Post updated weights.
  • the depth information is fused to the voxel according to the fusion weight w i (v) of the depth information and the projection relationship d i (v) between the voxel v and the depth information.
  • the fusion weight w i (v) of the depth information can be flexibly set according to the actual situation, and the projection relationship d i (v) between the voxels and the depth information can also be flexibly determined according to the actual situation of the voxels and the depth information .
  • d i (v) can be calculated by the following equations (4) and (5):
  • n d c (v)-X(v) (5);
  • d c (v) is the distance from the voxel v to the center of the camera
  • X(v) is the depth of the corresponding pixel obtained after the voxel v is projected to the depth information
  • ⁇ is a preset parameter.
  • the depth information is fused into the voxel information set based on the fusion weight of the depth information and the projection relationship between the voxels and the depth information.
  • the fusion method of the color information may refer to the fusion method of the above-mentioned depth information, the difference is that the truncation symbol function D(v) corresponding to the depth information is replaced by a color value C(v), and replace the projection relationship between voxels and depth information with the projection relationship between voxels and the target image, etc.
  • the fusion manner can be flexibly replaced and expanded with reference to the above disclosed embodiments.
  • the fusion voxel information set is obtained by performing information fusion on at least one voxel in the voxel information set according to the fusion weight of the collected data and the projection relationship between at least one voxel in the voxel information set and the collected data.
  • the collected data includes various forms of data
  • different information can be flexibly fused according to different forms of the collected data, which can not only improve the data integrity of the fusion voxel information set, but also improve the fusion efficiency.
  • the comprehensiveness of the obtained map data of the target scene and the efficiency of data generation can be improved.
  • step S14 may include:
  • the fusion voxel information sets corresponding to at least two images to be processed are combined to obtain map data of the target scene.
  • the map data of the target scene may be data obtained by summarizing each frame of target images in the target scene, and the data content contained therein may be flexibly determined according to the actual situation. Therefore, in some embodiments of the present disclosure, fused voxel information sets corresponding to different target images may be combined to form a data unit of fused voxel information, which is used as map data of the target scene. In this way, the map data is obtained by merging the fusion voxel information set corresponding to at least one target image, which can effectively improve the data integrity and comprehensiveness of the map data, and in the case of including multiple continuous target images, the map data can be A continuous representation of the target scene.
  • step S14 may further include:
  • the fusion voxel information set corresponding to at least one target image is stored in the map data of the target scene.
  • the fused voxel information sets corresponding to which target images are stored in the map data can be flexibly determined according to the actual situation.
  • the fused voxel information sets corresponding to each frame of the target image can be stored in the in the map data of the target scene.
  • target images may also be selected or screened to select fused voxel information sets corresponding to some target images to store in map data, etc.
  • the implementation can be flexibly determined according to actual conditions.
  • the voxel hashing method may be used to establish the voxel information set of the target image.
  • the map data may store, in addition to each fused voxel information set, a hash table or the like for searching each fused voxel information set.
  • the data integrity and comprehensiveness of the map data can be effectively improved.
  • the target scene can be represented continuously.
  • the data generation method proposed by the embodiments of the present disclosure may further include:
  • the map data of the target scene is updated.
  • the loopback detection may be to detect whether there are images for the same scene in the collected multi-frame target images.
  • the detection method of the loopback detection is not limited in this embodiment of the present disclosure, and any method used for loopback detection in the field of visual SLAM , can be used as the implementation manner of loopback detection in the embodiments of the present disclosure, and is not limited to the following disclosed embodiments.
  • loop closure detection may be implemented by establishing a bag-of-words model.
  • the target image in which the loopback is detected can be used as the loopback image, and according to the result of the loopback detection, the pose of the loopback image can be optimized to determine the second pose of the loopback image.
  • the method of optimizing the pose of the loopback image is not limited in the embodiments of the present disclosure, and any method of correcting and optimizing the pose of the loopback image based on the detection result in the loopback detection can be used as the second position.
  • the manner of determining the pose is not limited in this embodiment of the present disclosure. Since the loopback image can be used to correct and optimize the first pose corresponding to the target image in the target scene, the obtained second pose is compared with the first pose determined by the target image corresponding to the loopback image. , with higher accuracy, so that the map data updated based on the second pose has higher accuracy.
  • the map data of the target scene may be updated according to the second pose of the loopback image.
  • the update manner can be flexibly selected in the embodiments of the present disclosure. For details, please refer to the following disclosed embodiments, which will not be expanded here.
  • the second pose of the detected loopback image is determined, and the map data is updated according to the second pose.
  • the accumulated error in the map data can be effectively reduced and the map can be improved.
  • the precision of the data is provided.
  • updating the map data of the target scene according to the second pose of the loopback image includes:
  • the information fused in the first target set is re-fused to obtain the second target set;
  • the map data of the target scene is updated.
  • the fusion voxel information set corresponding to the loop closure image in the first pose is the voxel information set established according to the first pose determined by the loop closure image, and the fusion voxel information obtained by fusing the loop closure image information with the information of the loop closure image.
  • the set in this embodiment of the present disclosure, the set may be used as the first target set.
  • the pose of the loop closure image when the loop closure image is detected, the pose of the loop closure image can be optimized, and the second pose of the loop closure image can be re-determined.
  • the information contained in the first target set corresponding to the loop closure image in the first pose may have deviations from the data of the real target scene. Therefore, in some embodiments of the present disclosure, the information in the first target set may be re-fused according to the first pose and the second pose corresponding to the loopback image to obtain the second target set.
  • the fusion voxel information set may fuse various types of information, such as semantic information, depth information, or color information. Therefore, in addition to the depth information, the information fused in the first target set may also include one or more of semantic information and color information. With different types of information fused in the first target set, the manner of re-integration can also be flexibly changed. For details, see the following disclosed embodiments, which will not be expanded here.
  • the map data of the target scene can be updated according to the second target set, and the update method can be flexibly determined according to the actual situation.
  • the data of the second target set may be replaced by the data of the first target set to implement the update of the map data.
  • the fusion information in the first target set is processed. Re-integrate to obtain a second target set, so as to update the map data of the target scene according to the second target set.
  • the pose and various types of information in the map data can be optimized separately by means of re-integration.
  • the calculation in the process of updating the map data can be effectively reduced. It can improve the update efficiency of map data and realize real-time update of map data.
  • the information fused in the first target set is re-fused to obtain the second target set, including:
  • the information fused in the first target set is de-fused to obtain a third target set
  • the information of the loopback image is fused into the third target set to obtain the second target set.
  • the manner of re-merging the information fused in the first target set may include de-fusion performed according to the first pose, and re-fusion performed according to the second pose.
  • de-fusion can be regarded as a reverse process of fusion, so de-fusion can be implemented through a reverse operation of fusion.
  • the information fused in the first target set includes at least one of depth information and color information, in this case, the fusion process based on at least one of depth information and color information
  • the method for de-fusion of the information fused in the first target set may include:
  • the first target set is At least one voxel in the information is de-fused to obtain a third target set.
  • the process of performing de-fusion based on the fusion weight and the projection relationship of the voxels in the first pose can be flexibly determined.
  • the defusion process of the depth information in the first target set can be represented by the following formulas (6) and (7):
  • D"(v) is the depth information of voxel v after de-fusion
  • D'(v) is the depth information of voxel v after fusion in the first pose proposed in the above disclosed embodiments
  • W'(v ) is the weight updated after the fusion of voxels v in the first pose proposed in the above disclosed embodiment
  • w i (v) is the fusion weight proposed in the above disclosed embodiment
  • d i (v) is the above disclosed embodiment It is proposed that the distance between the corresponding back-projection point of voxel v in the depth information and the voxel v in the first pose
  • W"(v) is the updated weight of voxel v after de-fusion.
  • the implementation forms of the fusion weight w i (v) of the depth information and the projection relationship d i (v) between the voxels and the depth information may refer to the above disclosed embodiments.
  • the de-fusion method of the color information can refer to the above-mentioned de-fusion method of the depth information
  • the difference is that the truncation symbol corresponding to the depth information is The function D(v) is replaced by the color value C(v)
  • the projection relationship between the voxel and the depth information in the first pose is replaced by the projection relationship between the voxel and the target image in the first pose, and so on.
  • the de-fusion method can be flexibly replaced and extended with reference to the above disclosed embodiments.
  • the first target set is At least one voxel is de-fused to obtain a third target set.
  • the information fused in the first target set includes multiple forms of data, respectively, according to the different forms of the information fused in the first target set.
  • the information fused in the first target set may include semantic information.
  • the information fused in the first target set is removed. Fusion can include:
  • At least one voxel in the voxel information set can be projected to the target image, and according to the semantic information of at least one pixel in the target image, the project after projection can be determined.
  • Semantic probability distribution information of at least one voxel of Therefore, in the process of de-fusion of semantic information, the voxels in the first target set can also be projected to the loop closure image in the first pose, so as to determine the semantics of at least one pixel in the fused loop closure image based on the projection result.
  • the information is de-fused to obtain a third target set.
  • the semantic probability distribution information P(l) of the voxels before the semantic information of the current target image can be fused i
  • I 1,...,k ), and the semantic information probability distribution P(O u(v,k) l i
  • a third voxel is obtained. target collection.
  • the reverse operation of the semantic information fusion process can be used to realize the independent de-fusion of semantic information.
  • the information of the loopback image can be fused into the third target set according to the second pose corresponding to the loopback image to obtain the second target set.
  • the information content included in the information of the loopback image may not be limited in the embodiments of the present disclosure, and reference may be made to the implementation form of the information fused in the first target set in the above-mentioned disclosed embodiments, such as semantic information, depth information or color information, etc.
  • the method of merging the information of the loopback image into the third target set according to the second pose corresponding to the loopback image you can also refer to the above disclosed embodiments for various types of information (such as semantic information, depth information or color information). etc.) fusion method, the only difference is that the first pose in the above fusion process is replaced with the second pose.
  • at least one voxel in the third target set may be projected to the loopback image under the second pose, and at least one voxel in the target image may be projected according to at least one pixel in the target image.
  • the semantic information is determined, the semantic probability distribution information of at least one voxel after projection is determined, and the second target set is obtained.
  • the information fused in the first target set is de-fused to obtain a third target set, and then according to the second pose corresponding to the loopback image, the information of the loopback image is fused to the third target set.
  • a second target set is obtained.
  • map data After the map data is obtained by any combination of the above disclosed embodiments, corresponding applications can be executed based on the obtained map data, for example, in scenarios such as control of indoor robots or AR.
  • continuous semantic information is fused into the map data obtained in the embodiments of the present disclosure, it can be used to control the robot to perform some tasks related to semantic information in the target scene.
  • the robot can be controlled to perform corresponding operations on the target object in the target scene, such as picking up a water glass on the table.
  • the map data obtained in the embodiments of the present disclosure may also be applied to an AR platform with a semantic composition function, and the like.
  • FIG. 3 shows a schematic frame diagram of a data generating apparatus according to an embodiment of the present disclosure.
  • the data generating apparatus 30 may include:
  • the voxel information set establishment part 31 is configured to determine the first pose of the target image in the target scene, and according to the first pose, establish the voxel information set of the target image, wherein the target image includes the data obtained from the target scene. at least one frame of image.
  • the semantic segmentation part 32 is configured to perform semantic segmentation on the target image to obtain semantic information of the target image.
  • the fusion part 33 is configured to fuse the semantic information into the voxel information set to obtain the fusion voxel information set.
  • the data generating part 34 is configured to obtain map data of the target scene according to the fusion voxel information set corresponding to the target image.
  • the fusion part 33 is further configured to: project at least one voxel in the voxel information set to the target image, and determine at least one projected voxel according to the semantic information of at least one pixel in the target image The semantic probability distribution information of voxels is obtained to obtain a fusion voxel information set.
  • the fusion part 33 is further configured to: acquire collection data obtained by performing data collection on the target scene, and fuse the collection data into a voxel information set to obtain a fusion voxel information set.
  • the fusion part 33 is further configured to: according to the fusion weight of the collected data and the projection relationship between at least one voxel in the voxel information set and the collected data, perform at least one of the voxel information sets on the collected data.
  • a voxel is fused to obtain a fusion voxel information set.
  • the target image includes at least two to-be-processed images
  • the data generation part 34 is further configured to: combine the fusion voxel information sets corresponding to the at least two to-be-processed images to obtain a map of the target scene data.
  • the data generating apparatus 30 further includes: a detection part, configured to perform loop closure detection on at least one frame of the target image, and use the target image whose loop closure is detected as the loop closure image; a determination part, configured to determine the loop closure The second pose of the image, wherein the accuracy of the second pose is higher than the accuracy of the first pose of the target image corresponding to the loopback image; the update part is configured to, according to the second pose of the loopback image, update the target scene update the map data.
  • the updating part is further configured to: obtain a fusion voxel information set corresponding to the loopback image in the first pose as the first target set; based on the first pose and the first target set corresponding to the loopback image In two poses, the information fused in the first target set is re-fused to obtain a second target set; the map data of the target scene is updated according to the second target set.
  • the updating part is further configured to: de-fuse the information fused in the first target set according to the first pose corresponding to the loopback image to obtain a third target set; In the second pose, the information of the loopback image is fused into the third target set to obtain the second target set.
  • the embodiments of the present disclosure provide an example of an application scenario, that is, the application example of the present disclosure proposes a data generation method, which can generate high-quality map data including continuous semantic information.
  • FIG. 4 shows a schematic flowchart of obtaining map data according to an application example of the present disclosure. It can be seen from FIG. 4 that in the application example of the present disclosure, the data generation method may include the following processes:
  • the 6DOF pose is estimated using the target image and IMU data, and loop closure detection is performed.
  • the target image (RGB image) and IMU data obtained by data collection of the target scene by the sensor device, namely 402 and 403, can be obtained by means of tight coupling, relocation, self-calibration, nonlinear optimization and global positioning.
  • the second step is to perform semantic segmentation based on convolutional neural network.
  • the target image obtained in the first step can be subjected to semantic segmentation through a convolutional neural network for image segmentation, and a semantic segmentation result of each target image, ie, 406 , can be obtained.
  • the convolutional neural network used for image segmentation can be obtained by training the training image, wherein the training image contains the semantic labeling data of each pixel, and the labeling data can be obtained by the relevant labeling method or labeling tool.
  • the trained convolutional neural network for image segmentation has relatively accurate segmentation results for images collected in an indoor environment, and has a certain generalization ability.
  • the third step is to integrate the target image, depth information and semantic segmentation results of a single frame into the map data.
  • depth information of the target scene is also obtained, that is, 401 .
  • this step based on the first pose determined in the first step, that is, 404 and the current depth information 401, in the map data including the hash table and the voxel block data unit, create a corresponding map for the current target image A collection of voxel information.
  • the voxel information set corresponding to the current target image is updated to obtain a fusion voxel information set, that is, 409 . Since most of the voxels in the map data are invisible in the field of view corresponding to the current target image, the fusion process of various types of information can be accelerated.
  • the fusion voxel information sets corresponding to different target images can be combined to form a data unit of fusion voxel information, which can be used as the map data of the target scene, or the fusion voxel information set corresponding to at least one target image can be stored.
  • the map data 410 is obtained from the map data of the target scene.
  • the depth information and the color information may be fused in a sliding average manner.
  • semantic information may be fused based on Bayes' theorem, and the fusion process may also refer to the above disclosed embodiments. and formula (1).
  • FIG. 5 shows a schematic diagram of semantic effect fusion of the target scene in an application example of the present disclosure. As can be seen from FIG. 5 , after the target scene is fused with semantic information, the semantic information of different regions can be effectively and continuously expressed.
  • the pose optimization of the loopback image is performed and the map data is updated online in real time.
  • the optimized second pose of the loopback image namely 407
  • the map is updated online in real time by means of re-fusion, namely 408.
  • the re-fusion may include two processes of de-fusion and re-fusion, and reference may be made to the above disclosed embodiments for the implementation form.
  • map data can be represented by voxels, and the probability distribution of semantic labels is fused in the voxel information set corresponding to the voxels, so that a grid with semantic labels can be generated, and high-quality to represent consecutive scenes.
  • various types of information fused in the map data can be updated online in real time by re-fusion, such as the Signed Distance Field (SDF) value representing depth information, color Probability distributions of informative and semantic labels to remove distortions in the map in time.
  • SDF Signed Distance Field
  • the writing order of each step does not mean a strict execution order but constitutes any limitation on the implementation process, and the specific execution order of each step should be based on its function and possible Internal logic is determined.
  • Embodiments of the present disclosure further provide a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, implement the data generation method described in the foregoing method embodiments.
  • the computer-readable storage medium may be a volatile computer-readable storage medium or a non-volatile computer-readable storage medium.
  • Embodiments of the present disclosure further provide a computer program, where the computer program includes computer-readable codes, and when the computer-readable codes are executed in an electronic device, the processor of the electronic device executes any of the foregoing implementations The data generation method described in the example.
  • Embodiments of the present disclosure further provide another computer program product, including a computer-readable storage medium storing program codes, where the instructions included in the program codes can be configured to execute the data generation methods described in the foregoing method embodiments. Method Examples.
  • An embodiment of the present disclosure further provides an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to call instructions stored in the memory to execute the methods described in the foregoing method embodiments data generation method.
  • the above-mentioned memory may be a volatile memory (volatile memory), such as a random access memory (Random Access Memory, RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (Read Only Memory) -Only Memory, ROM), flash memory (flash memory), hard disk (Hard Disk Drive, HDD) or solid-state drive (Solid-State Drive, SSD); or a combination of the above types of memory, and provide instructions and data.
  • volatile memory such as a random access memory (Random Access Memory, RAM)
  • non-volatile memory such as a read-only memory (Read Only Memory) -Only Memory, ROM), flash memory (flash memory), hard disk (Hard Disk Drive, HDD) or solid-state drive (Solid-State Drive, SSD); or a combination of the above types of memory, and provide instructions and data.
  • the above processor may be an application specific integrated circuit (ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processor Device, DSPD), a programmable logic device (Programmable Logic Device) , PLD), at least one of field programmable gate array (Field Programmable Gate Array, FPGA), central processing unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor.
  • ASIC application specific integrated circuit
  • DSP digital signal processor
  • DSPD digital signal processing device
  • DSPD Digital Signal Processor Device
  • PLD programmable logic device
  • FPGA Field Programmable Gate Array
  • CPU Central Processing Unit
  • controller microcontroller
  • the electronic device may be provided as a terminal, server or other form of device.
  • an embodiment of the present disclosure further provides a computer program, which implements the above method when the computer program is executed by a processor.
  • FIG. 6 is a schematic structural block diagram of an electronic device 800 according to an embodiment of the present disclosure.
  • the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, and a terminal such as a PDU.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input/output (I/O) Interface 812 , sensor component 814 , and communication component 816 .
  • the processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations.
  • the processing component 802 can include one or more processors 820 to execute instructions to perform all or some of the steps of the methods described above.
  • processing component 802 may include one or more modules that facilitate interaction between processing component 802 and other components.
  • processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.
  • Memory 804 is configured to store various types of data to support operation at electronic device 800 . Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phonebook data, messages, pictures, videos, and the like.
  • the memory 804 may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (Electrically Erasable) Erasable Programmable Read Only Memory, EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), ROM, Magnetic Memory, Flash Memory , disk or disc.
  • SRAM Static Random-Access Memory
  • EEPROM Electrically Erasable Programmable Read-Only Memory
  • EPROM Erasable Programmable Read Only Memory
  • PROM Programmable Read Only Memory
  • ROM Magnetic Memory
  • Flash Memory disk or disc.
  • Power supply assembly 806 provides power to various components of electronic device 800 .
  • Power supply components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800 .
  • Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from a user.
  • the touch panel includes one or more touch sensors to sense touch, swipe, and gestures on the touch panel. The touch sensor may not only sense the boundaries of a touch or swipe action, but also detect the duration and pressure associated with the touch or swipe action.
  • the multimedia component 808 includes a front-facing camera and/or a rear-facing camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera may receive external multimedia data. Each of the front and rear cameras can be a fixed optical lens system or have focal length and optical zoom capability.
  • Audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (Microphone, MIC) configured to receive external audio signals when the electronic device 800 is in an operating mode, such as a calling mode, a recording mode, and a voice recognition mode.
  • the received audio signal may be stored in memory 804 or transmitted via communication component 816 .
  • the audio component 810 further includes a speaker configured to output audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module, which may be a keyboard, a click wheel, a button, or the like. These buttons may include, but are not limited to: home button, volume buttons, start button, and lock button.
  • Sensor assembly 814 includes one or more sensors configured to provide status assessment of various aspects of electronic device 800 .
  • the sensor assembly 814 can detect the on/off state of the electronic device 800, the relative positioning of the components, such as the display and the keypad of the electronic device 800, the sensor assembly 814 can also detect the electronic device 800 or one of the electronic device 800 Changes in the position of components, presence or absence of user contact with the electronic device 800 , orientation or acceleration/deceleration of the electronic device 800 and changes in the temperature of the electronic device 800 .
  • Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects in the absence of any physical contact.
  • Sensor assembly 814 may also include a light sensor, such as a Complementary Metal Oxide Semiconductor (CMOS) or Charge-coupled Device (CCD) image sensor, configured for use in imaging applications.
  • CMOS Complementary Metal Oxide Semiconductor
  • CCD Charge-coupled Device
  • the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices.
  • the electronic device 800 can access a wireless network based on a communication standard, such as a wireless network (Wi-Fi), a second-generation mobile communication technology (2-Generation, 2G) or a third-generation mobile communication technology (3rd-Generation, 3G), or their combination.
  • the communication component 816 receives broadcast signals or broadcast related personnel information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication.
  • the NFC module may be based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BitTorrent, BT) technology and other technology to achieve.
  • RFID Radio Frequency Identification
  • IrDA Infrared Data Association
  • UWB Ultra Wide Band
  • Bluetooth BitTorrent
  • electronic device 800 may be implemented by one or more ASICs, DSPs, DSPDs, PLDs, FPGAs, controllers, microcontrollers, microprocessors, or other electronic components for performing the above-described methods.
  • a non-volatile computer-readable storage medium such as a memory 804 comprising computer program instructions executable by the processor 820 of the electronic device 800 to perform the above method is also provided.
  • FIG. 7 is a schematic block diagram of the architecture of an electronic device 1900 according to an embodiment of the present disclosure.
  • the electronic device 1900 may be provided as a server.
  • an electronic device 1900 includes a processing component 1922, which in some embodiments of the present disclosure includes one or more processors, and a memory resource, represented by memory 1932, for storing instructions executable by the processing component 1922 , such as applications.
  • An application program stored in memory 1932 may include one or more modules, each corresponding to a set of instructions.
  • the processing component 1922 is configured to execute instructions to perform the above-described methods.
  • the electronic device 1900 may also include a power supply assembly 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input output (I/O) interface 1958 .
  • the electronic device 1900 can operate based on an operating system stored in the memory 1932, such as a Microsoft server operating system (Windows ServerTM), a graphical user interface based operating system (Mac OS XTM) introduced by Apple, a multi-user multi-process computer operating system (UnixTM). ), Free and Open Source Unix-like Operating System (LinuxTM), Open Source Unix-like Operating System (FreeBSDTM) or the like.
  • a non-volatile computer-readable storage medium such as memory 1932 comprising computer program instructions executable by processing component 1922 of electronic device 1900 to perform the above-described method.
  • Embodiments of the present disclosure may be systems, methods and/or computer program products.
  • the computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the embodiments of the present disclosure.
  • a computer-readable storage medium may be a tangible device that can hold and store instructions for use by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • Computer-readable storage media include: portable computer disks, hard disks, RAM, ROM, EPROM or flash memory, SRAM, portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROMs), Digital Video Discs (DVDs), memory sticks, floppy disks, mechanically encoded devices, such as punch cards or raised structures in grooves on which instructions are stored, and any suitable combination.
  • Computer-readable storage media, as used herein, are not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (eg, light pulses through fiber optic cables), or through electrical wires transmitted electrical signals.
  • the computer readable program instructions described herein may be downloaded to various computing/processing devices from a computer readable storage medium, or to an external computer or external storage device over a network such as the Internet, a local area network, a wide area network, and/or a wireless network.
  • the network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
  • a network adapter card or network interface in each computing/processing device receives computer-readable program instructions from a network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing/processing device .
  • Computer program instructions for performing operations of embodiments of the present disclosure may be assembly instructions, Industry Standard Architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or in a form of Source or object code in any combination of programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as the "C" language or similar programming language.
  • the computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server implement.
  • the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or Wide Area Network (WAN), or may be connected to an external computer (eg, use an internet service provider to connect via the internet).
  • LAN Local Area Network
  • WAN Wide Area Network
  • electronic circuits such as programmable logic circuits, FPGAs, or Programmable Logic Arrays (PLAs), that can execute computer Program instructions are readable to implement various aspects of the present disclosure.
  • PDAs Programmable Logic Arrays
  • These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer or other programmable data processing apparatus to produce a machine that causes the instructions when executed by the processor of the computer or other programmable data processing apparatus , resulting in means for implementing the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.
  • These computer readable program instructions can also be stored in a computer readable storage medium, these instructions cause a computer, programmable data processing apparatus and/or other equipment to operate in a specific manner, so that the computer readable medium on which the instructions are stored includes An article of manufacture comprising instructions for implementing various aspects of the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.
  • Computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other equipment to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other equipment to produce a computer-implemented process , thereby causing instructions executing on a computer, other programmable data processing apparatus, or other device to implement the functions/acts specified in one or more blocks of the flowcharts and/or block diagrams.
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more functions for implementing the specified logical function(s) executable instructions.
  • the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
  • each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations can be implemented in dedicated hardware-based systems that perform the specified functions or actions , or can be implemented in a combination of dedicated hardware and computer instructions.
  • Embodiments of the present disclosure provide a data generation method, apparatus, device, storage medium, and program.
  • the method includes: determining a first pose of a target image in a target scene, and establishing a voxel information set of the target image according to the first pose, wherein the target image includes data performed on the target scene. collecting the obtained at least one frame of image; performing semantic segmentation on the target image to obtain semantic information of the target image; fusing the semantic information into the voxel information set to obtain a fusion voxel information set; according to The fused voxel information set corresponding to the target image is used to obtain map data of the target scene.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Peptides Or Proteins (AREA)
  • Saccharide Compounds (AREA)
  • Medicines That Contain Protein Lipid Enzymes And Other Medicines (AREA)
  • Processing Or Creating Images (AREA)
  • Image Processing (AREA)

Abstract

本公开实施例提供了一种数据生成方法、装置、设备、存储介质及程序。所述方法包括：确定目标场景中目标图像的第一位姿，根据所述第一位姿，建立所述目标图像的体素信息集合，其中，所述目标图像包括对所述目标场景进行数据采集所得到的至少一帧图像；对所述目标图像进行语义分割，得到所述目标图像的语义信息；将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合；根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据。本公开实施例可以提升得到的地图数据的数据全面性和质量。

Description

数据生成方法、装置、设备、存储介质及程序
相关申请的交叉引用
本专利申请要求2021年03月02日提交的中国专利申请号为202110231700.1、申请人为浙江商汤科技开发有限公司，申请名称为“数据生成方法及装置、电子设备和存储介质”的优先权，该申请的全文以引用的方式并入本申请中。
技术领域
本公开涉及计算机视觉领域，尤其涉及一种数据生成方法、装置、设备、存储介质及程序。
背景技术
随着计算机视觉技术的快速发展，场景重建成为计算机视觉领域中一个愈加重要的应用。相关技术中，场景重建通常包含场景内各目标物体的形状位置等几何信息。
当重建的场景需要与真实世界交互的情况下，如应用于室内机器人或增强现实(Augmented Reality，AR)等场景中的情况下，可通过获取场景中目标物体的几何信息以实现交互效果。
发明内容
本公开实施例提出了一种数据生成方案。
本公开实施例提供了一种数据生成方法，所述方法由电子设备执行，所述方法包括：
确定目标场景中目标图像的第一位姿，根据所述第一位姿，建立所述目标图像的体素信息集合，其中，所述目标图像包括对所述目标场景进行数据采集所得到的至少一帧图像；
对所述目标图像进行语义分割，得到所述目标图像的语义信息；
将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合；
根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据。
如此，通过目标图像的体素信息集合融合目标图像中分割得到的语义信息，从而实现目标场景中语义信息的连续融合，得到包含连续融合的语义信息的地图数据，能够有效提升得到的地图数据的数据全面性和质量。
在本公开的一些实施例中，所述将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合，包括：将所述体素信息集合中至少一个体素投影至所述目标图像，根据所述目标图像中至少一个像素的语义信息，确定投影后的所述至少一个体素的语义概率分布信息，得到所述融合体素信息集合。如此，通过将体素信息集合中的体素投影至目标图像，同时将二维目标图像的语义信息融合至三维体素的二维投影之中，从而使得融合体素信息集合中包含语义信息的概率分布情况，可以高质量地对目标场景中的语义信息进行连续地表达，并使得基于该融合体素信息集合所得到的目标场景的地图数据可以更广泛地应用于需要语义信息的众多场景中。
在本公开的一些实施例中，所述将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合，还包括：获取对所述目标场景进行数据采集所得到的采集数据，将所述采集数据融合至所述体素信息集合中，得到所述融合体素信息集合。如此，可以增加融合体素信息结合中数据的全面性，基于该融合体素信息集合所得到的目标场景的地图数据可以更加地全面和完善，具有更高的质量。
在本公开的一些实施例中，所述将所述采集数据融合至所述体素信息集合中，得到融合体素信息集合，包括：根据所述采集数据的融合权重，以及所述体素信息集合中至少一个体素与所述采集数据之间的投影关系，对所述体素信息集合中至少一个体素进行信息融合，得到所述融合体素信息集合。如此，可以在采集数据包括多种形式数据的情 况下，分别根据采集数据的不同形式来对不同的信息进行灵活融合，既能够提升融合体素信息集合的数据完整性，且能够提高融合效率，从而能够提升得到的目标场景的地图数据的全面性以及数据生成的效率。
在本公开的一些实施例中，所述目标图像包括至少两个待处理图像，所述根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据，包括：对所述至少两个待处理图像对应的所述融合体素信息集合进行合并，得到所述目标场景的地图数据。如此，通过将至少一个目标图像对应的融合体素信息集合进行合并得到地图数据，能够有效提升地图数据的数据完整性和全面性，且在包含多个连续的目标图像的情况下，地图数据可以对目标场景进行连续的表示。
在本公开的一些实施例中，所述方法还包括：对至少一帧所述目标图像进行回环检测，将检测到回环的目标图像作为回环图像；确定所述回环图像的第二位姿，其中，所述第二位姿的准确度高于所述回环图像对应的目标图像的第一位姿的准确度；根据所述回环图像的第二位姿，对所述目标场景的地图数据进行更新。如此，得到的第二位姿相较于该回环图像对应的目标图像所确定的第一位姿来说，具有更高的准确度，从而使得基于第二位姿更新的地图数据具有更高的精度。
在本公开的一些实施例中，所述根据所述回环图像的第二位姿，对所述目标场景的地图数据进行更新，包括：获取所述回环图像在所述第一位姿下对应的融合体素信息集合，作为第一目标集合；基于所述回环图像对应的第一位姿和第二位姿，对所述第一目标集合中融合的信息进行重新融合，得到第二目标集合；根据所述第二目标集合，对所述目标场景的地图数据进行更新。如此，通过重新融合的方式，实现地图数据中位姿和各类信息的分别优化，与对地图数据中的各类信息进行联合优化更新相比，能够有效减小地图数据更新过程中的计算量，以及提高地图数据的更新效率，实现地图数据的实时更新。
在本公开的一些实施例中，所述基于所述回环图像对应的第一位姿和第二位姿，对所述第一目标集合中融合的信息进行重新融合，得到第二目标集合，包括：根据所述回环图像对应的第一位姿，对所述第一目标集合中融合的信息进行去融合，得到第三目标集合；根据所述回环图像对应的第二位姿，将所述回环图像的信息融合至所述第三目标集合中，得到所述第二目标集合。如此，可以在第一目标集合中融合的信息包括多种形式数据的情况下，分别根据第一目标集合中融合的信息的不同形式来对不同的信息进行灵活去融合，能够提高去融合的效率和灵活性，从而提升地图数据的更新效率。
以下装置、电子设备等的效果描述参见上述数据生成方法的说明。
本公开实施例提供了一种数据生成装置，所述装置包括：
体素信息集合建立部分，配置为确定目标场景中目标图像的第一位姿，根据所述第一位姿，建立所述目标图像的体素信息集合，其中，所述目标图像包括对所述目标场景进行数据采集所得到的至少一帧图像；
语义分割部分，配置为对所述目标图像进行语义分割，得到所述目标图像的语义信息；
融合部分，配置为将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合；
数据生成部分，配置为根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据。
在本公开的一些实施例中，所述融合部分，还配置为：将所述体素信息集合中至少一个体素投影至所述目标图像，根据所述目标图像中至少一个像素的语义信息，确定投影后的所述至少一个体素的语义概率分布信息，得到所述融合体素信息集合。
在本公开的一些实施例中，所述融合部分，还配置为：获取对所述目标场景进行数 据采集所得到的采集数据，将所述采集数据融合至所述体素信息集合中，得到所述融合体素信息集合。
在本公开的一些实施例中，所述融合部分，还配置为：根据所述采集数据的融合权重，以及所述体素信息集合中至少一个体素与所述采集数据之间的投影关系，对所述体素信息集合中至少一个体素进行信息融合，得到所述融合体素信息集合。
在本公开的一些实施例中，所述目标图像包括至少两个待处理图像，所述数据生成部分，还配置为：对所述至少两个待处理图像对应的所述融合体素信息集合进行合并，得到所述目标场景的地图数据。
在本公开的一些实施例中，所述装置还包括：检测部分，配置为对至少一帧所述目标图像进行回环检测，将检测到回环的目标图像作为回环图像；确定部分，配置为确定所述回环图像的第二位姿，其中，所述第二位姿的准确度高于所述回环图像对应的目标图像的第一位姿的准确度；更新部分，配置为根据所述回环图像的第二位姿，对所述目标场景的地图数据进行更新。
在本公开的一些实施例中，所述更新部分，还配置为：获取所述回环图像在所述第一位姿下对应的融合体素信息集合，作为第一目标集合；基于所述回环图像对应的第一位姿和第二位姿，对所述第一目标集合中融合的信息进行重新融合，得到第二目标集合；根据所述第二目标集合，对所述目标场景的地图数据进行更新。
在本公开的一些实施例中，所述更新部分，还配置为：根据所述回环图像对应的第一位姿，对所述第一目标集合中融合的信息进行去融合，得到第三目标集合；根据所述回环图像对应的第二位姿，将所述回环图像的信息融合至所述第三目标集合中，得到所述第二目标集合。
本公开实施例还提供了一种电子设备，包括：
处理器；用于存储处理器可执行指令的存储器；其中，所述处理器被配置为：执行上述任一实施例所述的数据生成方法。
本公开实施例还提供了一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述任一实施例所述的数据生成方法。
本公开实施例还提供一种计算机程序，所述计算机程序包括计算机可读代码，在所述计算机可读代码在电子设备中运行的情况下，所述电子设备的处理器执行如上述任一实施例所述的数据生成方法。
本公开实施例至少提供一种数据生成方法、装置、设备、存储介质及程序，通过确定目标场景中目标图像的第一位姿，根据第一位姿建立目标图像的体素信息集合，并将对目标图像进行语义分割得到的语义信息融合到体素信息集合中，得到融合体素信息集合并根据融合体素信息集合得到目标场景的地图数据。如此，可以通过建立体素信息集合来融合目标图像中分割得到的语义信息，从而实现目标场景中语义信息的连续融合，得到包含连续融合的语义信息的地图数据，能够有效提升得到的地图数据的数据全面性和质量。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本公开实施例。根据下面参考附图对示例性实施例的详细说明，本公开实施例的其它特征及方面将变得清楚。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。
图1示出根据本公开一实施例的数据生成方法的流程示意图；
图2示出应用本公开实施例的数据生成方法的一种系统架构示意图；
图3示出根据本公开一实施例的数据生成装置的框架示意图；
图4示出根据本公开一应用示例得到地图数据的流程示意图；
图5示出根据本公开一应用示例中目标场景的语义效果融合示意图；
图6示出根据本公开实施例的一种电子设备800的架构示意框图；
图7示出根据本公开实施例的一种电子设备1900的架构示意框图。
具体实施方式
以下将参考附图详细说明本公开的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括A、B、C中的至少一种，可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
本公开实施例中的多个或者多种可以分别指的是至少两个或者至少两种。
另外，为了更好地说明本公开，在下文的具体实施方式中给出了众多的实施细节。本领域技术人员应当理解，没有某些细节，本公开同样可以实施。在一些实施例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本公开的主旨。
图1示出根据本公开一实施例的数据生成方法的流程示意图，该方法可以应用于数据生成装置，数据生成装置可以为终端设备、服务器或者其他处理设备等。其中，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字处理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备以及可穿戴设备等。在本公开的一些实施例中，该数据生成方法可以应用于云端服务器或本地服务器，云端服务器可以为公有云服务器，也可以为私有云服务器，根据实际情况灵活选择即可。
在本公开的一些实施例中，该数据生成方法也可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
如图1所示，在本公开的一些实施例中，所述数据生成方法可以包括：
步骤S11，确定目标场景中目标图像的第一位姿，根据第一位姿，建立目标图像的体素信息集合。
其中，目标图像包括对目标场景进行数据采集所得到的至少一帧图像。
步骤S12，对目标图像进行语义分割，得到目标图像的语义信息。
步骤S13，将语义信息融合至体素信息集合中，得到融合体素信息集合。
步骤S14，根据目标图像对应的融合体素信息集合，得到目标场景的地图数据。
其中，目标场景可以是任意具有重建需求的场景，其位置以及范围等均可以根据实际情况灵活选择，在本公开实施例中不做限制，不局限于下述各公开实施例。
在本公开的一些实施例中，目标场景可以是室内场景，也可以是室外场景，还可以同时包含室内场景与室外场景等。
在本公开的一些实施例中，目标图像可以是对目标场景进行数据采集所得到的至少一帧图像。目标图像包含的图像数量在本公开实施例中不做限制，可以根据对目标场景 进行数据采集的实际情况灵活决定。在本公开的一些实施例中，可以将对目标场景进行数据采集所得到的每帧图像均作为目标图像，也可以从对目标场景进行数据采集所得到的多帧图像中，选定一帧或多帧来作为目标图像，选定的方式可以根据实际情况灵活选择。在本公开的一些实施例中，可以为随机选定，也可以按照一定的频率对采集的图像进行采样来选定得到目标图像等。
其中，对目标场景进行数据采集的方式在本公开实施例中不做限定，可以根据实际情况灵活决定，不局限于下述各公开实施例。在本公开的一些实施例中，可以通过摄像机或照相机等图像采集设备对目标场景进行数据采集。在本公开的一些实施例中，也可以通过包含图像采集设备的其他装置对目标场景进行数据采集，在本公开的一些实施例中，可以通过包含惯性测量单元(Inertial Measurement Unit，IMU)和作为图像采集设备的视觉传感器的集成设备对目标场景进行数据采集。比如：带有摄像头的智能手机等。在本公开的一些实施例中，在通过图像采集设备或是包含图像采集设备的其他装置来对目标场景进行数据采集的情况下，该图像采集设备还可以具有采集深度信息的功能，举例来说，在本公开的一些实施例中，图像采集设备可以包括飞行时间(Time Of Flight，TOF)相机。
随着数据采集方式的不同，采集到的采集数据也可以发生灵活的变化，可以根据实际情况灵活决定，不局限于下述各公开实施例。在本公开的一些实施例中，采集数据可以包括如上述公开实施例所述的至少一帧目标图像；在本公开的一些实施例中，采集数据也可以包括其他数据。在本公开的一些实施例中，在通过包含IMU的集成设备对目标场景进行数据采集的情况下，采集数据还可以包括IMU数据。在本公开的一些实施例中，在图像采集设备包括TOF相机的情况下，采集数据还可以包括目标图像的深度信息等。
基于上述实施例，步骤S11可以基于目标场景中的目标图像，来确定目标图像的第一位姿。其中，图像的位姿可以是采集该图像的设备的位姿，基于图像的位姿，可以确定图像与世界坐标系之间的坐标对应关系，从而确定图像中各像素点在空间中的位置等。在本公开的一些实施例中，第一位姿可以是在图像采集设备采集目标图像的情况下，图像采集设备的位姿。
第一位姿的确定方式在本公开实施例中不做限制，任何确定图像位姿的方法均可以作为步骤S11中确定第一位姿的方法，不局限于下述各公开实施例。在本公开的一些实施例中，可以仅基于目标场景中的目标图像，进行图像位姿估计，来确定目标图像的第一位姿。在本公开的一些实施例中，在采集数据包括IMU数据的情况下，也可以利用同步定位与建图(Simultaneous Localization And Mapping，SLAM)系统，通过视觉惯性里程计(Visual–Inertial Odometry，VIO)方法对目标图像进行位姿估计得到六自由度的第一位姿；在本公开的一些实施例中，在采集数据包括深度信息的情况下，还可以基于深度信息辅助进行位姿估计，来确定目标图像的第一位姿。
在本公开的一些实施例中，步骤S11还可以根据目标图像的第一位姿，来建立目标图像的体素信息集合。其中，体素(Voxel)是体积元素(Volume Pixel)的简称，是数字数据于三维空间分割上的最小单位。在本公开的一些实施例中，体素可以在本公开实施例中用来作为目标场景中各位置的表现方式。
其中，由于目标图像为目标场景中的图像，可以表示目标场景中在某个或某些视野下的场景内容。因此，在本公开的一些实施例中，目标图像可以对应目标场景中至少部分位置，因此，目标图像中的像素点与表示目标场景的至少部分体素相互对应。同时在本公开的一些实施例中，可以根据目标图像的第一位姿来建立目标图像的体素信息集合，该体素信息集合可以包含目标场景中至少部分体素的相关体素信息，来实现目标图像与表示目标场景的至少部分体素的对应。
其中，体素信息集合中包含的相关体素信息，其包含的信息内容可以根据实际情况 灵活决定，不局限于下述各公开实施例。在本公开的一些实施例中，体素信息集合可以包括：体素信息集合v在融合过程中的融合权重W(v)、目标图像的图像信息如颜色值C(v)以及目标图像的深度信息如截断符号函数D(v)等。在本公开的一些实施例中，目标图像的图像信息以及深度信息等也可以通过其他函数来进行表示；在本公开的一些实施例中，体素信息集合还可以包括其他地图数据所需的相关信息，可以根据实际情况灵活扩展，此不再一一列举。在本公开的一些实施例中，由于本公开实施例中的体素信息集合包含的是体素的相关信息，而作为三维数据的体素需要通过深度来确定相关信息，因此，以下各公开实施例均以采集数据与体素信息集合包含深度信息，且对体素信息集合进行融合的过程中，均需对深度信息进行融合为例来进行说明。
步骤S11中，根据第一位姿建立目标图像的体素信息集合的方式在本公开实施例中不做限制，可以根据实际情况灵活决定，不局限于下述各公开实施例。在本公开的一些实施例中，可以通过体素哈希(voxel hashing)方法来建立各目标图像的体素信息集合，并利用哈希表对各体素信息集合进行存储与查找。在本公开的一些实施例中，也可以通过其他数据结构来对体素信息集合进行建立、存储与查找。其中，选择哪些数据结构，以及如何基于这些结构来建立各体素信息集合，其实现形式可以根据实际情况灵活选择，在此不再一一列举。
在本公开的一些实施例中，还可以通过步骤S12对目标图像进行语义分割，得到目标图像的语义信息。其中，步骤S11与步骤S12的实现顺序在本公开实施例中不做限制，步骤S11和步骤S12可以同时进行，也可以按照预设顺序依次执行，其中，选择何种执行顺序可以根据实际情况灵活决定。
步骤S12中，对目标图像进行语义分割可以是将目标图像中所包含的不同类别的对象进行分割，目标图像的语义信息可以是目标图像中至少一个像素点所对应的对象的类别信息。在本公开的一些实施例中，在目标场景为室内场景的情况下，目标图像可以是室内场景中的图像，则目标图像中可能包含墙对象、地面对象、天花板对象、桌子对象以及椅子对象等，则通过对目标图像进行语义分割，可以将目标图像中包含的墙、地面、天花板、桌子以及椅子等对象进行分割，即可以确定目标图像中哪些像素点对应墙的类别信息，哪些像素点对应地面的类别信息等。在本公开的一些实施例中，随着目标场景的不同，目标图像的语义信息所对应的类别也可能发生变化，还可能包含其他的室内对象类别，比如：柜子或窗户等，也可能包含其他的室外对象类别，比如：天空、树木或是马路等。
对目标图像进行语义分割的方式在本公开实施例中不做限制，任何可以对目标图像进行分割的方式，均可以作为本公开实施例中语义分割的实现方式，不局限于下述各公开实施例。在本公开的一些实施例中，可以通过分割算法对目标图像进行处理，得到目标图像的语义信息。在本公开的一些实施例中，也可以将目标图像输入图像分割神经网络，来得到图像分割神经网络所输出的语义信息。
步骤S12中，对目标图像进行语义分割，可以是每次对一张目标图像进行语义分割，也可以是每次对多张目标图像进行语义分割，语义分割所对应的目标图像的数量可以根据实际情况灵活决定，在本公开实施例中不做限制。
在建立目标图像的体素信息集合，以及得到目标图像的语义信息以后，可以通过步骤S13，将语义信息融合至对应的体素信息集合中，得到融合体素信息集合。其中，如上述各公开实施例所述，体素信息集合中可以包含的信息内容可以根据实际情况灵活选择，因此融合体素信息集合中除了可以融合语义信息以外，也可以考虑融合其他信息。其中，融合哪些信息可以根据实际情况灵活选择。通过将语义信息融合至体素信息集合中，可以使得体素信息集合在包含各表示目标对象的形状位置等集合信息的基础上，能够融合进语义信息，从而能够更加高质量地对目标场景进行表达。
步骤S13中将语义信息融合至体素信息集合的方式在本公开实施例中不做限制，可以根据实际情况灵活选择，详见下述各公开实施例，在此先不做展开。
在得到融合体素信息集合后，可以根据目标图像对应的融合体素信息集合，来得到目标场景的地图数据。目标场景的地图数据可以是对目标场景中的各帧目标图像进行汇总所得到的数据，其包含的数据内容可以根据实际情况灵活决定。因此，步骤S14的实现方式可以根据地图数据的实际数据要求所灵活决定。步骤S14的实现方式，可以详见下述各公开实施例，在此同样先不做展开。
在本公开一些实施例中，通过确定目标场景中目标图像的第一位姿，根据第一位姿建立目标图像的体素信息集合，并将对目标图像进行语义分割得到的语义信息融合到体素信息集合中，得到融合体素信息集合并根据融合体素信息集合得到目标场景的地图数据。如此，通过建立体素信息集合来融合目标图像中分割得到的语义信息，由于体素信息集合中的体素可以对目标场景在三维空间中的位置进行连续地表达，因此基于体素信息集合来融合语义信息，能够实现目标场景中语义信息的连续融合，得到包含连续融合的语义信息的地图数据，进而能够有效提升得到的地图数据的数据全面性和质量。
图2示出可以应用本公开实施例的数据生成方法的一种系统架构示意图；如图2所示，该系统架构中包括：获取终端201、网络202和数据生成终端203。为实现支撑一个示例性应用，当获取终端201和数据生成终端203通过网络202建立通信连接，获取终端201通过网络202向数据生成终端203上报目标场景中目标图像。数据生成终端203响应于目标场景中目标图像，首先，确定目标场景中目标图像的第一位姿，根据第一位姿，建立目标图像的体素信息集合；以及对目标图像进行语义分割，得到目标图像的语义信息；其次，将语义信息融合至体素信息集合中，得到融合体素信息集合；最后，根据目标图像对应的融合体素信息集合，得到目标场景的地图数据。同时数据生成终端203将目标场景的地图数据上传至网络202，并通过网络202发送给获取终端201。
作为示例，获取终端201可以包括图像采集设备，数据生成终端203可以包括具有视觉信息处理能力的视觉处理设备或远程服务器。网络202可以采用有线或无线连接方式。其中，当数据生成终端203为视觉处理设备时，获取终端201可以通过有线连接的方式与视觉处理设备通信连接，例如通过总线进行数据通信；当数据生成终端203为远程服务器时，获取终端201可以通过无线网络与远程服务器进行数据交互。
或者，在一些场景中，当获取终端201可以是带有视频采集模组的视觉处理设备，可以是带有摄像头的主机。这时，本公开实施例的数据生成方法可以由获取终端201执行，上述系统架构可以不包含网络202和数据生成终端203。
如上述各公开实施例所述，步骤S13的实现方式可以根据实际情况灵活决定。在本公开的一些实施例中，步骤S13包括：
将体素信息集合中至少一个体素投影至目标图像，根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到融合体素信息集合。
其中，将体素信息集合中至少一个体素投影至目标图像，投影的方式以及角度等可以根据实际情况灵活选择，在本公开实施例中不做限制。其中，将体素信息集合中哪些体素投影至目标图像，也可以根据实际情况灵活选择，在本公开的一些实施例中，可以将体素信息集合中包含的每个体素均投影至目标图像；在本公开的一些实施例中，也可以随机选择或对体素进行一定比例的采样选择后，将选择的体素投影至目标图像等。
在将至少一个体素投影至目标图像以后，可以根据目标图像中至少一个像素的语义信息，来确定投影后的至少一个体素的语义概率分布信息。其中，语义概率分布信息可以是多个体素的语义信息的概率分布情况，如何根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，其实现形式可以根据实际情况灵活决定。在本公开的一些实施例中，可以将目标图像中至少一个像素与投影后的体素进 行一一对应，并依据对应情况融合语义信息后，基于多个体素的语义信息的融合结果来确定至少一个体素的语义概率分布信息。在本公开的一些实施例中，也可以根据目标图像中至少一个像素的语义信息，来获取目标图像中语义信息的概率分布情况，并将其与投影后的体素的语义概率分布信息进行融合等。其中，如何实现可以根据实际情况灵活选择。在本公开的一些实施例中，将体素信息集合中至少一个体素投影至目标图像，根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到融合体素信息集合的方式可以通过下述公式(1)来进行表示：
Figure PCTCN2021105485-appb-000001
其中，I 1,...,k为各帧目标图像，I k为当前目标图像，P(l i|I 1,...,k)为融合当前目标图像的语义信息后体素属于第i类别的语义概率分布信息，Z为归一化因子，P(l i|I 1,...,k-1)为在融合当前目标图像的语义信息以前的(即融合I 1至I k-1帧目标图像的语义信息)体素属于第i类别的语义概率分布信息，P(O u(v,k)＝l i|I k)为对当前目标图像进行语义分割所得到的语义信息概率分布，同时O u(v,k)为将体素v投影至当前目标图像的投影结果。
通过上述公式(1)可以看出，可以将当前目标图像进行语义分割所得到的语义分割信息P(O u(v,k)＝l i|I k)与在当前目标图像以前的k-1帧目标图像所对应的体素的语义概率分布信息相乘并进行归一化处理，同时将当前目标图像的语义信息融合至体素信息集合中，从而得到融合体素信息集合。
通过将体素信息集合中至少一个体素投影至目标图像，根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到融合体素信息集合。如此，可以通过将体素信息集合中的体素投影至目标图像，同时将二维目标图像的语义信息融合至三维体素的二维投影之中，从而使得融合体素信息集合中包含语义信息的概率分布情况，可以高质量地对目标场景中的语义信息进行连续地表达，并使得基于该融合体素信息集合所得到的目标场景的地图数据可以更广泛地应用于需要语义信息的众多场景中。
如上述各公开实施例所述，融合体素信息集合中除了可以融合语义信息以外，也可以考虑融合其他信息。因此，在本公开的一些实施例中，步骤S13还可以包括：
获取对目标场景进行数据采集所得到的采集数据，将采集数据融合至体素信息集合中，得到融合体素信息集合。
其中，采集数据的实现方式可以参考上述各公开实施例。如上述各公开实施例所述，采集数据包含的数据内容可以根据数据采集的方式灵活，比如包含深度信息、IMU数据或是其他信息等，详见上述各公开实施例。
在本公开的一些实施例中，可以根据采集数据的实际情况，将采集数据中的全部或部分数据融合至体素信息集合中。同时选择哪些采集数据进行融合在本公开实施例中不做限制，可以根据实际情况灵活选择，不局限于下述各公开实施例。随着采集数据的不同，融合的方式也可以灵活发生变化详见下述各公开实施例，在此先不做展开。
通过获取对目标场景进行数据采集所得到的采集数据，将采集数据融合至体素信息集合中，得到融合体素信息集合。如此，能够增加融合体素信息结合中数据的全面性，基于该融合体素信息集合所得到的目标场景的地图数据可以更加地全面和完善，具有更 高的质量。
在本公开的一些实施例中，采集数据包括目标图像的深度信息，在本公开的一些实施例中，采集数据还可以包括颜色信息等，将采集数据融合至体素信息集合中，得到融合体素信息集合，可以包括：
根据采集数据的融合权重，以及体素信息集合中至少一个体素与采集数据之间的投影关系，对体素信息集合中至少一个体素进行信息融合，得到融合体素信息集合。
其中，在本公开的一些实施例中，深度信息可以是如上述公开实施例所述的，通过TOF相机所采集得到的信息。在本公开的一些实施例中，该深度信息可以通过截断符号函数D(v)的形式进行表示。在本公开的一些实施例中，颜色信息可以是从采集到的目标图像本身所获取的RGB颜色信息。在本公开的一些实施例中，该颜色信息可以通过颜色值C(v)的形式进行表示等。同时上述深度信息与颜色信息的获取方式与表现形式等均仅为示例性的实现形式，在实际的应用中，可以根据实际情况灵活选择其他的实现形式，本公开实施例对此不做限制。
随着采集数据形式的不同，对采集数据融合的方式可能会灵活发生变化。通过上述公开实施例可以看出，在采集数据包括深度信息和颜色信息中的至少之一的情况下，可以根据采集数据的融合权重，以及体素信息集合中至少一个体素与采集数据之间的投影关系，对采集数据进行融合。同时如何基于融合权重和体素的投影关系来进行融合，其实现过程可以灵活决定。在本公开的一些实施例中，深度信息的融合过程可以通过下述公式(2)和(3)进行表示：
Figure PCTCN2021105485-appb-000002
W'(v)＝W(v)+w i(v)        (3)；
其中，D'(v)为体素v融合后的深度信息，D(v)为体素v在融合前的深度信息，W(v)为根据体素信息集合中的信息所确定的体素v的权重，w i(v)为融合权重，d i(v)为体素v在深度信息中对应的反投影点与体素v之间的距离，W'(v)为体素v融合后更新的权重。
通过公式(2)和(3)可以看出，根据深度信息的融合权重w i(v)，以及体素v与深度信息之间的投影关系d i(v)，将深度信息融合至体素信息集合中。其中，深度信息的融合权重w i(v)可以根据实际情况灵活设定，而体素与深度信息之间的投影关系d i(v)也可以根据体素与深度信息的实际情况来灵活确定。在本公开的一些实施例中，d i(v)可以通过下述公式(4)和(5)进行计算：
Figure PCTCN2021105485-appb-000003
η＝d c(v)-X(v)     (5)；
其中，d c(v)为体素v到相机中心的距离，X(v)为体素v投影到深度信息后得到的对应像素的深度，μ为预设参数。
通过上述公开实施例可以得出，基于深度信息的融合权重，以及体素与深度信息之间的投影关系，将深度信息融合到体素信息集合中。在本公开的一些实施例中，在采集数据包括颜色信息的情况下，颜色信息的融合方式可以参考上述深度信息的融合方式，区别在于将深度信息对应的截断符号函数D(v)替换为颜色值C(v)，以及将体素与深度信息之间的投影关系替换为体素与目标图像之间的投影关系等。在采集数据包含其他数据形式的情况下，其融合方式可以参考上述各公开实施例进行灵活替换与扩展。
通过根据所述采集数据的融合权重，以及体素信息集合中至少一个体素与采集数据之间的投影关系，对体素信息集合中至少一个体素进行信息融合，得到融合体素信息集合。如此，可以在采集数据包括多种形式数据的情况下，分别根据采集数据的不同形式来对不同的信息进行灵活融合，既能够提升融合体素信息集合的数据完整性，且能够提高融合效率，从而能够提升得到的目标场景的地图数据的全面性以及数据生成的效率。
在本公开的一些实施例中，在目标图像包括至少两个待处理图像，步骤S14可以包括：
对至少两个待处理图像对应的融合体素信息集合进行合并，得到目标场景的地图数据。
如上述各公开实施例所述，目标场景的地图数据可以是对目标场景中的各帧目标图像进行汇总所得到的数据，其包含的数据内容可以根据实际情况灵活决定。因此，在本公开的一些实施例中，可以将不同目标图像对应的融合体素信息集合进行合并，构成一个融合体素信息的数据单元，作为目标场景的地图数据。如此，通过将至少一个目标图像对应的融合体素信息集合进行合并得到地图数据，能够有效提升地图数据的数据完整性和全面性，且在包含多个连续的目标图像的情况下，地图数据可以对目标场景进行连续的表示。
在本公开的一些实施例中，步骤S14还可以包括：
将至少一个目标图像对应的融合体素信息集合存储至目标场景的地图数据中。
其中，在地图数据中存储哪些目标图像对应的融合体素信息集合，可以根据实际情况灵活决定，在本公开的一些实施例中，可以将每帧目标图像对应的融合体素信息集合均存储至目标场景的地图数据中。在本公开的一些实施例中，也可以对目标图像进行选定或筛选，来选择部分目标图像对应的融合体素信息集合来存储至地图数据等，如何实现可以根据实际情况灵活决定。
在本公开的一些实施例中，如上述各公开实施例所述，可以通过voxel hashing方法来建立目标图像的体素信息集合。在这种情况下，地图数据除了可以存储各融合体素信息集合以外，还可以存储用于查找各融合体素信息集合的哈希表等。
通过将至少一个目标图像对应的融合体素信息集合存储至目标场景的地图数据中，可以有效提升地图数据的数据完整性和全面性，且在包含多个连续的目标图像的情况下，地图数据可以对目标场景进行连续的表示。
在本公开的一些实施例中，本公开实施例提出的数据生成方法还可以包括：
对至少一帧目标图像进行回环检测，将检测到回环的目标图像作为回环图像；
确定回环图像的第二位姿，其中，第二位姿的准确度高于回环图像对应的目标图像的第一位姿的准确度；
根据回环图像的第二位姿，对目标场景的地图数据进行更新。
其中，回环检测可以是检测采集到的多帧目标图像中是否有针对相同场景的图像，回环检测的检测方式在本公开实施例中不做限制，任何视觉SLAM领域中用于进行回环检测的方式，均可以作为本公开实施例中回环检测的实现方式，不局限于下述各公开实施例。在本公开的一些实施例中，可以通过建立词袋模型的方式来实现回环检测。
在检测到回环以后，可以将检测到回环的目标图像作为回环图像，根据回环检测的结果，可以对回环图像的位姿进行优化，来确定回环图像的第二位姿。其中，对回环图像的位姿进行优化的方式在本公开实施例中不做限制，任何回环检测中基于检测结果，来对回环图像的位姿进行修正和优化的方式，均可以作为第二位姿的确定方式，本公开实施例中不对此过程进行限制。由于回环图像可以用于对目标场景中目标图像对应的第一位姿进行修正和优化，因此，得到的第二位姿相较于该回环图像对应的目标图像所确定的第一位姿来说，具有更高的准确度，从而使得基于第二位姿更新的地图数据具有更高的精度。
在确定回环图像的第二位姿以后，可以根据回环图像的第二位姿，对目标场景的地图数据进行更新。更新的方式在本公开实施例中可以灵活选择，详见下述各公开实施例，在此先不做展开。
通过对至少一帧目标图像进行回环检测，确定检测到的回环图像的第二位姿，并根据第二位姿对地图数据进行更新，如此，可以有效减小地图数据中的累计误差，提升地图数据的精度。
在本公开的一些实施例中，根据回环图像的第二位姿，对目标场景的地图数据进行更新，包括：
获取回环图像在第一位姿下对应的融合体素信息集合，作为第一目标集合；
基于回环图像对应的第一位姿和第二位姿，对第一目标集合中融合的信息进行重新融合，得到第二目标集合；
根据第二目标集合，对目标场景的地图数据进行更新。
其中，回环图像在第一位姿下对应的融合体素信息集合，是根据回环图像确定的第一位姿所建立的体素信息集合，与回环图像的信息进行融合所得到的融合体素信息集合，在本公开实施例中，可以将该集合作为第一目标集合。
如上述各公开实施例所述，在检测到回环图像的情况下，可以对回环图像的位姿进行优化，重新确定回环图像的第二位姿。在这种情况下，与回环图像在第一位姿下对应的第一目标集合，其包含的信息可能与真实的目标场景的数据具有偏差。因此，在本公开的一些实施例中，可以根据回环图像对应的第一位姿和第二位姿，对第一目标集合中的信息进行重新融合，得到第二目标集合。
如上述各公开实施例所述，融合体素信息集合可以融合多种类型的信息，比如语义信息、深度信息或颜色信息等。因此，第一目标集合中融合的信息包括深度信息的基础上，还可以包括语义信息以及颜色信息中的一种或多种。随着第一目标集合中融合的信息的种类的不同，重新融合的方式也可以灵活发生变化，详见下述各公开实施例，在此先不做展开。
在得到第二目标集合以后，可以根据第二目标集合对目标场景的地图数据进行更新，更新的方式可以根据实际情况灵活决定。在本公开的一些实施例中，可以用第二目标集合的数据替代第一目标集合的数据，来实现地图数据的更新。
通过获取回环图像在第一位姿下对应的融合体素信息集合，作为第一目标集合，并基于回环图像对应的第一位姿和第二位姿，对第一目标集合中融合的信息进行重新融合，来得到第二目标集合，从而根据第二目标集合，对目标场景的地图数据进行更新。如此，可以通过重新融合的方式，实现地图数据中位姿和各类信息的分别优化，与对地图数据 中的各类信息进行联合优化更新相比，能够有效减小地图数据更新过程中的计算量，以及提高地图数据的更新效率，实现地图数据的实时更新。
在本公开的一些实施例中，基于回环图像对应的第一位姿和第二位姿，对第一目标集合中融合的信息进行重新融合，得到第二目标集合，包括：
根据回环图像对应的第一位姿，对第一目标集合中融合的信息进行去融合，得到第三目标集合；
根据回环图像对应的第二位姿，将回环图像的信息融合至第三目标集合中，得到第二目标集合。
其中，第一目标集合中融合的信息的实现方式详见上述各公开实施例。
在本公开的一些实施例中，对第一目标集合中融合的信息进行重新融合的方式可以包括根据第一位姿所执行的去融合，以及依据第二位姿所执行的再次融合。
其中，去融合的方式可以根据第一目标集合中融合的信息的实现形式的不同而灵活发生变化。在本公开的一些实施例中，去融合可以看作融合的反向过程，因此去融合的可以通过融合的反向操作来实现。在本公开的一些实施例中，第一目标集合中融合的信息包括深度信息和颜色信息中的至少之一，在这种情况下，基于深度信息和颜色信息中的至少之一的融合过程的反向操作，对第一目标集合中融合的信息进行去融合的方式可以包括：
根据第一目标集合中融合的信息的融合权重，以及在第一位姿下，第一目标集合中至少一个体素与第一目标集合中融合的信息之间的投影关系，对第一目标集合中的至少一个体素进行信息去融合，得到第三目标集合。
其中，基于融合权重和在第一位姿下体素的投影关系来执行去融合的过程可以灵活决定。在本公开的一些实施例中，参考上述公式(2)和(3)的融合过程，第一目标集合中深度信息的去融合过程可以通过下述公式(6)和(7)进行表示：
Figure PCTCN2021105485-appb-000004
W”(v)＝W'(v)-w i(v)        (7)；
其中，D”(v)为体素v在去融合后的深度信息，D'(v)为上述公开实施例中提出的在第一位姿下体素v融合后的深度信息，W'(v)为上述公开实施例中提出的在第一位姿下体素v融合后更新的权重，w i(v)为上述公开实施例中提出的融合权重，d i(v)为上述公开实施例中提出的，在第一位姿下体素v在深度信息中对应的反投影点与体素v之间的距离，W”(v)为体素v去融合后更新的权重。
其中，深度信息的融合权重w i(v)，以及体素与深度信息之间的投影关系d i(v)的实现形式均可以参考上述各公开实施例。
通过上述公开实施例可以看出，基于深度信息的融合权重，以及在第一位姿下体素与深度信息之间的投影关系，对第一目标集合中已融合的深度信息进行去融合。在本公开的一些实施例中，在第一目标集合中融合的信息包括颜色信息的情况下，颜色信息的去融合方式可以参考上述深度信息的去融合方式，区别在于将深度信息对应的截断符号函数D(v)替换为颜色值C(v)，以及将在第一位姿下体素与深度信息之间的投影关系 替换为在第一位姿下体素与目标图像之间的投影关系等。在第一目标集合中融合的信息包含其他数据形式的情况下，其去融合方式可以参考上述各公开实施例进行灵活替换与扩展。
通过根据第一目标集合中融合的信息的融合权重，以及在第一位姿下体素信息集合中至少一个体素与第一目标集合中融合的信息之间的投影关系，对第一目标集合中至少一个体素进行信息去融合，得到第三目标集合，如此，可以在第一目标集合中融合的信息包括多种形式数据的情况下，分别根据第一目标集合中融合的信息的不同形式来对不同的信息进行灵活去融合，能够提高去融合的效率和灵活性，从而提升地图数据的更新效率。
在本公开的一些实施例中，第一目标集合中融合的信息可以包括语义信息，在这种情况下，基于语义信息的融合过程的反向操作，对第一目标集合中融合的信息进行去融合的方式可以包括：
在第一位姿下将第一目标集合中至少一个体素投影至回环图像，根据回环图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到第三目标集合。
根据上述各公开实施例可以看出，对语义信息进行融合的过程中，可以将体素信息集合中至少一个体素投影至目标图像，并根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息。因此，在对语义信息进行去融合的过程中，可以同样在第一位姿下将第一目标集合中的体素投影至回环图像，从而基于投影结果对融合的回环图像中至少一个像素的语义信息进行去融合，得到第三目标集合。
在本公开的一些实施例中，参考上述公式(1)可以看出，在对语义信息进行融合的过程中，可以基于融合当前目标图像的语义信息以前的体素的语义概率分布信息P(l i|I 1,...,k)，以及对当前目标图像进行语义分割所得到的语义信息概率分布P(O u(v,k)＝l i|I k)，来得到融合体素信息集合。因此，在本公开的一些实施例中，可以直接根据融合当前目标图像的语义信息以前的体素的语义概率分布信息P(l i|I 1,...,k)，得到第三目标集合。同时选择哪种方式来获取对语义信息进行去融合后的第三目标集合，其实现方式可以根据实际情况灵活决定，在本公开实施例中不做限制。
通过在第一位姿下将第一目标集合中至少一个体素投影至回环图像，根据回环图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到第三目标集合。同时可以利用语义信息融合过程的反向操作，实现语义信息的独立去融合，在不影响其他信息的融合与去融合的过程的基础上，便于后续重新融合语义信息来实现地图数据的更新，能够有效提升地图数据更新的可行性、灵活性以及效率。
在得到第三目标集合以后，可以根据回环图像对应的第二位姿，将回环图像的信息融合至第三目标集合中，得到第二目标集合。其中，回环图像的信息包含的信息内容可以在本公开实施例中不做限制，可以参考上述公开实施例中的第一目标集合中融合的信息的实现形式，比如可以为语义信息、深度信息或颜色信息等。
而根据回环图像对应的第二位姿，将回环图像的信息融合至第三目标集合的方式，同样可以参考上述各公开实施例中，对各类信息(比如：语义信息、深度信息或颜色信息等)进行融合的方式，唯一区别在于将上述融合过程中的第一位姿替换为第二位姿。在本公开的一些实施例中，在对语义信息进行融合的过程中，可以在第二位姿下，将第三目标集合中至少一个体素投影至回环图像，并根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到第二目标集合。融合深度信 息以及颜色信息的方式同样可以参考上述各公开实施例。
通过根据回环图像对应的第一位姿，对第一目标集合中融合的信息进行去融合，得到第三目标集合，再根据回环图像对应的第二位姿，将回环图像的信息融合至第三目标集合中，得到第二目标集合。如此，可以利用位姿与体素之间的对应关系，在位姿优化后快速地对融合的各类信息进行去融合与再融合，从而在更准确的位姿下对信息进行重新融合，高效地对地图数据进行更新，能够提高地图数据的准确性和更新效率。
在通过上述各公开实施例的任意组合形式得到地图数据以后，可以基于得到的地图数据，执行相应的应用，比如应用于对室内机器人的控制或是AR等场景中。在本公开的一些实施例中，由于本公开实施例中得到的地图数据中融合了连续的语义信息，因此可以用于控制机器人在目标场景中执行一些与语义信息相关的任务。在本公开的一些实施例中，可以控制机器人在目标场景中对目标物体执行相应操作，比如拿起桌子上的水杯等。在本公开的一些实施例中，本公开实施例得到的地图数据也可以应用于具有语义构图功能的AR平台等。
图3示出根据本公开一实施例的数据生成装置的框架示意图。如图3所示，所述数据生成装置30可以包括：
体素信息集合建立部分31，配置为确定目标场景中目标图像的第一位姿，根据第一位姿，建立目标图像的体素信息集合，其中，目标图像包括对目标场景进行数据采集所得到的至少一帧图像。
语义分割部分32，配置为对目标图像进行语义分割，得到目标图像的语义信息。
融合部分33，配置为将语义信息融合至体素信息集合中，得到融合体素信息集合。
数据生成部分34，配置为根据目标图像对应的融合体素信息集合，得到目标场景的地图数据。
在本公开的一些实施例中，融合部分33，还配置为：将体素信息集合中至少一个体素投影至目标图像，根据目标图像中至少一个像素的语义信息，确定投影后的至少一个体素的语义概率分布信息，得到融合体素信息集合。
在本公开的一些实施例中，融合部分33，还配置为：获取对目标场景进行数据采集所得到的采集数据，将采集数据融合至体素信息集合中，得到融合体素信息集合。
在本公开的一些实施例中，融合部分33，还配置为：根据采集数据的融合权重，以及体素信息集合中至少一个体素与采集数据之间的投影关系，对体素信息集合中至少一个体素进行信息融合，得到融合体素信息集合。
在本公开的一些实施例中，目标图像包括至少两个待处理图像，数据生成部分34，还配置为：对至少两个待处理图像对应的融合体素信息集合进行合并，得到目标场景的地图数据。
在本公开的一些实施例中，数据生成装置30，还包括：检测部分，配置为对至少一帧目标图像进行回环检测，将检测到回环的目标图像作为回环图像；确定部分，配置为确定回环图像的第二位姿，其中，第二位姿的准确度高于回环图像对应的目标图像的第一位姿的准确度；更新部分，配置为根据回环图像的第二位姿，对目标场景的地图数据进行更新。
在本公开的一些实施例中，更新部分，还配置为：获取回环图像在第一位姿下对应的融合体素信息集合，作为第一目标集合；基于回环图像对应的第一位姿和第二位姿，对第一目标集合中融合的信息进行重新融合，得到第二目标集合；根据第二目标集合，对目标场景的地图数据进行更新。
在本公开的一些实施例中，更新部分，还配置为：根据回环图像对应的第一位姿，对第一目标集合中融合的信息进行去融合，得到第三目标集合；根据回环图像对应的第二位姿，将回环图像的信息融合至第三目标集合中，得到第二目标集合。
基于上述实施例，本公开实施例提供一种应用场景示例，即本公开应用示例提出了一种数据生成方法，可以生成高质量的包含连续语义信息的地图数据。
图4示出根据本公开一应用示例得到地图数据的流程示意图，通过图4可以看出，本公开应用示例中，数据生成方法可以包括如下过程：
第一步，使用目标图像和IMU数据估计六自由度位姿，并且进行回环检测。
在本步骤中，可以利用传感器设备对目标场景进行数据采集得到的目标图像(RGB图像)和IMU数据，即402和403，并通过具有紧耦合、重定位、自标定、非线性优化以及全局位姿图优化等功能的单目VIO方法，来获取设备在采集各帧目标图像过程中所对应的六自由度位姿(即上述公开实施例中的第一位姿)，即404的位姿估计，并判断目标图像中是否有回环图像，即405。
第二步，基于卷积神经网络进行语义分割。
在本步骤中，可以将第一步中得到的目标图像，即402通过用于图像分割的卷积神经网络，进行语义分割，得到各目标图像的语义分割结果，即406。其中，用于图像分割的卷积神经网络可以通过训练图像进行训练来得到，其中，训练图像中包含各像素的语义标注数据，这些标注数据可以通过相关的标注方法或标注工具来进行获得。在本公开的一些实施例中，训练得到的用于图像分割的卷积神经网络对在室内环境下采集的图像具有较为准确的分割结果，并具有一定的泛化能力。
第三步，将单帧的目标图像、深度信息以及语义分割结果等融合到地图数据中。
如图4所示，在本公开应用示例中，除了在第一步中对目标场景进行数据采集得到目标图像和IMU数据以外，还得到目标场景的深度信息，即401。在本步骤中，可以基于第一步确定的第一位姿，即404以及当前的深度信息401，在包括哈希表和体素块数据单元的地图数据中，为当前的目标图像建立对应的体素信息集合。然后，根据语义信息、目标图像以及深度信息等，对当前目标图像对应的体素信息集合进行更新，来得到融合体素信息集合，即409。由于地图数据中绝大部分体素在当前目标图像对应的视野中都是不可见的，如此，能够加快各类信息的融合过程。
同时，可以将不同目标图像对应的融合体素信息集合进行合并，构成一个融合体素信息的数据单元，来作为目标场景的地图数据，或，将至少一个目标图像对应的融合体素信息集合存储至目标场景的地图数据中即得到地图数据410。
融合的方式在本公开应用示例中不做限制，在本公开的一些实施例中，可以通过滑动平均的方式融合深度信息和颜色信息。其中，融合过程可以参考上述各公开实施例以及公式(2)至(5)；在本公开的一些实施例中，可以基于贝叶斯定理融合语义信息，融合过程同样可以参考上述各公开实施例以及公式(1)。图5示出根据本公开一应用示例中目标场景的语义效果融合示意图，从图5中可以看出，目标场景在融合语义信息后，可以对不同区域的语义信息进行有效且连续地表达。
第四步，当检测到回环图像的情况下，对回环图像进行位姿优化后实时在线更新地图数据。
地图数据的生成过程中可能会出现位姿漂移甚至是错误估计，这样会导致地图数据存在扭曲，而回环检测可以消除位姿累计误差。在本步骤中，在检测到回环图像后，可以获取回环图像经过优化的第二位姿即407，并基于第二位姿通过重新融合的方式即408，实时在线更新地图。重新融合可以包括去融合和再融合两个过程，实现形式可以参考上述各公开实施例。
在本公开应用示例中，可以通过体素的方式表示地图数据，同时在体素对应的体素信息集合中融合语义标签的概率分布，从而可以生成带有语义标签的网格，并可以高质量地表示连续的场景。同时本公开应用示例在检测到回环并更新位姿以后，可以通过重新融合实时在线更新地图数据中融合的各类信息，如表征深度信息的有向距离场(Signed  Distance Field，SDF)值、颜色信息和语义标签的概率分布，从而及时消除地图中的扭曲。
可以理解，本公开提及的上述各个方法实施例，在不违背原理逻辑的情况下，均可以彼此相互结合形成结合后的实施例。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
本公开实施例还提出一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述方法实施例中所述数据生成方法。计算机可读存储介质可以是易失性计算机可读存储介质或非易失性计算机可读存储介质。
本公开实施例还提供一种计算机程序，所述计算机程序包括计算机可读代码，在所述计算机可读代码在电子设备中运行的情况下，所述电子设备的处理器执行如上述任一实施例所述数据生成方法。
本公开实施例还提供另一种计算机程序产品，包括存储了程序代码的计算机可读存储介质，所述程序代码包括的指令可配置为执行上述方法实施例中所述数据生成方法，可参见上述方法实施例。
本公开实施例还提出一种电子设备，包括：处理器；用于存储处理器可执行指令的存储器；其中，所述处理器配置为调用存储器存储的指令，以执行上述方法实施例中所述的数据生成方法。
在实际应用中，上述存储器可以是易失性存储器(volatile memory)，例如随机存取存储器(Random Access Memory，RAM)；或者非易失性存储器(non-volatile memory)，例如只读存储器(Read-Only Memory，ROM)，快闪存储器(flash memory)，硬盘(Hard Disk Drive，HDD)或固态硬盘(Solid-State Drive，SSD)；或者上述种类的存储器的组合，并向处理器提供指令和数据。
上述处理器可以为专用集成电路(Application Specific Integrated Circuit，ASIC)、数字信号处理器(Digital Signal Processor，DSP)、数字信号处理设备(Digital Signal Processor Device，DSPD)、可编程逻辑器件(Programmable Logic Device，PLD)、现场可编程门阵列(Field Programmable Gate Array，FPGA)、中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器中的至少一种。可以理解地，对于不同的设备，用于实现上述处理器功能的电子器件还可以为其它，本公开实施例不作限定。
电子设备可以被提供为终端、服务器或其它形态的设备。
基于前述实施例相同的技术构思，本公开实施例还提供了一种计算机程序，该计算机程序被处理器执行时实现上述方法。
图6是根据本公开实施例的一种电子设备800的架构示意框图。例如，电子设备800可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备以及PDU等终端。
参照图6所示，电子设备800可以包括以下一个或多个组件：处理组件802，存储器804，电源组件806，多媒体组件808，音频组件810，输入/输出(Input/Output，I/O)的接口812，传感器组件814，以及通信组件816。
处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。处理组件802可以包括一个或多个处理器820来执行指令，以完成上述的方法的全部或部分步骤。此外，处理组件802可以包括一个或多个模块，便于处理组件802和其他组件之间的交互。例如，处理组件802可以包括多媒体模块，以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电 话簿数据，消息，图片，视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(Static Random-Access Memory，SRAM)，电可擦除可编程只读存储器(Electrically Erasable Programmable Read Only Memory，EEPROM)，可擦除可编程只读存储器(Electrical Programmable Read Only Memory，EPROM)，可编程只读存储器(Programmable Read-Only Memory，PROM)，ROM，磁存储器，快闪存储器，磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力。电源组件806可以包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(Liquid Crystal Display，LCD)和触摸面板(Touch Panel，TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(Microphone，MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被存储在存储器804或经由通信组件816发送。在本公开的一些实施例中，音频组件810还包括一个扬声器，配置为输出音频信号。
I/O接口812为处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器，配置为为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到电子设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变，用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如互补金属氧化物半导体(Complementary Metal Oxide Semiconductor，CMOS)或电荷耦合装置(Charge-coupled Device，CCD)图像传感器，配置为在成像应用中使用。在本公开的一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如无线网络(Wi-Fi)，第二代移动通信技术(2-Generation，2G)或第三代移动通信技术(3rd-Generation，3G)，或它们的组合。在本公开的一些实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关人员信息。在本公开的一些实施例中，所述通信组件816还包括近场通信(Near Field Communication，NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(Radio Frequency Identification，RFID)技术，红外数据协会(Infrared Data Association，IrDA)技术，超宽带(Ultra Wide Band，UWB)技术，蓝牙(BitTorrent，BT)技术和其他技术来实现。
在示例性实施例中，电子设备800可以被一个或多个ASIC、DSP、DSPD、PLD、FPGA、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器804，上述计算机程序指令可由电子设备800的处理器820执行以完成上述方法。
图7是根据本公开实施例的一种电子设备1900的架构示意框图。例如，电子设备1900可以被提供为一服务器。参照图7，电子设备1900包括处理组件1922，其在本公开的一些实施例中包括一个或多个处理器，以及由存储器1932所代表的存储器资源，用于存储可由处理组件1922的执行的指令，例如应用程序。存储器1932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1922被配置为执行指令，以执行上述方法。
电子设备1900还可以包括一个电源组件1926被配置为执行电子设备1900的电源管理，一个有线或无线网络接口1950被配置为将电子设备1900连接到网络，和一个输入输出(I/O)接口1958。电子设备1900可以操作基于存储在存储器1932的操作系统，例如微软服务器操作系统(Windows ServerTM)，苹果公司推出的基于图形用户界面操作系统(Mac OS XTM)，多用户多进程的计算机操作系统(UnixTM)，自由和开放原代码的类Unix操作系统(LinuxTM)，开放原代码的类Unix操作系统(FreeBSDTM)或类似。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器1932，上述计算机程序指令可由电子设备1900的处理组件1922执行以完成上述方法。
本公开实施例可以是系统、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质，其上载有用于使处理器实现本公开实施例的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是但不限于电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：便携式计算机盘、硬盘、RAM、ROM、可EPROM或闪存、SRAM、便携式压缩盘只读存储器(Compact Disc Read-Only Memory，CD-ROM)、数字多功能盘(Digital Video Disc，DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备，或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从网络接收计算机可读程序指令，并转发该计算机可读程序指令，以供存储在各个计算/处理设备中的计算机可读存储介质中。
用于执行本公开实施例操作的计算机程序指令可以是汇编指令、指令集架构(Industry Standard Architecture，ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码，所述编程语言包括面向对象的编程语言，诸如Smalltalk、C++等，以及常规的过程式编程语言，诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机 上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络，包括局域网(Local Area Network，LAN)或广域网(Wide Area Network，WAN)，连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中，通过利用计算机可读程序指令的状态人员信息来个性化定制电子电路，例如可编程逻辑电路、FPGA或可编程逻辑阵列(Programmable Logic Arrays，PLA)，该电子电路可以执行计算机可读程序指令，从而实现本公开的各个方面。
这里参照根据本公开实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本公开的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上，使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤，以产生计算机实现的过程，从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本公开的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分，所述模块、程序段或指令的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
以上已经描述了本公开的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。
工业实用性
本公开实施例提供了一种数据生成方法、装置、设备、存储介质及程序。所述方法包括：确定目标场景中目标图像的第一位姿，根据所述第一位姿，建立所述目标图像的体素信息集合，其中，所述目标图像包括对所述目标场景进行数据采集所得到的至少一帧图像；对所述目标图像进行语义分割，得到所述目标图像的语义信息；将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合；根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据。

Claims (19)

  1. 一种数据生成方法，包括：
    确定目标场景中目标图像的第一位姿，根据所述第一位姿，建立所述目标图像的体素信息集合，其中，所述目标图像包括对所述目标场景进行数据采集所得到的至少一帧图像；
    对所述目标图像进行语义分割，得到所述目标图像的语义信息；
    将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合；
    根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据。
  2. 根据权利要求1所述的方法，其中，所述将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合，包括：
    将所述体素信息集合中至少一个体素投影至所述目标图像，根据所述目标图像中至少一个像素的语义信息，确定投影后的所述至少一个体素的语义概率分布信息，得到所述融合体素信息集合。
  3. 根据权利要求1或2所述的方法，其中，所述将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合，还包括：
    获取对所述目标场景进行数据采集所得到的采集数据，将所述采集数据融合至所述体素信息集合中，得到所述融合体素信息集合。
  4. 根据权利要求3所述的方法，其中，所述将所述采集数据融合至所述体素信息集合中，得到融合体素信息集合，包括：
    根据所述采集数据的融合权重，以及所述体素信息集合中至少一个体素与所述采集数据之间的投影关系，对所述体素信息集合中至少一个体素进行信息融合，得到所述融合体素信息集合。
  5. 根据权利要求1至4任一所述的方法，其中，所述目标图像包括至少两个待处理图像，所述根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据，包括：
    对所述至少两个待处理图像对应的所述融合体素信息集合进行合并，得到所述目标场景的地图数据。
  6. 根据权利要求1至5中任一项所述的方法，其中，所述方法还包括：
    对至少一帧所述目标图像进行回环检测，将检测到回环的目标图像作为回环图像；
    确定所述回环图像的第二位姿，其中，所述第二位姿的准确度高于所述回环图像对应的目标图像的第一位姿的准确度；
    根据所述回环图像的第二位姿，对所述目标场景的地图数据进行更新。
  7. 根据权利要求6所述的方法，其中，所述根据所述回环图像的第二位姿，对所述目标场景的地图数据进行更新，包括：
    获取所述回环图像在所述第一位姿下对应的融合体素信息集合，作为第一目标集合；
    基于所述回环图像对应的第一位姿和第二位姿，对所述第一目标集合中融合的信息进行重新融合，得到第二目标集合；
    根据所述第二目标集合，对所述目标场景的地图数据进行更新。
  8. 根据权利要求7所述的方法，其中，所述基于所述回环图像对应的第一位姿和第二位姿，对所述第一目标集合中融合的信息进行重新融合，得到第二目标集合，包括：
    根据所述回环图像对应的第一位姿，对所述第一目标集合中融合的信息进行去融合，得到第三目标集合；
    根据所述回环图像对应的第二位姿，将所述回环图像的信息融合至所述第三目标集合中，得到所述第二目标集合。
  9. 一种数据生成装置，所述装置包括：
    体素信息集合建立部分，配置为确定目标场景中目标图像的第一位姿，根据所述第一位姿，建立所述目标图像的体素信息集合，其中，所述目标图像包括对所述目标场景进行数据采集所得到的至少一帧图像；
    语义分割部分，配置为对所述目标图像进行语义分割，得到所述目标图像的语义信息；
    融合部分，配置为将所述语义信息融合至所述体素信息集合中，得到融合体素信息集合；
    数据生成部分，配置为根据所述目标图像对应的所述融合体素信息集合，得到所述目标场景的地图数据。
  10. 根据权利要求9所述的装置，其中，所述融合部分，还配置为：将所述体素信息集合中至少一个体素投影至所述目标图像，根据所述目标图像中至少一个像素的语义信息，确定投影后的所述至少一个体素的语义概率分布信息，得到所述融合体素信息集合。
  11. 根据权利要求9或10所述的装置，其中，所述融合部分，还配置为：获取对所述目标场景进行数据采集所得到的采集数据，将所述采集数据融合至所述体素信息集合中，得到所述融合体素信息集合。
  12. 根据权利要求11所述的装置，其中，所述融合部分，还配置为：根据所述采集数据的融合权重，以及所述体素信息集合中至少一个体素与所述采集数据之间的投影关系，对所述体素信息集合中至少一个体素进行信息融合，得到所述融合体素信息集合。
  13. 根据权利要求9至12中任一项所述的装置，其中，所述目标图像包括至少两个待处理图像，所述数据生成部分，还配置为：对所述至少两个待处理图像对应的所述融合体素信息集合进行合并，得到所述目标场景的地图数据。
  14. 根据权利要求9至13中任一项所述的装置，其中，所述装置还包括：检测部分，配置为对至少一帧所述目标图像进行回环检测，将检测到回环的目标图像作为回环图像；确定部分，配置为确定所述回环图像的第二位姿，其中，所述第二位姿的准确度高于所述回环图像对应的目标图像的第一位姿的准确度；更新部分，配置为根据所述回环图像的第二位姿，对所述目标场景的地图数据进行更新。
  15. 根据权利要求14所述的装置，其中，所述更新部分，还配置为：获取所述回环图像在所述第一位姿下对应的融合体素信息集合，作为第一目标集合；基于所述回环图像对应的第一位姿和第二位姿，对所述第一目标集合中融合的信息进行重新融合，得到第二目标集合；根据所述第二目标集合，对所述目标场景的地图数据进行更新。
  16. 根据权利要求15所述的装置，其中，所述更新部分，还配置为：根据所述回环图像对应的第一位姿，对所述第一目标集合中融合的信息进行去融合，得到第三目标集合；根据所述回环图像对应的第二位姿，将所述回环图像的信息融合至所述第三目标集合中，得到所述第二目标集合。
  17. 一种电子设备，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为调用所述存储器存储的指令，以执行权利要求1至8中任一项所述的方法。
  18. 一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现权利要求1至8中任一项所述的方法。
  19. 一种计算机程序，所述计算机程序包括计算机可读代码，在所述计算机可读代码在电子设备中运行的情况下，所述电子设备的处理器执行用于实现如权利要求1至8任一所述的方法。
PCT/CN2021/105485 2021-03-02 2021-07-09 数据生成方法、装置、设备、存储介质及程序 WO2022183656A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020227014409A KR20220125715A (ko) 2021-03-02 2021-07-09 데이터 생성 방법, 장치, 기기, 저장 매체 및 프로그램

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110231700.1A CN112837372A (zh) 2021-03-02 2021-03-02 数据生成方法及装置、电子设备和存储介质
CN202110231700.1 2021-03-02

Publications (1)

Publication Number Publication Date
WO2022183656A1 true WO2022183656A1 (zh) 2022-09-09

Family

ID=75934338

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/105485 WO2022183656A1 (zh) 2021-03-02 2021-07-09 数据生成方法、装置、设备、存储介质及程序

Country Status (3)

Country Link
KR (1) KR20220125715A (zh)
CN (1) CN112837372A (zh)
WO (1) WO2022183656A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115393386A (zh) * 2022-10-25 2022-11-25 杭州华橙软件技术有限公司 三维场景图的生成方法、装置、设备及可读存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112837372A (zh) * 2021-03-02 2021-05-25 浙江商汤科技开发有限公司 数据生成方法及装置、电子设备和存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104732587A (zh) * 2015-04-14 2015-06-24 中国科学技术大学 一种基于深度传感器的室内3d语义地图构建方法
US20190162856A1 (en) * 2017-11-29 2019-05-30 VoxelMaps Inc. Method and system for positioning of autonomously operating entities
CN110781262A (zh) * 2019-10-21 2020-02-11 中国科学院计算技术研究所 基于视觉slam的语义地图的构建方法
CN111462324A (zh) * 2020-05-18 2020-07-28 南京大学 一种在线时空语义融合方法和系统
CN111742310A (zh) * 2018-02-20 2020-10-02 祖克斯有限公司 创建包含语义信息的干净的地图
CN112348921A (zh) * 2020-11-05 2021-02-09 上海汽车集团股份有限公司 一种基于视觉语义点云的建图方法及系统
CN112837372A (zh) * 2021-03-02 2021-05-25 浙江商汤科技开发有限公司 数据生成方法及装置、电子设备和存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116659526A (zh) * 2016-03-15 2023-08-29 康多尔收购第二分公司 用于提供车辆认知的系统和方法
CN110044354B (zh) * 2019-03-28 2022-05-20 东南大学 一种双目视觉室内定位与建图方法及装置
CN110807782B (zh) * 2019-10-25 2021-08-20 中山大学 一种视觉机器人的地图表示系统及其构建方法
CN111462300A (zh) * 2020-03-05 2020-07-28 清华大学 衣物物理模型驱动的实时人体动态三维重建方法及系统
CN111047548B (zh) * 2020-03-12 2020-07-03 腾讯科技（深圳）有限公司 姿态变换数据处理方法、装置、计算机设备和存储介质
CN111476907A (zh) * 2020-04-14 2020-07-31 青岛小鸟看看科技有限公司 基于虚拟现实技术的定位及三维场景重建装置、方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104732587A (zh) * 2015-04-14 2015-06-24 中国科学技术大学 一种基于深度传感器的室内3d语义地图构建方法
US20190162856A1 (en) * 2017-11-29 2019-05-30 VoxelMaps Inc. Method and system for positioning of autonomously operating entities
CN111742310A (zh) * 2018-02-20 2020-10-02 祖克斯有限公司 创建包含语义信息的干净的地图
CN110781262A (zh) * 2019-10-21 2020-02-11 中国科学院计算技术研究所 基于视觉slam的语义地图的构建方法
CN111462324A (zh) * 2020-05-18 2020-07-28 南京大学 一种在线时空语义融合方法和系统
CN112348921A (zh) * 2020-11-05 2021-02-09 上海汽车集团股份有限公司 一种基于视觉语义点云的建图方法及系统
CN112837372A (zh) * 2021-03-02 2021-05-25 浙江商汤科技开发有限公司 数据生成方法及装置、电子设备和存储介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115393386A (zh) * 2022-10-25 2022-11-25 杭州华橙软件技术有限公司 三维场景图的生成方法、装置、设备及可读存储介质

Also Published As

Publication number Publication date
KR20220125715A (ko) 2022-09-14
CN112837372A (zh) 2021-05-25

Similar Documents

Publication Publication Date Title
CN110674719B (zh) 目标对象匹配方法及装置、电子设备和存储介质
CN111612009B (zh) 文本识别方法、装置、设备和存储介质
TWI706379B (zh) 圖像處理方法及裝置、電子設備和儲存介質
TW202105331A (zh) 一種人體關鍵點檢測方法及裝置、電子設備和電腦可讀儲存介質
US20210097715A1 (en) Image generation method and device, electronic device and storage medium
TWI767596B (zh) 場景深度和相機運動預測方法、電子設備和電腦可讀儲存介質
CN111538855B (zh) 视觉定位方法及装置、电子设备和存储介质
WO2022183656A1 (zh) 数据生成方法、装置、设备、存储介质及程序
CN111401230B (zh) 姿态估计方法及装置、电子设备和存储介质
CN111563138B (zh) 定位方法及装置、电子设备和存储介质
WO2022134475A1 (zh) 点云地图构建方法及装置、电子设备、存储介质和程序
TW202013316A (zh) 人臉圖像的處理方法及裝置、電子設備和儲存介質
WO2022017140A1 (zh) 目标检测方法及装置、电子设备和存储介质
CN112432637B (zh) 定位方法及装置、电子设备和存储介质
WO2022179013A1 (zh) 目标定位方法、装置、电子设备、存储介质及程序
WO2022151686A1 (zh) 场景图像展示方法、装置、设备、存储介质、程序及产品
JP2022546201A (ja) ターゲット検出方法および装置、電子機器並びに記憶媒体
WO2023273498A1 (zh) 深度检测方法及装置、电子设备和存储介质
WO2023273499A1 (zh) 深度检测方法及装置、电子设备和存储介质
CN111860388A (zh) 图像处理方法及装置、电子设备和存储介质
WO2022193456A1 (zh) 目标跟踪方法及装置、电子设备和存储介质
WO2022222379A1 (zh) 一种位置确定方法及装置、电子设备和存储介质
CN114581525A (zh) 姿态确定方法及装置、电子设备和存储介质
WO2022134390A1 (zh) 标注方法及装置、电子设备和存储介质
TWI770531B (zh) 人臉識別方法、電子設備和儲存介質

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2022525475

Country of ref document: JP

Kind code of ref document: A

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

Ref document number: 21928734

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: 21928734

Country of ref document: EP

Kind code of ref document: A1