技术领域Technical field
本发明涉及数据处理的领域，尤其涉及基于AI视觉识别的三维建模方法和系统。The present invention relates to the field of data processing, and in particular to a three-dimensional modeling method and system based on AI visual recognition.
背景技术Background technique
目前，三维建模通常是对目标场景进行倾斜拍摄或激光扫描方式得到目标场景的点云数据，再利用点云数据进行关于目标场景的三维建模。可见三维建模的数据来源主要是倾斜拍摄或激光扫描，使得三维建模的数据来源单一，若倾斜拍摄或激光扫描存在较大误差，会导致最终三维建模得到的三维模型的失真程度较高，使得三维模型与目标场景存在不匹配的情况，降低三维建模的准确性和效率。At present, 3D modeling usually involves oblique shooting or laser scanning of the target scene to obtain point cloud data of the target scene, and then using the point cloud data to perform 3D modeling of the target scene. It can be seen that the data source of 3D modeling is mainly oblique shooting or laser scanning, which makes the data source of 3D modeling single. If there are large errors in oblique shooting or laser scanning, it will lead to a high degree of distortion of the 3D model obtained by the final 3D modeling. , causing a mismatch between the 3D model and the target scene, reducing the accuracy and efficiency of 3D modeling.
发明内容Contents of the invention
本发明的目的在于提供基于AI视觉识别的三维建模方法和系统，其对目标场景进行拍摄，得到目标场景影像，并对每个目标场景子影像进行AI识别处理，得到目标场景的视觉标定点云数据；在拍摄目标场景时同步进行激光扫描，得到目标场景的激光扫描数据，并对其进行AI识别处理，得到目标场景的激光标定点云数据；基于视觉标定点云数据，构建目标场景的初始三维场景模型；再利用激光标定点云数据，对初始三维场景模型相应的三维场景部分进行校正，通过拍摄和激光扫描形成不同三维建模数据源，避免采用单一数据源进行三维建模过程中容易产生建模失真的情况，能够对三维模型进行针对性校正，提高三维建模的准确性和效率。The purpose of the present invention is to provide a three-dimensional modeling method and system based on AI visual recognition, which shoots the target scene to obtain the target scene image, and performs AI recognition processing on each target scene sub-image to obtain the visual calibration point of the target scene. Cloud data; perform laser scanning simultaneously when shooting the target scene to obtain the laser scanning data of the target scene, and perform AI recognition processing on it to obtain the laser calibration point cloud data of the target scene; build a target scene based on the visual calibration point cloud data Initial 3D scene model; then use laser calibration point cloud data to correct the corresponding 3D scene part of the initial 3D scene model. Different 3D modeling data sources are formed through shooting and laser scanning to avoid using a single data source in the 3D modeling process. In situations where modeling distortion is prone to occur, the 3D model can be corrected in a targeted manner to improve the accuracy and efficiency of 3D modeling.
本发明是通过以下技术方案实现：The present invention is realized through the following technical solutions:
基于AI视觉识别的三维建模方法，包括：Three-dimensional modeling methods based on AI visual recognition include:
对目标场景进行拍摄，得到目标场景影像，并将所述目标场景影像进行分割处理，得到若干目标场景子影像；对每个目标场景子影像进行AI识别处理，得到所述目标场景的视觉标定点云数据；Shoot the target scene to obtain the target scene image, segment the target scene image to obtain several target scene sub-images, and perform AI recognition processing on each target scene sub-image to obtain the visual calibration point of the target scene. cloud data;
在对所述目标场景进行拍摄过程中同步对所述目标场景进行激光扫描，得到所述目标场景的激光扫描数据；对所述激光扫描数据进行AI识别处理，得到所述目标场景的激光标定点云数据；基于对所述目标场景进行拍摄和激光扫描同步执行属性信息，构建所述视觉标定点云数据和所述激光标定点云数据的关联映射信息；During the shooting process of the target scene, the target scene is synchronously laser scanned to obtain the laser scanning data of the target scene; the laser scanning data is subjected to AI recognition processing to obtain the laser calibration point of the target scene. Cloud data; based on the synchronous execution attribute information of shooting and laser scanning of the target scene, constructing the associated mapping information of the visual calibration point cloud data and the laser calibration point cloud data;
基于所述视觉标定点云数据，构建所述目标场景对应的初始三维场景模型；对所述初始三维场景模型进行场景状态识别，从所述初始三维场景模型中提取需要进行校正的三维场景部分；基于所述关联映射信息，查找与所述三维场景部分对应的激光标定点云数据部分；基于所述激光标定点云数据部分，对所述三维场景部分进行校正。Based on the visual calibration point cloud data, construct an initial three-dimensional scene model corresponding to the target scene; perform scene state identification on the initial three-dimensional scene model, and extract the three-dimensional scene part that needs to be corrected from the initial three-dimensional scene model; Based on the associated mapping information, the laser calibration point cloud data part corresponding to the three-dimensional scene part is searched; based on the laser calibration point cloud data part, the three-dimensional scene part is corrected.
可选地，对目标场景进行拍摄，得到目标场景影像，并将所述目标场景影像进行分割处理，得到若干目标场景子影像；对每个目标场景子影像进行AI识别处理，得到所述目标场景的视觉标定点云数据，包括：Optionally, shoot the target scene to obtain the target scene image, segment the target scene image to obtain several target scene sub-images, and perform AI recognition processing on each target scene sub-image to obtain the target scene. Visual calibration point cloud data, including:
对目标场景进行扫描拍摄，得到目标场景全景影像；对所述目标场景全景影像依次进行像素锐化处理和像素轮廓识别处理，得到所述目标场景全景影像的像素轮廓特征信息；Scan and shoot the target scene to obtain a panoramic image of the target scene; perform pixel sharpening processing and pixel contour recognition processing on the panoramic image of the target scene in sequence to obtain pixel contour feature information of the panoramic image of the target scene;
基于所述像素轮廓特征信息，对所述目标场景全景影像进行分割处理，得到关于所述目标场景的背景区域和非背景区域的若干目标场景子影像；Based on the pixel contour feature information, segment the target scene panoramic image to obtain several target scene sub-images regarding the background area and non-background area of the target scene;
对每个目标场景子影像进行AI识别处理，得到每个目标场景子影像的所有关键像素点的点云数据；其中，所述关键像素点是指所述目标场景子影像的画面上背景物体和非背景物体各自的边界轮廓点；再基于所有目标场景子影像在所述目标场景影像的位置信息，将所有目标场景子影像各自对应的点云数据组成所述目标场景的视觉标定点云数据。Perform AI recognition processing on each target scene sub-image to obtain point cloud data of all key pixels of each target scene sub-image; where the key pixels refer to the background objects and objects on the screen of the target scene sub-image. The respective boundary contour points of non-background objects; and then based on the position information of all target scene sub-images in the target scene image, the corresponding point cloud data of all target scene sub-images are composed into the visual calibration point cloud data of the target scene.
可选地，在对所述目标场景进行拍摄过程中同步对所述目标场景进行激光扫描，得到所述目标场景的激光扫描数据；对所述激光扫描数据进行AI识别处理，得到所述目标场景的激光标定点云数据；基于对所述目标场景进行拍摄和激光扫描同步执行属性信息，构建所述视觉标定点云数据和所述激光标定点云数据的关联映射信息，包括：Optionally, during the process of photographing the target scene, the target scene is synchronously laser scanned to obtain the laser scanning data of the target scene; the laser scanning data is subjected to AI recognition processing to obtain the target scene. The laser calibration point cloud data; based on the attribute information of shooting and laser scanning of the target scene, constructing the associated mapping information of the visual calibration point cloud data and the laser calibration point cloud data, including:
在对所述目标场景进行拍摄过程中同步对所述目标场景进行全景激光扫描，得到所述目标场景的全景激光扫描数据，并对所述全景激光扫描数据进行卡尔曼滤波处理；During the process of photographing the target scene, perform panoramic laser scanning on the target scene simultaneously to obtain panoramic laser scanning data of the target scene, and perform Kalman filtering processing on the panoramic laser scanning data;
对所述全景激光扫描数据进行AI识别处理，得到关于所述目标场景的背景区域和非背景区域各自对应的激光标定点云子数据，并将所有激光标定点云子数据组成激光标定点云数据；Perform AI recognition processing on the panoramic laser scanning data to obtain laser calibration point cloud sub-data corresponding to the background area and non-background area of the target scene, and combine all laser calibration point cloud sub-data into laser calibration point cloud data ;
基于所述目标场景进行拍摄和激光扫描同步执行过程中的执行方位信息，构建所述视觉标定点云数据和所述激光标定点云数据的一一对应映射信息。Based on the execution orientation information during the simultaneous execution of shooting and laser scanning of the target scene, one-to-one mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed.
可选地，基于所述视觉标定点云数据，构建所述目标场景对应的初始三维场景模型；对所述初始三维场景模型进行场景状态识别，从所述初始三维场景模型中提取需要进行校正的三维场景部分；基于所述关联映射信息，查找与所述三维场景部分对应的激光标定点云数据部分；基于所述激光标定点云数据部分，对所述三维场景部分进行校正，包括：Optionally, based on the visual calibration point cloud data, an initial three-dimensional scene model corresponding to the target scene is constructed; scene state recognition is performed on the initial three-dimensional scene model, and the elements that need to be corrected are extracted from the initial three-dimensional scene model. The three-dimensional scene part; based on the associated mapping information, search for the laser calibration point cloud data part corresponding to the three-dimensional scene part; based on the laser calibration point cloud data part, correct the three-dimensional scene part, including:
对所述视觉标定点云数据进行AI建模处理，构建所述目标场景对应的初始三维场景模型；Perform AI modeling processing on the visual calibration point cloud data to construct an initial three-dimensional scene model corresponding to the target scene;
对所述初始三维场景模型进行场景失真度识别，得到所述初始三维场景模型的背景部分和非背景部分各自对应的场景失真度；若所述场景失真度大于或等于预设失真度阈值，则将对应的背景部分/非背景部分确定为所述初始三维场景模型中需要进行校正的三维场景部分；Perform scene distortion recognition on the initial three-dimensional scene model to obtain scene distortion degrees corresponding to the background part and the non-background part of the initial three-dimensional scene model; if the scene distortion is greater than or equal to the preset distortion threshold, then Determine the corresponding background part/non-background part as the three-dimensional scene part that needs to be corrected in the initial three-dimensional scene model;
基于所述关联映射信息，查找与所述三维场景部分对应的激光标定点云数据部分；基于所述激光标定点云数据部分对所述三维场景部分进行三维场景轮廓校正。Based on the associated mapping information, search for the laser calibration point cloud data part corresponding to the three-dimensional scene part; perform three-dimensional scene contour correction on the three-dimensional scene part based on the laser calibration point cloud data part.
基于AI视觉识别的三维建模系统，包括：3D modeling system based on AI visual recognition, including:
场景影像拍摄与处理模块，用于对目标场景进行拍摄，得到目标场景影像，并将所述目标场景影像进行分割处理，得到若干目标场景子影像；The scene image shooting and processing module is used to shoot the target scene to obtain the target scene image, and segment the target scene image to obtain several target scene sub-images;
影像AI识别模块，用于对每个目标场景子影像进行AI识别处理，得到所述目标场景的视觉标定点云数据；The image AI recognition module is used to perform AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene;
场景激光扫描模块，用于在对所述目标场景进行拍摄过程中同步对所述目标场景进行激光扫描，得到所述目标场景的激光扫描数据；A scene laser scanning module, configured to synchronously laser scan the target scene during the shooting process of the target scene to obtain laser scanning data of the target scene;
激光扫描数据AI识别模块，用于对所述激光扫描数据进行AI识别处理，得到所述目标场景的激光标定点云数据；A laser scanning data AI recognition module is used to perform AI recognition processing on the laser scanning data to obtain laser calibration point cloud data of the target scene;
关联映射确定模块，用于基于对所述目标场景进行拍摄和激光扫描同步执行属性信息，构建所述视觉标定点云数据和所述激光标定点云数据的关联映射信息；An association mapping determination module, configured to construct association mapping information of the visual calibration point cloud data and the laser calibration point cloud data based on the synchronous execution attribute information of shooting and laser scanning of the target scene;
三维场景模型构建与识别模块，用于基于所述视觉标定点云数据，构建所述目标场景对应的初始三维场景模型；对所述初始三维场景模型进行场景状态识别，从所述初始三维场景模型中提取需要进行校正的三维场景部分；A three-dimensional scene model construction and identification module is used to construct an initial three-dimensional scene model corresponding to the target scene based on the visual calibration point cloud data; perform scene state recognition on the initial three-dimensional scene model, and perform scene state identification on the initial three-dimensional scene model. Extract the part of the three-dimensional scene that needs to be corrected;
三维场景模型校正模块，用于基于所述关联映射信息，查找与所述三维场景部分对应的激光标定点云数据部分；基于所述激光标定点云数据部分，对所述三维场景部分进行校正。A three-dimensional scene model correction module is configured to search for the laser calibration point cloud data part corresponding to the three-dimensional scene part based on the associated mapping information; and correct the three-dimensional scene part based on the laser calibration point cloud data part.
可选地，所述场景影像拍摄与处理模块用于对目标场景进行拍摄，得到目标场景影像，并将所述目标场景影像进行分割处理，得到若干目标场景子影像，包括：Optionally, the scene image shooting and processing module is used to shoot the target scene to obtain the target scene image, and segment the target scene image to obtain several target scene sub-images, including:
对目标场景进行扫描拍摄，得到目标场景全景影像；对所述目标场景全景影像依次进行像素锐化处理和像素轮廓识别处理，得到所述目标场景全景影像的像素轮廓特征信息；Scan and shoot the target scene to obtain a panoramic image of the target scene; perform pixel sharpening processing and pixel contour recognition processing on the panoramic image of the target scene in sequence to obtain pixel contour feature information of the panoramic image of the target scene;
基于所述像素轮廓特征信息，对所述目标场景全景影像进行分割处理，得到关于所述目标场景的背景区域和非背景区域的若干目标场景子影像；Based on the pixel contour feature information, segment the target scene panoramic image to obtain several target scene sub-images regarding the background area and non-background area of the target scene;
所述影像AI识别模块用于对每个目标场景子影像进行AI识别处理，得到所述目标场景的视觉标定点云数据，包括：The image AI recognition module is used to perform AI recognition processing on each target scene sub-image to obtain visual calibration point cloud data of the target scene, including:
对每个目标场景子影像进行AI识别处理，得到每个目标场景子影像的所有关键像素点的点云数据；其中，所述关键像素点是指所述目标场景子影像的画面上背景物体和非背景物体各自的边界轮廓点；再基于所有目标场景子影像在所述目标场景影像的位置信息，将所有目标场景子影像各自对应的点云数据组成所述目标场景的视觉标定点云数据。Perform AI recognition processing on each target scene sub-image to obtain point cloud data of all key pixels of each target scene sub-image; where the key pixels refer to the background objects and objects on the screen of the target scene sub-image. the respective boundary contour points of non-background objects; and then based on the position information of all target scene sub-images in the target scene image, the corresponding point cloud data of all target scene sub-images is composed into the visual calibration point cloud data of the target scene.
可选地，所述场景激光扫描模块用于在对所述目标场景进行拍摄过程中同步对所述目标场景进行激光扫描，得到所述目标场景的激光扫描数据，包括：Optionally, the scene laser scanning module is used to simultaneously perform laser scanning on the target scene during the shooting process of the target scene to obtain laser scanning data of the target scene, including:
在对所述目标场景进行拍摄过程中同步对所述目标场景进行全景激光扫描，得到所述目标场景的全景激光扫描数据，并对所述全景激光扫描数据进行卡尔曼滤波处理；During the process of photographing the target scene, perform panoramic laser scanning on the target scene simultaneously to obtain panoramic laser scanning data of the target scene, and perform Kalman filtering processing on the panoramic laser scanning data;
所述激光扫描数据AI识别模块用于对所述激光扫描数据进行AI识别处理，得到所述目标场景的激光标定点云数据，包括：The laser scanning data AI recognition module is used to perform AI recognition processing on the laser scanning data to obtain laser calibration point cloud data of the target scene, including:
对所述全景激光扫描数据进行AI识别处理，得到关于所述目标场景的背景区域和非背景区域各自对应的激光标定点云子数据，并将所有激光标定点云子数据组成激光标定点云数据；Perform AI recognition processing on the panoramic laser scanning data to obtain laser calibration point cloud sub-data corresponding to the background area and non-background area of the target scene, and combine all laser calibration point cloud sub-data into laser calibration point cloud data ;
所述关联映射确定模块用于基于对所述目标场景进行拍摄和激光扫描同步执行属性信息，构建所述视觉标定点云数据和所述激光标定点云数据的关联映射信息，包括：The association mapping determination module is used to construct association mapping information of the visual calibration point cloud data and the laser calibration point cloud data based on the attribute information of the synchronous execution of shooting and laser scanning of the target scene, including:
基于所述目标场景进行拍摄和激光扫描同步执行过程中的执行方位信息，构建所述视觉标定点云数据和所述激光标定点云数据的一一对应映射信息。Based on the execution orientation information during the simultaneous execution of shooting and laser scanning of the target scene, one-to-one mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed.
可选地，所述三维场景模型构建与识别模块用于基于所述视觉标定点云数据，构建所述目标场景对应的初始三维场景模型；对所述初始三维场景模型进行场景状态识别，从所述初始三维场景模型中提取需要进行校正的三维场景部分，包括：Optionally, the three-dimensional scene model construction and recognition module is used to construct an initial three-dimensional scene model corresponding to the target scene based on the visual calibration point cloud data; perform scene state recognition on the initial three-dimensional scene model, and perform scene state recognition from the initial three-dimensional scene model. The parts of the 3D scene that need to be corrected are extracted from the initial 3D scene model, including:
对所述视觉标定点云数据进行AI建模处理，构建所述目标场景对应的初始三维场景模型；Perform AI modeling processing on the visual calibration point cloud data to construct an initial three-dimensional scene model corresponding to the target scene;
对所述初始三维场景模型进行场景失真度识别，得到所述初始三维场景模型的背景部分和非背景部分各自对应的场景失真度；若所述场景失真度大于或等于预设失真度阈值，则将对应的背景部分/非背景部分确定为所述初始三维场景模型中需要进行校正的三维场景部分；Perform scene distortion recognition on the initial three-dimensional scene model to obtain scene distortion degrees corresponding to the background part and the non-background part of the initial three-dimensional scene model; if the scene distortion is greater than or equal to the preset distortion threshold, then Determine the corresponding background part/non-background part as the three-dimensional scene part that needs to be corrected in the initial three-dimensional scene model;
所述三维场景模型校正模块用于基于所述关联映射信息，查找与所述三维场景部分对应的激光标定点云数据部分；基于所述激光标定点云数据部分，对所述三维场景部分进行校正，包括：The three-dimensional scene model correction module is used to find the laser calibration point cloud data part corresponding to the three-dimensional scene part based on the associated mapping information; and correct the three-dimensional scene part based on the laser calibration point cloud data part. ,include:
基于所述关联映射信息，查找与所述三维场景部分对应的激光标定点云数据部分；基于所述激光标定点云数据部分对所述三维场景部分进行三维场景轮廓校正。Based on the associated mapping information, search for the laser calibration point cloud data part corresponding to the three-dimensional scene part; perform three-dimensional scene contour correction on the three-dimensional scene part based on the laser calibration point cloud data part.
与现有技术相比，本发明具有如下有益效果：Compared with the prior art, the present invention has the following beneficial effects:
本申请提供的基于AI视觉识别的三维建模方法和系统对目标场景进行拍摄，得到目标场景影像，并对每个目标场景子影像进行AI识别处理，得到目标场景的视觉标定点云数据；在拍摄目标场景时同步进行激光扫描，得到目标场景的激光扫描数据，并对其进行AI识别处理，得到目标场景的激光标定点云数据；基于视觉标定点云数据，构建目标场景的初始三维场景模型；再利用激光标定点云数据，对初始三维场景模型相应的三维场景部分进行校正，通过拍摄和激光扫描形成不同三维建模数据源，避免采用单一数据源进行三维建模过程中容易产生建模失真的情况，能够对三维模型进行针对性校正，提高三维建模的准确性和效率。The three-dimensional modeling method and system based on AI visual recognition provided by this application shoots the target scene to obtain the target scene image, and performs AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene; in When shooting the target scene, laser scanning is performed simultaneously to obtain the laser scanning data of the target scene, and AI recognition processing is performed on it to obtain the laser calibration point cloud data of the target scene; based on the visual calibration point cloud data, an initial three-dimensional scene model of the target scene is constructed. ; Then use laser calibration point cloud data to correct the corresponding 3D scene part of the initial 3D scene model. Different 3D modeling data sources are formed through shooting and laser scanning to avoid modeling that is easy to occur when using a single data source for 3D modeling. In case of distortion, the 3D model can be corrected in a targeted manner to improve the accuracy and efficiency of 3D modeling.
附图说明Description of drawings
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。其中：In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without exerting creative efforts. in:
图1为本发明提供的基于AI视觉识别的三维建模方法的流程示意图。Figure 1 is a schematic flow chart of the three-dimensional modeling method based on AI visual recognition provided by the present invention.
图2为本发明提供的基于AI视觉识别的三维建模系统的结构示意图。Figure 2 is a schematic structural diagram of the three-dimensional modeling system based on AI visual recognition provided by the present invention.
具体实施方式Detailed ways
为使本申请的上述目的、特征和优点能够更为明显易懂，下面结合附图，对本申请的具体实施方式做详细的说明。可以理解的是，此处所描述的具体实施例仅用于解释本申请，而非对本申请的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本申请相关的部分而非全部结构。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本申请保护的范围。In order to make the above objects, features and advantages of the present application more obvious and easy to understand, the specific implementation modes of the present application will be described in detail below with reference to the accompanying drawings. It can be understood that the specific embodiments described here are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for convenience of description, only some but not all structures related to the present application are shown in the drawings. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.
本申请中的术语“包括”和“具有”以及它们任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备没有限定于已列出的步骤或单元，而是可选地还包括没有列出的步骤或单元，或可选地还包括对于这些过程、方法、产品或设备固有的其它步骤或单元。The terms "including" and "having" and any variations thereof in this application are intended to cover a non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes Other steps or units inherent to such processes, methods, products or devices.
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。Reference herein to "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearances of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those skilled in the art understand, both explicitly and implicitly, that the embodiments described herein may be combined with other embodiments.
请参阅图1所示，本申请一实施例提供的基于AI视觉识别的三维建模方法包括：Please refer to Figure 1. A three-dimensional modeling method based on AI visual recognition provided by an embodiment of the present application includes:
对目标场景进行拍摄，得到目标场景影像，并将该目标场景影像进行分割处理，得到若干目标场景子影像；对每个目标场景子影像进行AI识别处理，得到该目标场景的视觉标定点云数据；Shoot the target scene to obtain the target scene image, and segment the target scene image to obtain several target scene sub-images; perform AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene ;
在对该目标场景进行拍摄过程中同步对该目标场景进行激光扫描，得到该目标场景的激光扫描数据；对该激光扫描数据进行AI识别处理，得到该目标场景的激光标定点云数据；基于对该目标场景进行拍摄和激光扫描同步执行属性信息，构建该视觉标定点云数据和该激光标定点云数据的关联映射信息；During the shooting process of the target scene, laser scanning is performed simultaneously on the target scene to obtain laser scanning data of the target scene; AI recognition processing is performed on the laser scanning data to obtain laser calibration point cloud data of the target scene; based on The target scene is photographed and laser scanned to execute attribute information simultaneously, and the associated mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed;
基于该视觉标定点云数据，构建该目标场景对应的初始三维场景模型；对该初始三维场景模型进行场景状态识别，从该初始三维场景模型中提取需要进行校正的三维场景部分；基于该关联映射信息，查找与该三维场景部分对应的激光标定点云数据部分；基于该激光标定点云数据部分，对该三维场景部分进行校正。Based on the visual calibration point cloud data, an initial three-dimensional scene model corresponding to the target scene is constructed; the scene state is identified on the initial three-dimensional scene model, and the three-dimensional scene part that needs to be corrected is extracted from the initial three-dimensional scene model; based on the association mapping Information is used to find the laser calibration point cloud data part corresponding to the three-dimensional scene part; based on the laser calibration point cloud data part, the three-dimensional scene part is corrected.
上述实施例的有益效果，该基于AI视觉识别的三维建模方法对目标场景进行拍摄，得到目标场景影像，并对每个目标场景子影像进行AI识别处理，得到目标场景的视觉标定点云数据；在拍摄目标场景时同步进行激光扫描，得到目标场景的激光扫描数据，并对其进行AI识别处理，得到目标场景的激光标定点云数据；基于视觉标定点云数据，构建目标场景的初始三维场景模型；再利用激光标定点云数据，对初始三维场景模型相应的三维场景部分进行校正，通过拍摄和激光扫描形成不同三维建模数据源，避免采用单一数据源进行三维建模过程中容易产生建模失真的情况，能够对三维模型进行针对性校正，提高三维建模的准确性和效率。The beneficial effect of the above embodiment is that the three-dimensional modeling method based on AI visual recognition shoots the target scene to obtain the target scene image, and performs AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene. ; Perform laser scanning simultaneously when shooting the target scene to obtain the laser scanning data of the target scene, and perform AI recognition processing on it to obtain the laser calibration point cloud data of the target scene; based on the visual calibration point cloud data, construct an initial three-dimensional view of the target scene Scene model; then use laser calibration point cloud data to correct the corresponding 3D scene part of the initial 3D scene model, and form different 3D modeling data sources through shooting and laser scanning to avoid the problems that are easy to occur in the 3D modeling process using a single data source. In case of modeling distortion, the 3D model can be corrected in a targeted manner to improve the accuracy and efficiency of 3D modeling.
在另一实施例中，对目标场景进行拍摄，得到目标场景影像，并将该目标场景影像进行分割处理，得到若干目标场景子影像；对每个目标场景子影像进行AI识别处理，得到该目标场景的视觉标定点云数据，包括：In another embodiment, the target scene is photographed to obtain a target scene image, and the target scene image is segmented to obtain several target scene sub-images; AI recognition processing is performed on each target scene sub-image to obtain the target scene. Visual calibration point cloud data of the scene, including:
对目标场景进行扫描拍摄，得到目标场景全景影像；对该目标场景全景影像依次进行像素锐化处理和像素轮廓识别处理，得到该目标场景全景影像的像素轮廓特征信息；Scan and shoot the target scene to obtain a panoramic image of the target scene; perform pixel sharpening processing and pixel contour recognition processing on the panoramic image of the target scene in sequence to obtain the pixel contour feature information of the panoramic image of the target scene;
基于该像素轮廓特征信息，对该目标场景全景影像进行分割处理，得到关于该目标场景的背景区域和非背景区域的若干目标场景子影像；Based on the pixel contour feature information, perform segmentation processing on the target scene panoramic image to obtain several target scene sub-images regarding the background area and non-background area of the target scene;
对每个目标场景子影像进行AI识别处理，得到每个目标场景子影像的所有关键像素点的点云数据；其中，该关键像素点是指该目标场景子影像的画面上背景物体和非背景物体各自的边界轮廓点；再基于所有目标场景子影像在该目标场景影像的位置信息，将所有目标场景子影像各自对应的点云数据组成该目标场景的视觉标定点云数据。Perform AI recognition processing on each target scene sub-image to obtain point cloud data of all key pixels of each target scene sub-image; where the key pixels refer to background objects and non-background objects on the screen of the target scene sub-image. The respective boundary contour points of the object; and then based on the position information of all target scene sub-images in the target scene image, the corresponding point cloud data of all target scene sub-images is composed of the visual calibration point cloud data of the target scene.
上述实施例的有益效果，在对目标场景的实际扫描拍摄中，可利用360度摄像机对目标场景进行全景扫描拍摄，得到目标场景全景影像，再对该目标场景全景影像依次进行像素锐化处理和像素轮廓识别处理，得到相应的像素轮廓特征信息，以此对目标场景全景影像进行背景区域和非背景区域的标定区分，这样以该像素轮廓特征信息为基准，对该目标场景全景影像进行分割处理，得到关于该目标场景的背景区域和非背景区域的若干目标场景子影像，便于后续针对背景区域和非背景区域进行区分化的点云数据的提取。再利用相应的神经网络模型对每个目标场景子影像进行AI识别处理，得到每个目标场景子影像的所有关键像素点的点云数据，从而实现对目标场景全景影像的全面准确点云数据提取。The beneficial effect of the above embodiment is that in the actual scanning and shooting of the target scene, a 360-degree camera can be used to perform panoramic scanning and shooting of the target scene to obtain a panoramic image of the target scene, and then perform pixel sharpening and processing on the panoramic image of the target scene in sequence. Pixel contour recognition processing is used to obtain corresponding pixel contour feature information, so as to calibrate and distinguish the background area and non-background area of the target scene panoramic image. In this way, the target scene panoramic image is segmented based on the pixel outline feature information. , obtain several target scene sub-images about the background area and non-background area of the target scene, which facilitates the subsequent extraction of point cloud data that differentiates between background areas and non-background areas. Then use the corresponding neural network model to perform AI recognition processing on each target scene sub-image, and obtain the point cloud data of all key pixels of each target scene sub-image, thereby achieving comprehensive and accurate point cloud data extraction for the target scene panoramic image. .
在另一实施例中，在对该目标场景进行拍摄过程中同步对该目标场景进行激光扫描，得到该目标场景的激光扫描数据；对该激光扫描数据进行AI识别处理，得到该目标场景的激光标定点云数据；基于对该目标场景进行拍摄和激光扫描同步执行属性信息，构建该视觉标定点云数据和该激光标定点云数据的关联映射信息，包括：In another embodiment, during the process of shooting the target scene, laser scanning is performed simultaneously on the target scene to obtain the laser scanning data of the target scene; AI recognition processing is performed on the laser scanning data to obtain the laser scanning data of the target scene. Calibrate point cloud data; based on the synchronous execution attribute information of shooting and laser scanning of the target scene, construct the associated mapping information of the visual calibration point cloud data and the laser calibration point cloud data, including:
在对该目标场景进行拍摄过程中同步对该目标场景进行全景激光扫描，得到该目标场景的全景激光扫描数据，并对该全景激光扫描数据进行卡尔曼滤波处理；During the shooting process of the target scene, panoramic laser scanning is performed simultaneously on the target scene to obtain panoramic laser scanning data of the target scene, and Kalman filtering is performed on the panoramic laser scanning data;
对该全景激光扫描数据进行AI识别处理，得到关于该目标场景的背景区域和非背景区域各自对应的激光标定点云子数据，并将所有激光标定点云子数据组成激光标定点云数据；Perform AI recognition processing on the panoramic laser scanning data to obtain laser calibration point cloud sub-data corresponding to the background area and non-background area of the target scene, and combine all laser calibration point cloud sub-data into laser calibration point cloud data;
基于该目标场景进行拍摄和激光扫描同步执行过程中的执行方位信息，构建该视觉标定点云数据和该激光标定点云数据的一一对应映射信息。Based on the execution orientation information during the simultaneous execution of shooting and laser scanning of the target scene, one-to-one mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed.
上述实施例的有益效果，在对该目标场景进行拍摄过程同步对该目标场景进行全景激光扫描，即每当对该目标场景相应方位区域进行拍摄的同时对该方位区域进行激光扫描，从而得到该目标场景的全景激光扫描数据，再对该全景激光扫描数据进行卡尔曼滤波处理，以此降低全景激光扫描数据的噪声成分。再利用相应的神经网络模型对该全景激光扫描数据进行AI识别处理，得到关于该目标场景的背景区域和非背景区域各自对应的激光标定点云子数据，并将所有激光标定点云子数据组成激光标定点云数据，实现对该目标场景的背景区域和非背景区域的区分化激光标定点云数据确定。此外，还基于该目标场景进行拍摄和激光扫描同步执行过程中的执行方位信息，构建该视觉标定点云数据和该激光标定点云数据的一一对应映射信息，便于后续对利用视觉标定点云数据三维建模后得到的三维模型进行激光标定点云数据的校正，提高对三维模型的校正准确性。The beneficial effect of the above embodiment is that during the shooting process of the target scene, the panoramic laser scan of the target scene is performed simultaneously, that is, whenever the corresponding azimuth area of the target scene is photographed, the azimuth area is simultaneously laser scanned, thereby obtaining the The panoramic laser scanning data of the target scene is then processed by Kalman filtering to reduce the noise component of the panoramic laser scanning data. Then use the corresponding neural network model to perform AI recognition processing on the panoramic laser scanning data to obtain laser calibration point cloud sub-data corresponding to the background area and non-background area of the target scene, and combine all laser calibration point cloud sub-data into Laser calibration point cloud data is used to determine the differentiated laser calibration point cloud data of the background area and non-background area of the target scene. In addition, based on the execution orientation information during the simultaneous execution of shooting and laser scanning of the target scene, a one-to-one mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed to facilitate subsequent use of the visual calibration point cloud. The three-dimensional model obtained after three-dimensional modeling of the data is corrected with laser calibration point cloud data to improve the accuracy of the correction of the three-dimensional model.
在另一实施例中，基于该视觉标定点云数据，构建该目标场景对应的初始三维场景模型；对该初始三维场景模型进行场景状态识别，从该初始三维场景模型中提取需要进行校正的三维场景部分；基于该关联映射信息，查找与该三维场景部分对应的激光标定点云数据部分；基于该激光标定点云数据部分，对该三维场景部分进行校正，包括：In another embodiment, based on the visual calibration point cloud data, an initial three-dimensional scene model corresponding to the target scene is constructed; scene state recognition is performed on the initial three-dimensional scene model, and the three-dimensional scene model that needs to be corrected is extracted from the initial three-dimensional scene model. Scene part; based on the associated mapping information, find the laser calibration point cloud data part corresponding to the three-dimensional scene part; based on the laser calibration point cloud data part, correct the three-dimensional scene part, including:
对该视觉标定点云数据进行AI建模处理，构建该目标场景对应的初始三维场景模型；Perform AI modeling processing on the visual calibration point cloud data to construct an initial three-dimensional scene model corresponding to the target scene;
对该初始三维场景模型进行场景失真度识别，得到该初始三维场景模型的背景部分和非背景部分各自对应的场景失真度；若该场景失真度大于或等于预设失真度阈值，则将对应的背景部分/非背景部分确定为该初始三维场景模型中需要进行校正的三维场景部分；Perform scene distortion recognition on the initial three-dimensional scene model to obtain scene distortion degrees corresponding to the background part and the non-background part of the initial three-dimensional scene model; if the scene distortion is greater than or equal to the preset distortion threshold, the corresponding scene distortion will be The background part/non-background part is determined to be the three-dimensional scene part that needs to be corrected in the initial three-dimensional scene model;
基于该关联映射信息，查找与该三维场景部分对应的激光标定点云数据部分；基于该激光标定点云数据部分对该三维场景部分进行三维场景轮廓校正。Based on the associated mapping information, the laser calibration point cloud data part corresponding to the three-dimensional scene part is searched; the three-dimensional scene contour correction is performed on the three-dimensional scene part based on the laser calibration point cloud data part.
上述实施例的有益效果，先以视觉标定点云数据数据为源数据进行AI建模处理，构建该目标场景对应的初始三维场景模型，再对该初始三维场景模型进行关于目标场景的背景区域部分和非背景区域部分各自对应的场景失真度；若该场景失真度大于或等于预设失真度阈值，则将对应的背景区域部分/非背景区域部分确定为该初始三维场景模型中需要进行校正的三维场景部分，从而对场景失真度较大的背景区域部分和非背景区域部分进行准确的提取识别，便于后续有针对性利用背景区域部分和非背景区域部分的激光标定点云数据部分进行关于三维场景轮廓的校正，从而对三维模型进行针对性校正，提高三维建模的准确性。The beneficial effect of the above embodiment is that the visual calibration point cloud data is first used as the source data for AI modeling processing to construct an initial three-dimensional scene model corresponding to the target scene, and then the initial three-dimensional scene model is used to perform the background area part of the target scene. and the corresponding scene distortion degree of the non-background area part; if the scene distortion degree is greater than or equal to the preset distortion threshold, the corresponding background area part/non-background area part is determined to be the part that needs to be corrected in the initial three-dimensional scene model Three-dimensional scene part, so as to accurately extract and identify the background area and non-background area parts with large scene distortion, so as to facilitate the subsequent targeted use of the laser calibration point cloud data of the background area part and non-background area part to perform three-dimensional Correction of scene contours to perform targeted corrections on 3D models and improve the accuracy of 3D modeling.
请参阅图2所示，本申请一实施例提供的基于AI视觉识别的三维建模系统包括：Please refer to Figure 2. A three-dimensional modeling system based on AI visual recognition provided by an embodiment of the present application includes:
场景影像拍摄与处理模块，用于对目标场景进行拍摄，得到目标场景影像，并将该目标场景影像进行分割处理，得到若干目标场景子影像；The scene image shooting and processing module is used to shoot the target scene to obtain the target scene image, and segment the target scene image to obtain several target scene sub-images;
影像AI识别模块，用于对每个目标场景子影像进行AI识别处理，得到该目标场景的视觉标定点云数据；The image AI recognition module is used to perform AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene;
场景激光扫描模块，用于在对该目标场景进行拍摄过程中同步对该目标场景进行激光扫描，得到该目标场景的激光扫描数据；The scene laser scanning module is used to simultaneously perform laser scanning on the target scene during the shooting process of the target scene, and obtain the laser scanning data of the target scene;
激光扫描数据AI识别模块，用于对该激光扫描数据进行AI识别处理，得到该目标场景的激光标定点云数据；The laser scanning data AI recognition module is used to perform AI recognition processing on the laser scanning data to obtain the laser calibration point cloud data of the target scene;
关联映射确定模块，用于基于对该目标场景进行拍摄和激光扫描同步执行属性信息，构建该视觉标定点云数据和该激光标定点云数据的关联映射信息；The associated mapping determination module is used to construct the associated mapping information of the visual calibration point cloud data and the laser calibration point cloud data based on the simultaneous execution attribute information of shooting and laser scanning of the target scene;
三维场景模型构建与识别模块，用于基于该视觉标定点云数据，构建该目标场景对应的初始三维场景模型；对该初始三维场景模型进行场景状态识别，从该初始三维场景模型中提取需要进行校正的三维场景部分；The three-dimensional scene model construction and recognition module is used to construct an initial three-dimensional scene model corresponding to the target scene based on the visual calibration point cloud data; perform scene state recognition on the initial three-dimensional scene model, and extract the required steps from the initial three-dimensional scene model Corrected 3D scene portion;
三维场景模型校正模块，用于基于该关联映射信息，查找与该三维场景部分对应的激光标定点云数据部分；基于该激光标定点云数据部分，对该三维场景部分进行校正。The three-dimensional scene model correction module is used to find the laser calibration point cloud data part corresponding to the three-dimensional scene part based on the associated mapping information; and correct the three-dimensional scene part based on the laser calibration point cloud data part.
上述实施例的有益效果，该基于AI视觉识别的三维建模系统对目标场景进行拍摄，得到目标场景影像，并对每个目标场景子影像进行AI识别处理，得到目标场景的视觉标定点云数据；在拍摄目标场景时同步进行激光扫描，得到目标场景的激光扫描数据，并对其进行AI识别处理，得到目标场景的激光标定点云数据；基于视觉标定点云数据，构建目标场景的初始三维场景模型；再利用激光标定点云数据，对初始三维场景模型相应的三维场景部分进行校正，通过拍摄和激光扫描形成不同三维建模数据源，避免采用单一数据源进行三维建模过程中容易产生建模失真的情况，能够对三维模型进行针对性校正，提高三维建模的准确性和效率。The beneficial effect of the above embodiment is that the three-dimensional modeling system based on AI visual recognition shoots the target scene to obtain the target scene image, and performs AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene. ; Perform laser scanning simultaneously when shooting the target scene to obtain the laser scanning data of the target scene, and perform AI recognition processing on it to obtain the laser calibration point cloud data of the target scene; based on the visual calibration point cloud data, construct an initial three-dimensional view of the target scene Scene model; then use laser calibration point cloud data to correct the corresponding 3D scene part of the initial 3D scene model, and form different 3D modeling data sources through shooting and laser scanning to avoid the problems that are easy to occur in the 3D modeling process using a single data source. In case of modeling distortion, the 3D model can be corrected in a targeted manner to improve the accuracy and efficiency of 3D modeling.
在另一实施例中，该场景影像拍摄与处理模块用于对目标场景进行拍摄，得到目标场景影像，并将该目标场景影像进行分割处理，得到若干目标场景子影像，包括：In another embodiment, the scene image shooting and processing module is used to shoot the target scene to obtain the target scene image, and segment the target scene image to obtain several target scene sub-images, including:
对目标场景进行扫描拍摄，得到目标场景全景影像；对该目标场景全景影像依次进行像素锐化处理和像素轮廓识别处理，得到该目标场景全景影像的像素轮廓特征信息；Scan and shoot the target scene to obtain a panoramic image of the target scene; perform pixel sharpening processing and pixel contour recognition processing on the panoramic image of the target scene in sequence to obtain the pixel contour feature information of the panoramic image of the target scene;
基于该像素轮廓特征信息，对该目标场景全景影像进行分割处理，得到关于该目标场景的背景区域和非背景区域的若干目标场景子影像；Based on the pixel contour feature information, perform segmentation processing on the target scene panoramic image to obtain several target scene sub-images regarding the background area and non-background area of the target scene;
该影像AI识别模块用于对每个目标场景子影像进行AI识别处理，得到该目标场景的视觉标定点云数据，包括：The image AI recognition module is used to perform AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene, including:
对每个目标场景子影像进行AI识别处理，得到每个目标场景子影像的所有关键像素点的点云数据；其中，该关键像素点是指该目标场景子影像的画面上背景物体和非背景物体各自的边界轮廓点；再基于所有目标场景子影像在该目标场景影像的位置信息，将所有目标场景子影像各自对应的点云数据组成该目标场景的视觉标定点云数据。Perform AI recognition processing on each target scene sub-image to obtain point cloud data of all key pixels of each target scene sub-image; where the key pixels refer to background objects and non-background objects on the screen of the target scene sub-image. The respective boundary contour points of the object; and then based on the position information of all target scene sub-images in the target scene image, the corresponding point cloud data of all target scene sub-images is composed of the visual calibration point cloud data of the target scene.
上述实施例的有益效果，在对目标场景的实际扫描拍摄中，可利用360度摄像机对目标场景进行全景扫描拍摄，得到目标场景全景影像，再对该目标场景全景影像依次进行像素锐化处理和像素轮廓识别处理，得到相应的像素轮廓特征信息，以此对目标场景全景影像进行背景区域和非背景区域的标定区分，这样以该像素轮廓特征信息为基准，对该目标场景全景影像进行分割处理，得到关于该目标场景的背景区域和非背景区域的若干目标场景子影像，便于后续针对背景区域和非背景区域进行区分化的点云数据的提取。再利用相应的神经网络模型对每个目标场景子影像进行AI识别处理，得到每个目标场景子影像的所有关键像素点的点云数据，从而实现对目标场景全景影像的全面准确点云数据提取。The beneficial effect of the above embodiment is that in the actual scanning and shooting of the target scene, a 360-degree camera can be used to perform panoramic scanning and shooting of the target scene to obtain a panoramic image of the target scene, and then perform pixel sharpening and processing on the panoramic image of the target scene in sequence. Pixel contour recognition processing is used to obtain corresponding pixel contour feature information, so as to calibrate and distinguish the background area and non-background area of the target scene panoramic image. In this way, the target scene panoramic image is segmented based on the pixel outline feature information. , obtain several target scene sub-images about the background area and non-background area of the target scene, which facilitates the subsequent extraction of point cloud data that differentiates between background areas and non-background areas. Then use the corresponding neural network model to perform AI recognition processing on each target scene sub-image, and obtain the point cloud data of all key pixels of each target scene sub-image, thereby achieving comprehensive and accurate point cloud data extraction for the target scene panoramic image. .
在另一实施例中，该场景激光扫描模块用于在对该目标场景进行拍摄过程中同步对该目标场景进行激光扫描，得到该目标场景的激光扫描数据，包括：In another embodiment, the scene laser scanning module is used to simultaneously perform laser scanning on the target scene during the shooting process of the target scene, and obtain laser scanning data of the target scene, including:
在对该目标场景进行拍摄过程中同步对该目标场景进行全景激光扫描，得到该目标场景的全景激光扫描数据，并对该全景激光扫描数据进行卡尔曼滤波处理；During the shooting process of the target scene, panoramic laser scanning is performed simultaneously on the target scene to obtain panoramic laser scanning data of the target scene, and Kalman filtering is performed on the panoramic laser scanning data;
该激光扫描数据AI识别模块用于对该激光扫描数据进行AI识别处理，得到该目标场景的激光标定点云数据，包括：The laser scanning data AI recognition module is used to perform AI recognition processing on the laser scanning data to obtain laser calibration point cloud data of the target scene, including:
对该全景激光扫描数据进行AI识别处理，得到关于该目标场景的背景区域和非背景区域各自对应的激光标定点云子数据，并将所有激光标定点云子数据组成激光标定点云数据；Perform AI recognition processing on the panoramic laser scanning data to obtain laser calibration point cloud sub-data corresponding to the background area and non-background area of the target scene, and combine all laser calibration point cloud sub-data into laser calibration point cloud data;
该关联映射确定模块用于基于对该目标场景进行拍摄和激光扫描同步执行属性信息，构建该视觉标定点云数据和该激光标定点云数据的关联映射信息，包括：The association mapping determination module is used to construct the association mapping information of the visual calibration point cloud data and the laser calibration point cloud data based on the synchronous execution attribute information of shooting and laser scanning of the target scene, including:
基于该目标场景进行拍摄和激光扫描同步执行过程中的执行方位信息，构建该视觉标定点云数据和该激光标定点云数据的一一对应映射信息。Based on the execution orientation information during the simultaneous execution of shooting and laser scanning of the target scene, one-to-one mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed.
上述实施例的有益效果，在对该目标场景进行拍摄过程同步对该目标场景进行全景激光扫描，即每当对该目标场景相应方位区域进行拍摄的同时对该方位区域进行激光扫描，从而得到该目标场景的全景激光扫描数据，再对该全景激光扫描数据进行卡尔曼滤波处理，以此降低全景激光扫描数据的噪声成分。再利用相应的神经网络模型对该全景激光扫描数据进行AI识别处理，得到关于该目标场景的背景区域和非背景区域各自对应的激光标定点云子数据，并将所有激光标定点云子数据组成激光标定点云数据，实现对该目标场景的背景区域和非背景区域的区分化激光标定点云数据确定。此外，还基于该目标场景进行拍摄和激光扫描同步执行过程中的执行方位信息，构建该视觉标定点云数据和该激光标定点云数据的一一对应映射信息，便于后续对利用视觉标定点云数据三维建模后得到的三维模型进行激光标定点云数据的校正，提高对三维模型的校正准确性。The beneficial effect of the above embodiment is that during the shooting process of the target scene, the panoramic laser scan of the target scene is performed simultaneously, that is, whenever the corresponding azimuth area of the target scene is photographed, the azimuth area is simultaneously laser scanned, thereby obtaining the The panoramic laser scanning data of the target scene is then processed by Kalman filtering to reduce the noise component of the panoramic laser scanning data. Then use the corresponding neural network model to perform AI recognition processing on the panoramic laser scanning data to obtain laser calibration point cloud sub-data corresponding to the background area and non-background area of the target scene, and combine all laser calibration point cloud sub-data into Laser calibration point cloud data is used to determine the differentiated laser calibration point cloud data of the background area and non-background area of the target scene. In addition, based on the execution orientation information during the simultaneous execution of shooting and laser scanning of the target scene, a one-to-one mapping information of the visual calibration point cloud data and the laser calibration point cloud data is constructed to facilitate subsequent use of the visual calibration point cloud. The three-dimensional model obtained after three-dimensional modeling of the data is corrected with laser calibration point cloud data to improve the accuracy of the correction of the three-dimensional model.
在另一实施例中，该三维场景模型构建与识别模块用于基于该视觉标定点云数据，构建该目标场景对应的初始三维场景模型；对该初始三维场景模型进行场景状态识别，从该初始三维场景模型中提取需要进行校正的三维场景部分，包括：In another embodiment, the three-dimensional scene model construction and identification module is used to construct an initial three-dimensional scene model corresponding to the target scene based on the visual calibration point cloud data; perform scene state recognition on the initial three-dimensional scene model, and start from the initial three-dimensional scene model. Extract the parts of the 3D scene that need to be corrected from the 3D scene model, including:
对该视觉标定点云数据进行AI建模处理，构建该目标场景对应的初始三维场景模型；Perform AI modeling processing on the visual calibration point cloud data to construct an initial three-dimensional scene model corresponding to the target scene;
对该初始三维场景模型进行场景失真度识别，得到该初始三维场景模型的背景部分和非背景部分各自对应的场景失真度；若该场景失真度大于或等于预设失真度阈值，则将对应的背景部分/非背景部分确定为该初始三维场景模型中需要进行校正的三维场景部分；Perform scene distortion recognition on the initial three-dimensional scene model to obtain scene distortion degrees corresponding to the background part and the non-background part of the initial three-dimensional scene model; if the scene distortion is greater than or equal to the preset distortion threshold, the corresponding scene distortion will be The background part/non-background part is determined to be the three-dimensional scene part that needs to be corrected in the initial three-dimensional scene model;
该三维场景模型校正模块用于基于该关联映射信息，查找与该三维场景部分对应的激光标定点云数据部分；基于该激光标定点云数据部分，对该三维场景部分进行校正，包括：The three-dimensional scene model correction module is used to find the laser calibration point cloud data part corresponding to the three-dimensional scene part based on the associated mapping information; based on the laser calibration point cloud data part, correct the three-dimensional scene part, including:
基于该关联映射信息，查找与该三维场景部分对应的激光标定点云数据部分；基于该激光标定点云数据部分对该三维场景部分进行三维场景轮廓校正。Based on the associated mapping information, the laser calibration point cloud data part corresponding to the three-dimensional scene part is searched; the three-dimensional scene contour correction is performed on the three-dimensional scene part based on the laser calibration point cloud data part.
上述实施例的有益效果，先以视觉标定点云数据数据为源数据进行AI建模处理，构建该目标场景对应的初始三维场景模型，再对该初始三维场景模型进行关于目标场景的背景区域部分和非背景区域部分各自对应的场景失真度；若该场景失真度大于或等于预设失真度阈值，则将对应的背景区域部分/非背景区域部分确定为该初始三维场景模型中需要进行校正的三维场景部分，从而对场景失真度较大的背景区域部分和非背景区域部分进行准确的提取识别，便于后续有针对性利用背景区域部分和非背景区域部分的激光标定点云数据部分进行关于三维场景轮廓的校正，从而对三维模型进行针对性校正，提高三维建模的准确性。The beneficial effect of the above embodiment is that the visual calibration point cloud data is first used as the source data for AI modeling processing to construct an initial three-dimensional scene model corresponding to the target scene, and then the initial three-dimensional scene model is used to perform the background area part of the target scene. and the corresponding scene distortion degree of the non-background area part; if the scene distortion degree is greater than or equal to the preset distortion threshold, the corresponding background area part/non-background area part is determined to be the part that needs to be corrected in the initial three-dimensional scene model Three-dimensional scene part, so as to accurately extract and identify the background area part and non-background area part with large scene distortion, so as to facilitate the subsequent targeted use of the laser calibration point cloud data part of the background area part and non-background area part to conduct three-dimensional analysis. Correction of scene contours to perform targeted corrections on 3D models and improve the accuracy of 3D modeling.
总体而言，该基于AI视觉识别的三维建模方法和系统对目标场景进行拍摄，得到目标场景影像，并对每个目标场景子影像进行AI识别处理，得到目标场景的视觉标定点云数据；在拍摄目标场景时同步进行激光扫描，得到目标场景的激光扫描数据，并对其进行AI识别处理，得到目标场景的激光标定点云数据；基于视觉标定点云数据，构建目标场景的初始三维场景模型；再利用激光标定点云数据，对初始三维场景模型相应的三维场景部分进行校正，通过拍摄和激光扫描形成不同三维建模数据源，避免采用单一数据源进行三维建模过程中容易产生建模失真的情况，能够对三维模型进行针对性校正，提高三维建模的准确性和效率。Generally speaking, the three-dimensional modeling method and system based on AI visual recognition shoots the target scene to obtain the target scene image, and performs AI recognition processing on each target scene sub-image to obtain the visual calibration point cloud data of the target scene; Laser scanning is performed simultaneously when shooting the target scene to obtain the laser scanning data of the target scene, and AI recognition processing is performed on it to obtain the laser calibration point cloud data of the target scene; based on the visual calibration point cloud data, an initial three-dimensional scene of the target scene is constructed. model; then use laser calibration point cloud data to correct the corresponding 3D scene parts of the initial 3D scene model, and form different 3D modeling data sources through shooting and laser scanning to avoid the construction that is easy to occur in the 3D modeling process using a single data source. In case of model distortion, the 3D model can be corrected in a targeted manner to improve the accuracy and efficiency of 3D modeling.
上述仅为本发明的一个具体实施方式，其它基于本发明构思的前提下做出的任何改进都视为本发明的保护范围。The above is only a specific embodiment of the present invention, and any other improvements made based on the concept of the present invention are deemed to be within the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311426042.7ACN117351154A (en) | 2023-10-31 | 2023-10-31 | Three-dimensional modeling method and system based on AI visual recognition |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311426042.7ACN117351154A (en) | 2023-10-31 | 2023-10-31 | Three-dimensional modeling method and system based on AI visual recognition |
| Publication Number | Publication Date |
|---|---|
| CN117351154Atrue CN117351154A (en) | 2024-01-05 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311426042.7APendingCN117351154A (en) | 2023-10-31 | 2023-10-31 | Three-dimensional modeling method and system based on AI visual recognition |
| Country | Link |
|---|---|
| CN (1) | CN117351154A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118859238A (en)* | 2024-06-26 | 2024-10-29 | 建元未来城市投资发展有限公司 | Three-dimensional data acquisition method for underground pipe gallery equipment based on SLAM and AI |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118859238A (en)* | 2024-06-26 | 2024-10-29 | 建元未来城市投资发展有限公司 | Three-dimensional data acquisition method for underground pipe gallery equipment based on SLAM and AI |
| Publication | Publication Date | Title |
|---|---|---|
| CN111476709B (en) | Face image processing method and device and electronic equipment | |
| WO2021115071A1 (en) | Three-dimensional reconstruction method and apparatus for monocular endoscope image, and terminal device | |
| CN111079545A (en) | Three-dimensional target detection method and system based on image restoration | |
| CN113191954B (en) | Panoramic image stitching method based on binocular camera | |
| CN106403845A (en) | 3D sensor system and 3D data acquisition method | |
| CN101639947A (en) | Image-based plant three-dimensional shape measurement and reconstruction method and system | |
| CN106408597A (en) | Neighborhood entropy and consistency detection-based SAR (synthetic aperture radar) image registration method | |
| CN117173225B (en) | High-precision registration method for complex PCB | |
| CN112257713A (en) | Image processing method, image processing device, electronic equipment and computer readable storage medium | |
| CN105574533A (en) | Image feature extraction method and device | |
| JP2015201123A (en) | Subject recognition device, subject recognition method, and subject recognition program | |
| CN113592953B (en) | Binocular non-cooperative target pose measurement method based on feature point set | |
| CN113112531B (en) | Image matching method and device | |
| CN117351154A (en) | Three-dimensional modeling method and system based on AI visual recognition | |
| CN105654479A (en) | Multispectral image registering method and multispectral image registering device | |
| CN114626118A (en) | Building indoor model generation method and device | |
| CN108764230A (en) | A kind of bank's card number automatic identifying method based on convolutional neural networks | |
| CN114742705B (en) | An image stitching method based on halcon | |
| CN116385502A (en) | An Image Registration Method Based on Region Search under Geometric Constraints | |
| CN111681271A (en) | A multi-channel multi-spectral camera registration method, system and medium | |
| CN111626241A (en) | Face detection method and device | |
| CN111696161B (en) | Calibration method and system for external parameters of double-station camera | |
| CN118397489A (en) | A method, device, electronic terminal and medium for identifying and deduplicating defects in power scenarios | |
| CN117876713A (en) | A tower crane hook distance measurement method, device, equipment and storage medium | |
| US8363941B2 (en) | Method and program for extracting silhouette image and method and program for constructing three dimensional model |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |