WO2020211427A1 - 基于扫描点云数据的分割与识别方法、系统及存储介质 - Google Patents
基于扫描点云数据的分割与识别方法、系统及存储介质 Download PDFInfo
- Publication number
- WO2020211427A1 WO2020211427A1 PCT/CN2019/126978 CN2019126978W WO2020211427A1 WO 2020211427 A1 WO2020211427 A1 WO 2020211427A1 CN 2019126978 W CN2019126978 W CN 2019126978W WO 2020211427 A1 WO2020211427 A1 WO 2020211427A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- point cloud
- scene
- segmentation
- scanning
- data
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Definitions
- the invention relates to the field of three-dimensional reconstruction and pattern recognition, in particular to a method, system and storage medium for segmentation and recognition based on scanning point cloud data.
- scanning-based 3D reconstruction technology In the realization of large scenes such as smart parks and smart cities, scanning-based 3D reconstruction technology is widely used. This technology generally first scans or collects 3D information of the surrounding environment through scanning equipment such as cameras and aerial aircrafts, and then scans or collects The three-dimensional information reconstructs the three-dimensional model of the surrounding environment.
- the prior art requires point cloud segmentation to separate the point cloud data corresponding to different types of objects before performing 3D reconstruction. Perform point cloud modeling for each object separately.
- this point cloud segmentation method in the prior art adopts the segmentation method before modeling. When there are many types of objects, the point cloud segmentation is time-consuming and reduces the modeling efficiency.
- this point cloud segmentation method only uses the segmentation results for modeling, and does not use the segmentation results for further operations such as counting the number of objects of the same type.
- the degree of intelligence is not high, and it is difficult to meet application scenarios such as smart parks and smart cities. The high demands.
- the purpose of the embodiments of the present invention is to provide a method, system and storage medium for segmentation and recognition based on scanning point cloud data.
- the segmentation and recognition method based on scanning point cloud data includes the following steps:
- An artificial intelligence method is used to intelligently recognize the point cloud of the first object in the scene.
- the intelligent recognition includes recognizing the type and corresponding quantity of the first object, and the first object includes indoor objects and outdoor objects.
- the step of scanning the scene and generating a three-dimensional model of the scene according to the scanned data specifically includes:
- Scanning three-dimensional data of the scene by scanning equipment including aerial scanning equipment, indoor scanning equipment and outdoor scanning equipment;
- the server performs three-dimensional reconstruction to obtain the three-dimensional model of the scene, the corresponding link, and the corresponding point cloud data.
- the step of segmenting the scanned point cloud data to obtain the point cloud of the first object in the scene specifically includes:
- the point cloud of the first object in the scene is segmented from the scanned point cloud data.
- the step of using artificial intelligence to intelligently recognize the point cloud of the first object in the scene specifically includes:
- the step of inputting the point cloud of the first object in the scene into the point cloud recognition model to identify the type of the point cloud of the first object in the scene specifically includes:
- the point cloud of the first object in the scene is input into the point cloud recognition model, and the type of the point cloud of the indoor object is recognized.
- the type of the point cloud of the indoor object includes chair, table, computer, ceiling, floor, wall, glass mirror and window.
- the step of inputting the point cloud of the first object in the scene into the point cloud recognition model to identify the type of the point cloud of the first object in the scene specifically includes:
- the point cloud of the first object in the scene is input into the point cloud recognition model, and the type of the point cloud of the outdoor object is recognized.
- the type of the point cloud of the outdoor object includes cables, tables, trees, roads, buildings, lamp posts, and vehicles And video capture device.
- Segmentation and recognition system based on scanning point cloud data including:
- Scanning and modeling module used to scan the scene and generate a three-dimensional model of the scene according to the scanned data
- the acquisition module is used to acquire scan point cloud data from the 3D model of the scene
- the segmentation module is used to segment the point cloud data to obtain the point cloud of the first object in the scene;
- the intelligent recognition module is used to intelligently recognize the point cloud of the first object in the scene by using artificial intelligence.
- the intelligent recognition includes recognizing the type and corresponding quantity of the first object.
- the first object includes indoor objects and outdoor objects. Object.
- Segmentation and recognition system based on scanning point cloud data including:
- At least one processor At least one processor
- At least one memory for storing at least one program
- the at least one processor realizes the segmentation and recognition method based on scanning point cloud data according to the present invention.
- the storage medium stores therein instructions executable by the processor, and the instructions executable by the processor are used to implement the segmentation and recognition method based on scanning point cloud data according to the present invention when the instructions are executed by the processor.
- the embodiment of the present invention first obtains scanned point cloud data from the three-dimensional model of the scene generated after scanning, then performs point cloud segmentation, and finally performs intelligent recognition.
- the point cloud segmentation is performed after the modeling is completed, which will not reduce the modeling efficiency due to the time-consuming point cloud segmentation, and the modeling efficiency is high; after the point cloud segmentation is completed, the artificial intelligence method is used to intelligently identify the type and the first object in the scene
- the corresponding number is convenient for calculating the number of objects of the same type and other objects through artificial intelligence, and the degree of intelligence is high, which meets the requirements of application scenarios such as smart parks and smart cities.
- Figure 1 is a structural block diagram of a scanning modeling and intelligent recognition system according to an embodiment of the present invention
- FIG. 2 is a flowchart of a method for segmentation and recognition based on scanning point cloud data provided by an embodiment of the present invention
- FIG. 3 is a structural block diagram of a segmentation and recognition system based on scanning point cloud data provided by an embodiment of the invention
- FIG. 4 is another structural block diagram of a segmentation and recognition system based on scanning point cloud data provided by an embodiment of the invention.
- This embodiment describes the architecture of the scanning modeling and intelligent recognition system adopted in the present invention.
- the system mainly includes scanning equipment, servers and display modules.
- the scanning device is used to scan objects in scenes such as industrial parks and cities, and upload the scanned data to the server.
- the object can be a symmetrical object, an asymmetrical object with an uneven surface, or an environment or a person.
- the scanning device may be an aerial scanning device, an indoor scanning device, or an outdoor scanning device.
- the aerial scanning device may be an aerial photographing device such as an aerial photographing plane, which is used to scan the three-dimensional data of the area within the scene (such as the entire park).
- Indoor scanning equipment used to scan the three-dimensional data of the indoor environment (such as the interior of a certain floor of a building in the park).
- the indoor scanning device can be a handheld scanning device (such as a camera with a support frame) or other automatic scanning equipment (such as an automatic scanning robot).
- Outdoor scanning equipment used to scan 3D data in outdoor environments (such as a road in a park, etc.).
- the outdoor scanning equipment can be a handheld scanning equipment (such as a camera with a support frame) or other automatic scanning equipment (such as an automatic scanning robot).
- Three-dimensional data includes data such as two-dimensional pictures and depth information.
- the scanning device may be integrated with a GPU chip, which can perform preliminary processing on the collected data such as two-dimensional pictures and depth information locally (such as preliminary stitching of two-dimensional pictures according to depth information, etc.), which reduces the processing burden of the server .
- the server is used to perform 3D reconstruction according to the data uploaded by the scanning device to generate a 3D model of the scene.
- the content of 3D reconstruction includes model repair, editing, cropping, surface reduction, mold reduction, compression, material processing, texture processing and lighting processing.
- the server is also used to generate links to the three-dimensional model of the scene (such as URL links, etc.), so that any computing device that supports browsers (including smart phones, tablets, laptops, smart watches, smart TVs, computers, etc.)
- the 3D model can be accessed through this link.
- the server may be a background server, cloud server, etc. that can communicate with the scanning device in a wired or wireless manner.
- the three-dimensional model of the scene is composed of multiple point clouds (collections of points), so the server can also provide corresponding scanned point cloud data after generating the three-dimensional model of the scene to facilitate subsequent segmentation and intelligent recognition.
- the server can also be used to segment the point cloud of each object in the scene from the scanned point cloud data according to the relationship between the points in the scanned point cloud data and the neighboring points, and then combine the artificial intelligence method to identify each object. type.
- the recognition result server combined with artificial intelligence can automatically count the number of objects of the same type, eliminating the need for manual calculation of the number, greatly improving efficiency and convenience for users.
- the display module is used to display the three-dimensional model of the scene and the results of intelligent recognition (such as the names and corresponding numbers of various types of objects).
- the display module can be implemented by any of AR display devices, VR display devices, mobile terminals, tablet computers, PC computers, air screens, LED displays, LCD displays, OLED displays, and dot matrix displays, etc. .
- an embodiment of the present invention provides a method for segmentation and recognition based on scanning point cloud data, which includes the following steps:
- An artificial intelligence method is used to intelligently recognize the point cloud of the first object in the scene.
- the intelligent recognition includes recognizing the type and corresponding quantity of the first object, and the first object includes indoor objects and outdoor objects.
- the first object in the scene may include multiple objects, people, and other objects in a large scene.
- the scanned point cloud data is segmented, and the point cloud obtained for the first object in the scene also contains multiple point clouds, and the specific types of these segmented point clouds can be obtained after being recognized by artificial intelligence methods.
- each segmented point cloud is corresponding to the type, and the specific number of each type of object can be automatically counted, which is very convenient.
- this embodiment performs point cloud segmentation after the modeling is completed. Compared with the existing method of first performing point cloud segmentation and then modeling, it will not reduce the modeling efficiency due to the time-consuming point cloud segmentation. The mold efficiency is higher.
- the artificial intelligence method is used to intelligently identify the type and corresponding quantity of the first object in the scene, so that the artificial intelligence can count the number of objects of the same type and other objects, with a high degree of intelligence. It meets the requirements of application scenarios such as smart parks and smart cities.
- the step of scanning the scene and generating a three-dimensional model of the scene according to the scanned data specifically includes:
- Scanning three-dimensional data of the scene by scanning equipment including aerial scanning equipment, indoor scanning equipment and outdoor scanning equipment;
- the server performs three-dimensional reconstruction to obtain the three-dimensional model of the scene, the corresponding link, and the corresponding point cloud data.
- this embodiment can easily scan the 3D data of large scenes such as industrial parks and cities, and can quickly generate 3D models and point cloud data of the scenes through the 3D reconstruction of the server. For subsequent segmentation and intelligent recognition.
- the step of segmenting the scanned point cloud data to obtain the point cloud of the first object in the scene specifically includes:
- the point cloud of the first object in the scene is segmented from the scanned point cloud data.
- scanning the relationship between the points in the point cloud data and the neighboring points reflects the association between the points in the point cloud data.
- points that meet the division criteria can be divided into the same type of point cloud. For example, a point whose distance to a certain point is within a preset threshold can be included in the point cloud to which the point belongs.
- the classification standard may not only be based on distance, and other standards (such as attributes such as color and size) are also applicable to this embodiment.
- the step of intelligently identifying the point cloud of the first object in the scene using an artificial intelligence method specifically includes:
- the tag is used to identify the type of point cloud data in the sample, that is, the point cloud data in the input sample and the corresponding type are known, so that the artificial intelligence method can be used to identify the point cloud type.
- Model There will be new point cloud data input later, even if its type is unknown, it can be identified using this model.
- the input sample can be either the point cloud data of a predetermined object, or the point cloud data of an object newly generated after training or recognition, so that the point cloud recognition model can be continuously self-learning during training And update to improve the accuracy and accuracy of the recognition model.
- the number of point clouds of each type of first object in the scene represents the number of each type of first object in the scene.
- the step of inputting the point cloud of the first object in the scene into the point cloud recognition model to identify the type of the point cloud of the first object in the scene specifically includes:
- the point cloud of the first object in the scene is input into the point cloud recognition model, and the type of the point cloud of the indoor object is recognized.
- the type of the point cloud of the indoor object includes chair, table, computer, ceiling, floor, wall, glass mirror and window.
- the type of objects such as indoor objects can be recognized through the pre-trained point cloud recognition model of indoor objects, which is very convenient and fast.
- the step of inputting the point cloud of the first object in the scene into the point cloud recognition model to identify the type of the point cloud of the first object in the scene specifically includes:
- the point cloud of the first object in the scene is input into the point cloud recognition model, and the type of the point cloud of the outdoor object is recognized.
- the type of the point cloud of the outdoor object includes cables, tables, trees, roads, buildings, lamp posts, and vehicles And video capture device.
- the type of objects such as outdoor objects can be recognized through the pre-trained point cloud recognition model of outdoor objects, which is very convenient and fast.
- Cables include cables, optical cables, etc.
- the lamp post is used to place the lighting lamp.
- the video capture device can be a surveillance camera, CCTV closed-circuit television and other devices.
- the three-dimensional model of the scene can be intuitively displayed through the display module of FIG. 1, and the intelligent recognition results such as the type and corresponding number of each object in the scene can also be directly displayed, which is very convenient.
- an embodiment of the present invention provides a segmentation and recognition system based on scanning point cloud data, including:
- Scanning and modeling module used to scan the scene and generate a three-dimensional model of the scene according to the scanned data
- the acquisition module is used to acquire scan point cloud data from the 3D model of the scene
- the segmentation module is used to segment the scanned point cloud data to obtain the point cloud of the first object in the scene;
- the intelligent recognition module is used to intelligently recognize the point cloud of the first object in the scene using an artificial intelligence method.
- the intelligent recognition includes recognizing the type and corresponding number of the first object.
- the first object includes indoor objects and outdoor objects. Object.
- an embodiment of the present invention provides a segmentation and recognition system based on scanning point cloud data, including:
- At least one processor At least one processor
- At least one memory for storing at least one program
- the at least one processor realizes the segmentation and recognition method based on scanning point cloud data according to the present invention.
- the embodiment of the present invention also provides a storage medium in which instructions executable by the processor are stored. When the instructions executable by the processor are executed by the processor, they are used to implement the scanning point cloud data based on the present invention. Method of segmentation and recognition.
- Step 1 Scan the indoor and outdoor panorama of the park through aerial photography aircraft, indoor scanning equipment, etc., and upload the scanned data to the server;
- Step 2 The server performs three-dimensional reconstruction based on the uploaded data to reconstruct the three-dimensional model of the park through model repair, editing, cropping, surface reduction, model reduction, compression, material processing, texture processing and lighting processing, etc. link;
- Step 3 Obtain the point cloud data of the park from the three-dimensional model of the park;
- Step 4 Perform point cloud segmentation on the acquired point cloud data to obtain the point cloud of each object in the park;
- Step 6 Use artificial intelligence to intelligently identify the point cloud of each object in the park, and obtain the type (name) of each object in the park and the number of each type of object;
- the type of each point cloud (that is, the type of each object) can be identified through the artificial intelligence method. Separate each point cloud, so that only the type of point cloud and each point cloud need to be correlated, and the number of each type of object can be obtained by counting the number of point clouds of the same type. For example, it can identify how many chairs, how many tables, how many computers, which is the ceiling, which is the floor, which is the wall, which is contained in a certain indoor environment of the park (such as a certain room in a building) Is it a glass mirror, which is a window, etc.
- Step 7 Display the three-dimensional model of the park, the type (or name) of each object in the park, and the number of various types of objects and other intelligent recognition results through the display module.
Abstract
本发明公开了一种基于扫描点云数据的分割与识别方法、系统及存储介质，方法包括：扫描场景并根据扫描的数据生成场景的三维模型；从场景的三维模型获取扫描点云数据；对扫描点云数据进行分割，得到场景内第一对象的点云；采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。本发明在完成建模之后再进行点云分割，不会因点云分割耗时而降低建模效率，建模效率高；在点云分割完成后通过人工智能的方法智能识别出场景内第一对象的类型及对应的数量，便于通过人工智能统计出同类型物体等对象的数量，智能化程度高，可广泛应用于三维重建与模式识别领域。
Description
本发明涉及三维重建与模式识别领域，尤其是一种基于扫描点云数据的分割与识别方法、系统及存储介质。
在智慧园区、智慧城市等大场景的实现方案中，基于扫描的三维重建技术被广泛应用，该技术一般先通过相机、航拍飞机等扫描设备扫描或采集周围环境的三维信息，然后根据扫描或采集的三维信息重建出周围环境的三维模型。
由于扫描的场景包含不同类型的物体，例如地面、建筑物、树木、车辆等，现有技术在进行三维重建之前，需要通过点云分割将不同类型的物体对应的点云数据彼此分割开，以便对各个物体分别进行点云建模。然而，现有技术的这种点云分割方式采用了建模之前的分割方式，在物体的类型较多时会因点云分割耗时而降低建模效率。此外，这种点云分割方式只是利用分割结果来建模，并未利用分割结果来进行同类型物体的数量统计等进一步的操作，智能化程度不高，难以满足智慧园区、智慧城市等应用场景的高要求。
发明内容
为解决上述技术问题，本发明实施例的目的在于：提供一种基于扫描点云数据的分割与识别方法、系统及存储介质。
第一方面，本发明实施例所采取的技术方案是：
基于扫描点云数据的分割与识别方法，包括以下步骤：
扫描场景并根据扫描的数据生成场景的三维模型；
从场景的三维模型获取扫描点云数据；
对扫描点云数据进行分割，得到场景内第一对象的点云；
采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。
进一步，所述扫描场景并根据扫描的数据生成场景的三维模型这一步骤，具体包括：
通过扫描设备扫描场景的三维数据，所述扫描设备包括航拍扫描设备、室内扫描设备和室外扫描设备；
将场景的三维数据上传服务器；
根据场景的三维数据通过服务器进行三维重建，得到场景的三维模型、对应的链接以及对应的点云数据。
进一步，所述对扫描点云数据进行分割，得到场景内第一对象的点云这一步骤，具体为：
根据扫描点云数据中点与近邻点间的关系，从扫描点云数据中分割出场景内第一对象的点云。
进一步，所述采用人工智能的方法对场景内第一对象的点云进行智能识别这一步骤，具体包括：
根据输入的样本和标签，采用人工智能的方法训练点云识别模型；
将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型；
计算场景内各类型第一对象的点云的数量，从而得到第一对象的数量。
进一步，所述将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型这一步骤，具体包括：
将场景内第一对象的点云输入点云识别模型，识别出室内对象的点云的类型，所述室内对象的点云的类型包括椅子、桌子、电脑、天花板、地板、墙、玻璃镜面和窗口。
进一步，所述将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型这一步骤，具体包括：
将场景内第一对象的点云输入点云识别模型，识别出室外对象的点云的类型，所述室外对象的点云的类型包括线缆、桌子、树、道路、建筑、灯柱、车辆和视频采集装置。
进一步，还包括以下步骤：
展示场景的三维模型和智能识别的结果。
第二方面，本发明实施例所采取的技术方案是：
基于扫描点云数据的分割与识别系统，包括：
扫描与建模模块，用于扫描场景并根据扫描的数据生成场景的三维模型；
获取模块，用于从场景的三维模型获取扫描点云数据；
分割模块，用于对点云数据进行分割，得到场景内第一对象的点云；
智能识别模块，用于采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。
第三方面，本发明实施例所采取的技术方案是：
基于扫描点云数据的分割与识别系统，包括：
至少一个处理器；
至少一个存储器，用于存储至少一个程序；
当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如本发明所述的基于扫描点云数据的分割与识别方法。
第四方面，本发明实施例所采取的技术方案是：
存储介质，其中存储有处理器可执行的指令，所述处理器可执行的指令在由处理器执行时用于实现如本发明所述的基于扫描点云数据的分割与识别方法。
上述本发明实施例中的一个或多个技术方案具有如下优点：本发明实施例先从扫描后生成的场景的三维模型中获取扫描点云数据，再进行点云分割，最后进行智能识别，在完成建模之后再进行点云分割，不会因点云分割耗时而降低建模效率，建模效率高；在点云分割完成后通过人工智能的方法智能识别出场景内第一对象的类型及对应的数量，便于通过人工智能统计出同类型物体等对象的数量，智能化程度高，满足了智慧园区、智慧城市等应用场景的要求。
图1为本发明实施例扫描建模与智能识别系统的结构框图；
图2为本发明实施例提供的基于扫描点云数据的分割与识别方法流程图；
图3为发明实施例提供的基于扫描点云数据的分割与识别系统的一种结构框图；
图4为发明实施例提供的基于扫描点云数据的分割与识别系统的另一种结构框图。
下面结合说明书附图和具体实施例对本发明作进一步解释和说明。
本实施例对本发明所采用的扫描建模与智能识别系统的架构进行说明。如图1所示，该系统主要包括扫描设备、服务器和展示模块。
其中，扫描设备，用于对工业园区、城市等场景内的对象进行扫描，并将扫描的数据上传给服务器。对象可以是对称物体，具有不平坦表面的不对称物体，还可以是环境或人物。扫描设备可以是航拍扫描设备、室内扫描设备或室外扫描设备。航拍扫描设备，可以是航拍飞机等航拍设备，用于扫描场景内区域范围(如整个园区)的三维数据。室内扫描设备，用于扫描室内环境(如园区内某栋建筑某层楼的内部)的三维数据。室内扫描设备，可以是手持扫描设备(如带支撑架的相机)或其他自动扫描设备(如自动扫描机器人)。室外扫描设备，用于扫描室外环境(如园区内的某条马路等)的三维数据。室外扫描设备，可以是手持扫描设备(如带支撑架的相机)或其他自动扫描设备(如自动扫描机器人)。三维数据包括二维图片和深度信息等数据。优选地，扫描设备可集成有GPU芯片，能在本地对采集的二维图片和 深度信息等数据进行初步的处理(如将二维图片按深度信息进行初步拼接等)，减轻了服务器的处理负担。
服务器，用于根据扫描设备上传的数据进行三维重建，以生成场景的三维模型。其中，三维重建的内容包括模型修复、剪辑、裁剪、减面、减模、压缩、处理材质、处理贴图和处理灯光。优选地，服务器还用于生成场景的三维模型的链接(如URL链接等)，这样任何支持浏览器的计算设备(包括智能手机、平板电脑、笔记本电脑、智能手表、智能电视、计算机等)都可以通过该链接访问该三维模型。服务器可以是能通过有线或无线的方式与扫描设备通讯的后台服务器、云端服务器等。场景的三维模型是由多个点云(点的集合)组成的，故服务器也可以在生成场景的三维模型后提供对应的扫描点云数据，以便于后续的分割与智能识别。
优选地，服务器还可用于根据扫描点云数据中点与近邻点间的关系，从扫描点云数据中分割出场景内各个对象的点云，再结合人工智能的方法进行识别，从而得到各个对象的类型。同时由于分割时已对点云进行了划分，结合人工智能的识别结果服务器就可以自动统计出同一类型对象的数量，省去了人工计算数量的过程，极大地提升了效率和方便了用户。
展示模块，用于展示场景的三维模型以及智能识别的结果(如各种类型对象的名称和对应的数量等)。展示模块可以采用AR显示设备、VR显示设备、移动终端、平板电脑端、PC电脑端、空气屏、LED显示屏、LCD显示屏、OLED显示屏和点阵显示屏等中的任一种来实现。
如图2所示，本发明实施例提供了一种基于扫描点云数据的分割与识别方法，包括以下步骤：
扫描场景并根据扫描的数据生成场景的三维模型；
从场景的三维模型获取扫描点云数据；
对扫描点云数据进行分割，得到场景内第一对象的点云；
采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。
具体地，场景内第一对象可包含多个物体、人物等大场景内的对象。相应地，对扫描点云数据进行分割，得到场景内第一对象的点云也包含了多个点云，这些分割出的点云的具体类型经人工智能的方法识别后即可得出。这样将分割出的每个点云与类型对应起来，即可自动地统计出各个类型对象的具体数量，十分方便。
由上述的内容可知，本实施例在完成建模之后再进行点云分割，与现有先进行点云分割 再建模的方式相比，不会因点云分割耗时而降低建模效率，建模效率更高。同时，本实施例点云分割完成后通过人工智能的方法智能识别出场景内第一对象的类型及对应的数量，以便于通过人工智能统计出同类型物体等对象的数量，智能化程度高，满足了智慧园区、智慧城市等应用场景的要求。
进一步作为优选的实施方式，所述扫描场景并根据扫描的数据生成场景的三维模型这一步骤，具体包括：
通过扫描设备扫描场景的三维数据，所述扫描设备包括航拍扫描设备、室内扫描设备和室外扫描设备；
将场景的三维数据上传服务器；
根据场景的三维数据通过服务器进行三维重建，得到场景的三维模型、对应的链接以及对应的点云数据。
具体地，基于图1的扫描设备和服务器，本实施例可以很方便地扫描工业园区、城市等大场景的三维数据，并可以通过服务器的三维重建快速生成场景的三维模型以及点云数据，以便于后续的分割与智能识别。
进一步作为优选的实施方式，所述对扫描点云数据进行分割，得到场景内第一对象的点云这一步骤，具体为：
根据扫描点云数据中点与近邻点间的关系，从扫描点云数据中分割出场景内第一对象的点云。
具体地，扫描点云数据中点与近邻点间的关系，反映了点云数据中点与点之间的关联。本实施例可基于该关系，将满足划分标准的点划分到同一类型的点云中去。例如，可将与某点间的距离在预设阈值内的点归入该点所属的点云中。本领域技术人员可以理解的是，划分标准可以不仅仅依据距离，其它的标准(如颜色、大小等属性)同样适用于本实施例。
进一步作为优选的实施方式，所述采用人工智能的方法对场景内第一对象的点云进行智能识别这一步骤，具体包括：
根据输入的样本和标签，采用人工智能的方法训练点云识别模型；
将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型；
计算场景内各类型第一对象的点云的数量，从而得到第一对象的数量。
具体地，标签用于标识样本中点云数据的类型，也就是说，输入的样本中点云数据及对应的类型是已知的，这样通过人工智能的方法可训练出用于识别点云类型的模型。后面再有新的点云数据输入，即使它的类型是未知的，也可以使用该模型来识别出来。另外，输入的 样本既可以是预先给定的对象的点云数据，也可以是在训练或识别后新生成的对象的点云数据，这样点云识别模型在训练时就可以通过不断的自我学习和更新来提高识别模型的精度和准确度。
而场景内各类型第一对象的点云的数量则代表了场景内各类型第一对象的数量。
进一步作为优选的实施方式，所述将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型这一步骤，具体包括：
将场景内第一对象的点云输入点云识别模型，识别出室内对象的点云的类型，所述室内对象的点云的类型包括椅子、桌子、电脑、天花板、地板、墙、玻璃镜面和窗口。
具体地，本实施例在对室内对象进行智能识别时，可通过预训练的室内对象的点云识别模型来识别出室内的物体等对象的类型，十分方便和快捷。
进一步作为优选的实施方式，所述将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型这一步骤，具体包括：
将场景内第一对象的点云输入点云识别模型，识别出室外对象的点云的类型，所述室外对象的点云的类型包括线缆、桌子、树、道路、建筑、灯柱、车辆和视频采集装置。
具体地，本实施例在对室外对象进行智能识别时，可通过预训练的室外对象的点云识别模型来识别出室外的物体等对象的类型，十分方便和快捷。线缆包括电缆、光缆等。灯柱用于安置照明灯。视频采集装置可以是监控摄像头、CCTV闭路电视等装置。
进一步作为优选的实施方式，还包括以下步骤：
展示场景的三维模型和智能识别的结果。
具体地，本实施例在智能识别完成后，可以通过图1的展示模块直观地展示场景的三维模型，还可以直接展示场景内各个对象的类型和对应的数量等智能识别的结果，十分方便。
如图3所示，本发明实施例提供了一种基于扫描点云数据的分割与识别系统，包括：
扫描与建模模块，用于扫描场景并根据扫描的数据生成场景的三维模型；
获取模块，用于从场景的三维模型获取扫描点云数据；
分割模块，用于对扫描点云数据进行分割，得到场景内第一对象的点云；
智能识别模块，用于采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。
上述方法实施例中的内容均适用于本系统实施例中，本系统实施例所具体实现的功能与上述方法实施例相同，并且达到的有益效果与上述方法实施例所达到的有益效果也相同。
如图4所示，本发明实施例提供了一种基于扫描点云数据的分割与识别系统，包括：
至少一个处理器；
至少一个存储器，用于存储至少一个程序；
当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如本发明所述的基于扫描点云数据的分割与识别方法。
上述方法实施例中的内容均适用于本系统实施例中，本系统实施例所具体实现的功能与上述方法实施例相同，并且达到的有益效果与上述方法实施例所达到的有益效果也相同。
本发明实施例还提供了一种存储介质，其中存储有处理器可执行的指令，所述处理器可执行的指令在由处理器执行时用于实现如本发明所述的基于扫描点云数据的分割与识别方法。
以某个工业园区为例，应用本发明的方法对园区进行点云分割与智能识别，具体实现步骤如下：
步骤1：通过航拍飞机、室内扫描设备等对该园区的室内室外全景进行扫描，并将扫描的数据上传服务器；
步骤2：服务器根据上传的数据进行三维重建，以通过模型修复、剪辑、裁剪、减面、减模、压缩、处理材质、处理贴图和处理灯光等处理重建出该园区的三维模型并生成对应的链接；
步骤3：从该园区的三维模型中获取该园区的点云数据；
步骤4：对获取的点云数据进行点云分割，得到该园区内各对象的点云；
步骤6：采用人工智能的方法对该园区内各对象的点云进行智能识别，得到该园区内各对象的类型(名称)以及各类型对象的数量；
具体地，以该园区内各对象为室内环境对象为例，在点云分割完成后，可通过人工智能的方法识别出各点云的类型(即各对象的类型)，同时由于点云分割已经分割出各个点云，这样只需将点云的类型和各个点云对应起来，即可通过统计同一类型的点云的数量得到各个类型对象的数量。例如，可识别出该园区的某个室内环境(如某栋楼的某一个房间)内包含有多少把椅子、多少个桌子、多少个电脑、哪个是天花板、哪个是地板、哪个是墙壁、哪个是玻璃镜面、哪个是窗口等等。
同理，对于该园区的室外环境，可识别该园区的室外环境包含的电缆、树木、马路、灯柱、CCDV摄像头、建筑、车辆和行人等对象及每个对象的数量。
步骤7：通过展示模块展示该园区的三维模型，以及该园区内各对象的类型(或名称)以及各类型对象的数量等智能识别的结果。
以上是对本发明的较佳实施进行了具体说明，但本发明并不限于所述实施例，熟悉本领域的技术人员在不违背本发明精神的前提下还可做作出种种的等同变形或替换，这些等同的变形或替换均包含在本申请权利要求所限定的范围内。
Claims (10)
- 基于扫描点云数据的分割与识别方法，其特征在于：包括以下步骤：扫描场景并根据扫描的数据生成场景的三维模型；从场景的三维模型获取扫描点云数据；对扫描点云数据进行分割，得到场景内第一对象的点云；采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。
- 根据权利要求1所述的基于扫描点云数据的分割与识别方法，其特征在于：所述扫描场景并根据扫描的数据生成场景的三维模型这一步骤，具体包括：通过扫描设备扫描场景的三维数据，所述扫描设备包括航拍扫描设备、室内扫描设备和室外扫描设备；将场景的三维数据上传服务器；根据场景的三维数据通过服务器进行三维重建，得到场景的三维模型、对应的链接以及对应的点云数据。
- 根据权利要求1所述的基于扫描点云数据的分割与识别方法，其特征在于：所述对扫描点云数据进行分割，得到场景内第一对象的点云这一步骤，具体为：根据扫描点云数据中点与近邻点间的关系，从扫描点云数据中分割出场景内第一对象的点云。
- 根据权利要求1所述的基于扫描点云数据的分割与识别方法，其特征在于：所述采用人工智能的方法对场景内第一对象的点云进行智能识别这一步骤，具体包括：根据输入的样本和标签，采用人工智能的方法训练点云识别模型；将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型；计算场景内各类型第一对象的点云的数量，从而得到第一对象的数量。
- 根据权利要求4所述的基于扫描点云数据的分割与识别方法，其特征在于：所述将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型这一步骤，具体包括：将场景内第一对象的点云输入点云识别模型，识别出室内对象的点云的类型，所述室内对象的点云的类型包括椅子、桌子、电脑、天花板、地板、墙、玻璃镜面和窗口。
- 根据权利要求4所述的基于扫描点云数据的分割与识别方法，其特征在于：所述将场景内第一对象的点云输入点云识别模型，识别出场景内第一对象的点云的类型这一步骤，具体包括：将场景内第一对象的点云输入点云识别模型，识别出室外对象的点云的类型，所述室外对象的点云的类型包括线缆、桌子、树、道路、建筑、灯柱、车辆和视频采集装置。
- 根据权利要求1所述的基于扫描点云数据的分割与识别方法，其特征在于：还包括以下步骤：展示场景的三维模型和智能识别的结果。
- 基于扫描点云数据的分割与识别系统，其特征在于：包括：扫描与建模模块，用于扫描场景并根据扫描的数据生成场景的三维模型；获取模块，用于从场景的三维模型获取扫描点云数据；分割模块，用于对扫描点云数据进行分割，得到场景内第一对象的点云；智能识别模块，用于采用人工智能的方法对场景内第一对象的点云进行智能识别，所述智能识别包括识别第一对象的类型及对应的数量，所述第一对象包括室内对象和室外对象。
- 基于扫描点云数据的分割与识别系统，其特征在于：包括：至少一个处理器；至少一个存储器，用于存储至少一个程序；当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如权利要求1-7任一项所述的基于扫描点云数据的分割与识别方法。
- 存储介质，其中存储有处理器可执行的指令，其特征在于：所述处理器可执行的指令在由处理器执行时用于实现如权利要求1-7任一项所述的基于扫描点云数据的分割与识别方法。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910305106.5A CN110223297A (zh) | 2019-04-16 | 2019-04-16 | 基于扫描点云数据的分割与识别方法、系统及存储介质 |
| CN201910305106.5 | 2019-04-16 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020211427A1 true WO2020211427A1 (zh) | 2020-10-22 |
Family
ID=67822607
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/126978 WO2020211427A1 (zh) | 2019-04-16 | 2019-12-20 | 基于扫描点云数据的分割与识别方法、系统及存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110223297A (zh) |
| WO (1) | WO2020211427A1 (zh) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110223297A (zh) * | 2019-04-16 | 2019-09-10 | 广东康云科技有限公司 | 基于扫描点云数据的分割与识别方法、系统及存储介质 |
| CN111080799A (zh) * | 2019-12-04 | 2020-04-28 | 广东康云科技有限公司 | 基于三维建模的场景漫游方法、系统、装置和存储介质 |
| CN112102375B (zh) * | 2020-07-22 | 2024-04-12 | 广州视源电子科技股份有限公司 | 一种点云配准可靠性检测的方法、装置、移动智慧设备 |
| CN113503815A (zh) * | 2021-07-07 | 2021-10-15 | 思灵机器人科技(哈尔滨)有限公司 | 基于光栅的喷涂外型识别方法 |
| CN116774195B (zh) * | 2023-08-22 | 2023-12-08 | 国网天津市电力公司滨海供电分公司 | 多传感器联合标定的激励判断与参数自调节方法及系统 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366250A (zh) * | 2013-07-12 | 2013-10-23 | 中国科学院深圳先进技术研究院 | 基于三维实景数据的市容环境检测方法及系统 |
| CN108052914A (zh) * | 2017-12-21 | 2018-05-18 | 中国科学院遥感与数字地球研究所 | 一种基于slam和图像识别的森林林木资源调查方法 |
| CN108389256A (zh) * | 2017-11-23 | 2018-08-10 | 千寻位置网络有限公司 | 二三维交互式无人机电力杆塔巡检辅助方法 |
| CN110223297A (zh) * | 2019-04-16 | 2019-09-10 | 广东康云科技有限公司 | 基于扫描点云数据的分割与识别方法、系统及存储介质 |
Family Cites Families (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101877128B (zh) * | 2009-12-23 | 2012-09-05 | 中国科学院自动化研究所 | 一种三维场景中不同物体的分割方法 |
| CN102930246B (zh) * | 2012-10-16 | 2015-04-08 | 同济大学 | 一种基于点云片段分割的室内场景识别方法 |
| CN105205858B (zh) * | 2015-09-18 | 2018-04-13 | 天津理工大学 | 一种基于单个深度视觉传感器的室内场景三维重建方法 |
| HUE038849T2 (hu) * | 2016-01-18 | 2018-11-28 | Sick Ag | Detektáló készülék és eljárás jármûtengelyek detektálására |
| CN106570903B (zh) * | 2016-10-13 | 2019-06-18 | 华南理工大学 | 一种基于rgb-d摄像头的视觉识别与定位方法 |
| CN206322194U (zh) * | 2016-10-24 | 2017-07-11 | 杭州非白三维科技有限公司 | 一种基于三维扫描的反欺诈人脸识别系统 |
| CN207037750U (zh) * | 2017-07-21 | 2018-02-23 | 湖南拓视觉信息技术有限公司 | 全场景扫描装置及三维扫描建模系统 |
| CN107748890A (zh) * | 2017-09-11 | 2018-03-02 | 汕头大学 | 一种基于深度图像的视觉抓取方法、装置及其可读存储介质 |
| CN107894911B (zh) * | 2017-11-13 | 2021-04-30 | 中海油常州涂料化工研究院有限公司 | 海上平台的信息管理方法和装置 |
| CN109085966B (zh) * | 2018-06-15 | 2020-09-08 | 广东康云多维视觉智能科技有限公司 | 一种基于云计算的三维展示系统及方法 |
| CN109102537B (zh) * | 2018-06-25 | 2020-03-20 | 中德人工智能研究院有限公司 | 一种二维激光雷达和球幕相机结合的三维建模方法和系统 |
| CN109143207B (zh) * | 2018-09-06 | 2020-11-10 | 百度在线网络技术（北京）有限公司 | 激光雷达内参精度验证方法、装置、设备及介质 |
| CN109345510A (zh) * | 2018-09-07 | 2019-02-15 | 百度在线网络技术（北京）有限公司 | 物体检测方法、装置、设备、存储介质及车辆 |
| CN109344750B (zh) * | 2018-09-20 | 2021-10-22 | 浙江工业大学 | 一种基于结构描述子的复杂结构三维对象识别方法 |
| CN109344813B (zh) * | 2018-11-28 | 2023-11-28 | 北醒（北京）光子科技有限公司 | 一种基于rgbd的目标识别和场景建模方法 |
-
2019
- 2019-04-16 CN CN201910305106.5A patent/CN110223297A/zh active Pending
- 2019-12-20 WO PCT/CN2019/126978 patent/WO2020211427A1/zh active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366250A (zh) * | 2013-07-12 | 2013-10-23 | 中国科学院深圳先进技术研究院 | 基于三维实景数据的市容环境检测方法及系统 |
| CN108389256A (zh) * | 2017-11-23 | 2018-08-10 | 千寻位置网络有限公司 | 二三维交互式无人机电力杆塔巡检辅助方法 |
| CN108052914A (zh) * | 2017-12-21 | 2018-05-18 | 中国科学院遥感与数字地球研究所 | 一种基于slam和图像识别的森林林木资源调查方法 |
| CN110223297A (zh) * | 2019-04-16 | 2019-09-10 | 广东康云科技有限公司 | 基于扫描点云数据的分割与识别方法、系统及存储介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110223297A (zh) | 2019-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020211427A1 (zh) | 基于扫描点云数据的分割与识别方法、系统及存储介质 | |
| US11687382B2 (en) | Systems and methods for motion correction in synthetic images | |
| WO2020228766A1 (zh) | 基于实景建模与智能识别的目标跟踪方法、系统及介质 | |
| US11729495B2 (en) | Directed image capture | |
| CN110009561B (zh) | 一种监控视频目标映射到三维地理场景模型的方法及系统 | |
| WO2020134528A1 (zh) | 目标检测方法及相关产品 | |
| CN108234927B (zh) | 视频追踪方法和系统 | |
| CN108388882B (zh) | 基于全局-局部rgb-d多模态的手势识别方法 | |
| CN109816745B (zh) | 人体热力图展示方法及相关产品 | |
| CN108053449A (zh) | 双目视觉系统的三维重建方法、装置及双目视觉系统 | |
| US20180357819A1 (en) | Method for generating a set of annotated images | |
| WO2023280038A1 (zh) | 一种三维实景模型的构建方法及相关装置 | |
| CN110232731A (zh) | 一种智慧城市系统及其实现方法 | |
| CN110660125B (zh) | 用于电力配网系统的三维建模装置 | |
| WO2020228767A1 (zh) | 基于视频融合的场景动态模拟方法、系统及存储介质 | |
| CN114125310B (zh) | 拍照方法、终端设备及云端服务器 | |
| CN114202622B (zh) | 虚拟建筑生成方法、装置、设备及计算机可读存储介质 | |
| CN113572962A (zh) | 室外自然场景光照估计方法及装置 | |
| CN113487723B (zh) | 基于可量测全景三维模型的房屋在线展示方法及系统 | |
| WO2020228347A1 (zh) | 基于超像素的对象三维模型生成方法、系统及存储介质 | |
| JP7092615B2 (ja) | 影検出装置、影検出方法、影検出プログラム、学習装置、学習方法、及び学習プログラム | |
| CN113228626B (zh) | 视频监控系统和方法 | |
| CN114387445A (zh) | 对象关键点识别方法及装置、电子设备和存储介质 | |
| CN112598780A (zh) | 实例对象模型构建方法及装置、可读介质和电子设备 | |
| Yu et al. | Intelligent visual-IoT-enabled real-time 3D visualization for autonomous crowd management |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19925391 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: 19925391 Country of ref document: EP Kind code of ref document: A1 |