WO2014101600A1 - 三维模型创建方法和设备 - Google Patents

三维模型创建方法和设备 Download PDF

Info

Publication number
WO2014101600A1
WO2014101600A1 PCT/CN2013/087662 CN2013087662W WO2014101600A1 WO 2014101600 A1 WO2014101600 A1 WO 2014101600A1 CN 2013087662 W CN2013087662 W CN 2013087662W WO 2014101600 A1 WO2014101600 A1 WO 2014101600A1
Authority
WO
WIPO (PCT)
Prior art keywords
dimensional
data
checked baggage
volume data
imaging system
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2013/087662
Other languages
English (en)
French (fr)
Inventor
陈志强
张丽
赵自然
李强
顾建平
黄清萍
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Nuctech Co Ltd
Original Assignee
Tsinghua University
Nuctech Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University, Nuctech Co Ltd filed Critical Tsinghua University
Publication of WO2014101600A1 publication Critical patent/WO2014101600A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V5/00—Prospecting or detecting by the use of ionising radiation, e.g. of natural or induced radioactivity
    • G01V5/20—Detecting prohibited goods, e.g. weapons, explosives, hazardous substances, contraband or smuggled objects
    • G01V5/22—Active interrogation, i.e. by irradiating objects or goods using external radiation sources, e.g. using gamma rays or cosmic rays
    • G01V5/226—Active interrogation, i.e. by irradiating objects or goods using external radiation sources, e.g. using gamma rays or cosmic rays using tomography
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional [3D] objects
    • G06V20/653—Three-dimensional [3D] objects by matching three-dimensional models, e.g. conformal mapping of Riemann surfaces
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three-dimensional [3D] modelling for computer graphics
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30112—Baggage; Luggage; Suitcase
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/05—Recognition of patterns representing particular kinds of hidden objects, e.g. weapons, explosives, drugs

Definitions

  • the invention relates to a security inspection of an object, in particular to a method and a device for creating a three-dimensional model. Background technique
  • Perspective imaging is an important means in the field of security inspection.
  • the typical application process is as follows: The equipment scans and images the baggage items in perspective, and the inspector judges the figure, manually marks the suspect area, and makes a semantic description of the area, such as "lighter”. "A bottle of wine” and so on. This process relies too much on human factors. Under the influence of factors such as the occurrence of real dangerous objects with limited probability, limited experience of inspectors, and fatigue, it will lead to missed inspections and serious consequences.
  • the typical means of solving this problem is based on automatic auxiliary detection, which is supplemented by the interaction between the equipment and the security inspector.
  • automatic detection is the main method, the effect is not satisfactory.
  • Typical technologies such as explosives detection, high-density alarms, etc. do not adequately meet application needs.
  • the reason is that on the one hand, technical conditions are limited, such as the overlap of perspectives in DEDR (Dual Energy Digital Radiography), which causes aliasing of objects, and on the other hand, there are few studies in the academic world. Newer technologies such as DECT (Dual Energy Computed Tomography) ) Requires new detection algorithm support.
  • DECT is an advantageous technology to solve this problem. It is developed by DR and CT technology. It can obtain the effective atomic number and equivalent electron density inside the object while acquiring the three-dimensional structure information of the scanned object. Conditions of understanding. However, the existing research objectives are often the target detection of a specific object, and mainly rely on density, atomic number information, and lack of understanding of "object" information. Summary of the invention
  • a method and device for creating a three-dimensional model are proposed.
  • a method of creating a three-dimensional model of an object in an inspected baggage in a CT imaging system comprising the steps of: acquiring tomographic data of the checked baggage using the CT imaging system; Performing interpolation to generate three-dimensional volume data of the checked baggage; unsupervised segmentation of the three-dimensional volume data of the checked baggage to obtain a plurality of divided regions; performing isosurface extraction on the plurality of divided regions, Obtaining a corresponding isosurface; performing three-dimensional surface segmentation on the isosurface to form a three-dimensional model of each object in the checked baggage.
  • an apparatus for creating a three-dimensional model of an object in an inspected baggage in a CT imaging system comprising: means for acquiring tomographic data of an inspected baggage using the CT imaging system; Interpolating the fault data to generate three-dimensional volume data of the checked baggage; performing unsupervised segmentation on the three-dimensional volume data of the checked baggage to obtain a plurality of divided regions; The region performs isosurface extraction to obtain a corresponding isosurface device; the isosurface is subjected to three-dimensional surface segmentation to form a three-dimensional model of each object in the checked baggage.
  • FIG. 1 is a schematic structural view of a CT system according to an embodiment of the present invention.
  • FIG. 2 is a schematic structural diagram of a computer data processor according to an embodiment of the present invention.
  • FIG. 3 is a schematic structural diagram of a controller according to an embodiment of the present invention.
  • FIG. 4 is a schematic diagram for describing depth projection from three perspectives of View1, View2, and View3;
  • FIG. 5 is a flowchart describing a method for inspecting an object to be inspected in a CT imaging system;
  • FIG. 6 is a view for describing another method according to the present invention. A flowchart of a method of displaying an object in a CT imaging system of an embodiment;
  • FIG. 7 is a flow chart depicting a method of creating a three-dimensional model of an object in a baggage in a CT imaging system in accordance with another embodiment of the present invention. detailed description
  • embodiments of the present invention propose a method of inspecting baggage in a CT imaging system. After the CT imaging system acquires the tomographic data of the checked baggage, three-dimensional volume data of at least one object in the checked baggage is generated from the tomographic data of the checked baggage. And calculating, according to the three-dimensional volume data, the first depth projection image, the second depth projection image, and the third depth projection image of the object in three directions, wherein the projection direction of the third depth projection image and the first and second depth projection images The projection direction is orthogonal.
  • the respective symmetry metrics of the first, second, and third depth projection images, the similarity metric values between the two, and the duty ratio and aspect ratio are calculated.
  • a shape characteristic parameter of the object to be inspected is generated based on at least a symmetry metric of each of the first to third depth projection images, a similarity measure between the two and a duty ratio and an aspect ratio.
  • the shape feature parameters are classified by a classifier based on shape feature parameters to obtain a quantifier description embodying the shape of the object.
  • the output includes at least a semantic description of the quantifier description of the object.
  • the shape characteristics of the object are obtained by processing the data of the object obtained by the CT imaging system, and outputted in the manner of semantic description, so that the inspector can intuitively and accurately obtain the specific description of the object in the checked baggage, thereby Reduced the rate of missed detection.
  • a method of displaying an object in a CT imaging system After acquiring the tomographic data of the checked baggage using the CT imaging system, three-dimensional volume data of each object in the checked baggage is generated from the tomographic data. Then, for each object, a semantic description including at least the quantifier description of the object is determined based on the three-dimensional volume data. Receiving a user's selection of an object, while displaying a three-dimensional image of the selected object, presenting a semantic description of the object.
  • an embodiment of the present invention in order to more accurately extract the shape feature of an object in the checked baggage, provides a method of creating a three-dimensional model of an object in a checked baggage in a CT imaging system .
  • the tomographic data is interpolated to generate three-dimensional volume data of the checked baggage.
  • the three-dimensional volume data of the checked baggage is unsupervised and divided to obtain a plurality of divided regions, and then the equal-surface extraction is performed on the plurality of divided regions to obtain corresponding isosurfaces.
  • the isosurface is then subjected to three-dimensional surface segmentation to form a three-dimensional model of each object.
  • Such a three-dimensional model of the object in the checked baggage can more accurately describe the three-dimensional surface of the object in the checked baggage As described, it provides a good basis for the subsequent extraction of three-dimensional shape features, and thus can improve the accuracy of security inspection.
  • a CT apparatus includes: a chassis 20, a carrier mechanism 40, a controller 50, a computer data processor 60, and the like.
  • the gantry 20 includes a source 10 for emitting X-rays for inspection, such as an X-ray machine, and a detection and acquisition device 30.
  • the carrier mechanism 40 carries the scanned area between the source 10 of the inspected baggage 70 passing through the frame 20 and the detecting and collecting device 30, while the frame 20 is rotated about the direction of advancement of the inspected baggage 70, thereby being emitted by the source 10
  • the radiation can pass through the inspected baggage 70 to perform a CT scan of the inspected baggage 70.
  • the detecting and collecting device 30 is, for example, a detector having an integral module structure and a data collector, such as a flat panel detector, for detecting rays transmitted through the liquid object to be inspected, obtaining an analog signal, and converting the analog signal into a digital signal, thereby outputting
  • the checked baggage 70 is directed to X-ray projection data.
  • the controller 50 is used to control the synchronization of various parts of the entire system.
  • the computer data processor 60 is used to process the data collected by the data collector, process and reconstruct the data, and output the results.
  • the radiation source 10 is placed on the side where the object to be inspected can be placed, and the detecting and collecting device 30 is placed on the other side of the checked baggage 70, including a detector and a data collector for acquiring the checked baggage.
  • the data collector includes a data amplification forming circuit that operates in either (current) integration mode or pulse (count) mode.
  • the data output cable of the detection and acquisition device 30 is coupled to the controller 50 and the computer data processor 60, and the acquired data is stored in the computer data processor 60 in accordance with a trigger command.
  • Fig. 2 is a block diagram showing the structure of a computer data processor 60 as shown in Fig. 1.
  • the data collected by the data collector is stored in the memory 61 through the interface unit 68 and the bus 64.
  • the read-only memory (ROM) 62 stores configuration information and programs of the computer data processor.
  • a random access memory (RAM) 63 is used to temporarily store various data during the operation of the processor 66.
  • a computer program for performing data processing is also stored in the memory 61.
  • the internal bus 64 is connected to the above-described memory 61, read only memory 62, random access memory 63, input device 65, processor 66, display device 67, and interface unit 68.
  • the instruction code of the computer program instructs the processor 66 to execute a predetermined data processing algorithm, and after obtaining the data processing result, displays it on, for example, an LCD display.
  • the processing result is outputted on the display device 67 of the class, or directly in the form of a hard copy such as printing.
  • FIG. 3 shows a block diagram of a controller in accordance with an embodiment of the present invention.
  • the controller 50 includes: a control unit 51 that controls the radiation source 10, the carrier mechanism 40, and the probe according to an instruction from the computer 60. a detection and acquisition device 30; a trigger signal generating unit 52 for generating a trigger command for triggering the action of the radiation source 10, the detecting and collecting device 30, and the carrying mechanism 40 under the control of the control unit; the first driving device 53, which The instructed carrier 40 transmits the inspected baggage 70 according to a trigger command generated by the trigger signal generating unit 52 under the control of the control unit 51; the second driving device 54 is generated according to the trigger signal generating unit 52 under the control of the control unit 51.
  • the trigger command frame 20 is rotated.
  • the projection data obtained by the detecting and collecting device 30 is stored in the computer 60 for CT tomographic image reconstruction, thereby obtaining tomographic image data of the checked baggage 70.
  • the computer 60 then extracts the three-dimensional shape parameters of at least one of the objects in the checked baggage 70 from the tomographic image data, for example, by executing software, thereby performing a security check.
  • the CT imaging system described above may also be a dual energy CT system, that is, the X-ray source 10 of the gantry 20 is capable of emitting both high and low energy rays, and the detection and acquisition device 30 detects projections at different energy levels.
  • the dual-energy CT reconstruction is performed by the computer data processor 60 to obtain the equivalent atomic number and equivalent electron density data of the respective faults of the checked baggage 70.
  • FIG. 4 shows a schematic diagram of definitions of various perspectives in a method in accordance with an embodiment of the present invention.
  • Figure 5 is a flow chart depicting a method for inspecting luggage in a CT imaging system.
  • the CT imaging system acquires the tomographic data of the checked baggage.
  • the CT imaging system acquires the tomographic data of the checked baggage.
  • the tomographic data generally includes tomographic density map data and tomographic atomic number map data.
  • linear attenuation coefficient image data is obtained.
  • step S52 three-dimensional volume data of at least one object in the checked baggage is generated from the tomographic data.
  • inter-layer interpolation of the tomographic data is performed to obtain three-dimensional volume data of the checked baggage.
  • three-dimensional interpolation is performed on the two, so that the resolution of the images is consistent within the faults and between the faults.
  • algorithms for 3D interpolation such as the commercial software Intel IPP (Intel Integrated Performance Primitives) function library, open source software Kitware VTK (Visualization Toolkit) function library, all provide this function.
  • the two-dimensional tomographic data is converted into three-dimensional volume data.
  • a first depth projection image, a second depth projection image, and a third depth projection image of the object in three directions are calculated based on the three-dimensional volume data, wherein a projection direction of the third depth projection image is first It is orthogonal to the projection direction of the second depth projection image.
  • the projection directions of the first depth projection image and the second depth projection image are as orthogonal as possible (e.g., substantially orthogonal), respectively approaching the largest and smallest directions of the projected area of the object.
  • Deep projection Depth Buffer also known as Z-Buffering
  • Z-Buffering is the basic technology for 3D surface display. Judge object The occlusion relationship between them, and the unoccluded portion is displayed on the screen. This technology is a typical technology used in 3DOR, but it often involves dozens of projections with high complexity. As shown in FIG. 4, in the present embodiment, only three depth projection maps are used.
  • the first projection is defined for the purpose of getting the "main view" where it is approximated by the largest area projection.
  • the projection shown in the XOY plane in Figure 2 is ⁇
  • the second projection is defined as 1 ⁇ 2, the purpose of which is to obtain a "top view” where the projection with the smallest area is approached.
  • the projection shown in the XOZ plane in Figure 2 is ⁇ .
  • the two projection directions in Fig. 2 are orthogonal, but this condition is not necessarily satisfied, so the angle formed by the two projection directions is also one of the features.
  • the third projection is defined for the purpose of getting a "side view”. After the projection direction of ⁇ is obtained, the projection is orthogonal to their projection direction, and the projection on the YOZ plane in FIG. 4 is obtained.
  • Rectilinearity algorithm is used to quickly estimate the first two projection directions using a genetic algorithm.
  • the symmetry of the image can reflect the self-similarity of the object and is an important shape feature.
  • the PCA Principal Component Analyze
  • the two-dimensional image is maximized in the X-axis divergence, that is, the upper and lower symmetry of the image is the strongest.
  • the following refers to the image that has been aligned.
  • step S54 the respective symmetry metrics of the first depth projection image, the second depth projection image, and the third depth projection image, the similarity metric values between the two and the duty ratio and the aspect ratio are calculated.
  • the symmetry, the two-two similarity, the duty cycle, the aspect ratio of ⁇ are extracted as one or more of the shape features or any combination thereof as the shape feature parameter.
  • the angle between the above-mentioned '1 ⁇ 2 projection directions also reflects the shape of the object, and also serves as one of the characteristic parameters.
  • the volume of the object reflects the size of the object and is also one of the features.
  • the method for obtaining the above features may be:
  • a shape characteristic of the object in the checked baggage is generated based on at least a symmetry metric value of each of the first to third depth projection images, a similarity metric value between the two and a duty ratio and an aspect ratio. parameter. For example, in some embodiments, one or more of the symmetry values, the similarity values, the duty cycle and the aspect ratio, and the angle between the projection and the two directions calculated from i) to iv) above. To form shape feature parameters.
  • the shape feature parameters are classified using a classifier based on shape feature parameters to obtain a quantifier description embodying the shape of the object.
  • the process of constructing the classifier conforms to the general process of pattern recognition, and various types of classifiers can be used, such as linear classifiers, support vector machines, decision trees, neural networks, integrated classifiers, and the like.
  • various types of classifiers can be used, such as linear classifiers, support vector machines, decision trees, neural networks, integrated classifiers, and the like.
  • the shape classification and recognition of unknown targets can be realized.
  • the embodiment implements a classifier using RF (Random Forest).
  • RF Random Forest
  • Many well-known function libraries contain RF algorithms, such as the open source tool Opencv, which has implementation functions, and the process is not described here.
  • the object obtained in the fifth step needs to be labeled as "bag, piece, block, bottle, can, tube, root, bag, box, one" by manual judgment. .
  • Bag, piece, block, bottle, can, tube, root, bag, box, one by manual judgment.
  • “Bag” refers to flat packaging, and the aspect ratio is an important feature. Such as soft milk, flat bags of uniform food items, etc.;
  • Porture refers to an object with a very low thickness, and the aspect ratio is an important feature. Thin books, luggage fillings, and knives are all in this range;
  • Block refers to objects with low similarity and low duty cycle, such as homogeneous objects in plastic bags, which will form “blocks” if they are not flat-packed;
  • Bottom refers to items similar to mineral water bottles. Main and side projection similarity, duty cycle, and aspect ratio are all important features;
  • Can refers to a similar canned item, similar to a bottle, but with a higher duty cycle and aspect ratio
  • Root refers to a long object, and its aspect ratio is an important feature. Such as sausage, wood, iron pipe, are in this range;
  • “Tube” is an object that is shorter than “root” and has good symmetry, such as facial cleanser, glue, etc.; "box” refers to a rectangular item with a certain thickness, which is larger than the “block” duty cycle. Duty cycle, aspect ratio is its main feature, such as soap, many cosmetics, food have similar characteristics;
  • Bog refers to a larger object, and volume is an important feature. Such as computers, large and thick books, large objects that can be judged to be other shapes are not such;
  • the cup shown in Figure 4 may be “can” with the above predicate.
  • This definition is for the specific application of security inspection, such as solid explosives, generally in the form of “bags”, “pieces”, “blocks”, liquids are mainly “bottles”, “cans", “tubes”, and There are many forms of “pieces” and “roots” for control devices. Others such as “package”, “box”, and “piece” are complementary to common shapes.
  • a semantic description of the quantifier description including at least the object is output.
  • the user can interact with the user in a variety of ways. For example, the outline of the object is directly displayed in the result to remind the user to pay attention; or the object is extracted on the screen and the description information is displayed when the user clicks, so that the user can further understand and mark the object.
  • limiting the semantics of objects and highlighting objects that satisfy specific semantics can reduce the labor intensity of the inspectors and improve work efficiency.
  • the position and weight of the object have been known, and the shape of the object has been known. Therefore, it is only necessary to calculate the average atomic number and electron density (or the linear attenuation coefficient in the case of single-energy CT) to complete the description. After each The predicate information has been obtained, and it is sorted to obtain the semantic description of the object, ie "shape + weight + density + atomic number + position"
  • FIG. 6 is a flow chart depicting a method of displaying an object in a CT imaging system in accordance with another embodiment of the present invention.
  • automatic detection of an object in the checked baggage is achieved by automatic detection and description.
  • the object description result is a necessary supplement to the manual detection, and is also a means to strengthen human-computer interaction. It has a strong application value in reducing the important problem of missed detection.
  • the CT imaging system is used to acquire the tomographic data of the checked baggage.
  • step S62 three-dimensional volume data of each object in the checked baggage is generated from the tomographic data. Then, in step S63, for each object, a semantic description of the quantifier description including at least the object is determined based on the three-dimensional volume data.
  • step S64 the user's selection of an object is received, and while the three-dimensional image of the selected object is displayed, a semantic description of the object is presented.
  • all the detected object positions are marked in the display window, and when the inspector uses a tool such as a mouse to select a position within a certain object range, the complete semantic description of the object is displayed.
  • the inspector can use a mouse and other tools to select objects, and then select further objects in detail to add semantic description content.
  • the semantic description can show only objects that meet the qualification. For example, an object with a shape of "bottle" and a weight of 200 g or more is limited, and the position of the suspect object can be represented in a two-dimensional, three-dimensional image, and the auxiliary inspector performs the judgment.
  • objects can be highlighted when they are selected, and all other content can be masked, showing only the contents of the object.
  • some of the limitations of the above embodiments may be enhanced, such as reinforcement volume threshold definition, object shape limitation, and may be used as an automatic detection of specific objects such as explosives, contraband.
  • a three-dimensional model of an object in the checked baggage may also be created before the depth projection is performed to further extract the shape features and perform a security check.
  • Figure 7 is a flow chart depicting a method of creating a three-dimensional model of an object in a checked baggage in a CT imaging system in accordance with another embodiment of the present invention.
  • the CT data system is used to acquire the tomographic data of the checked baggage.
  • the tomographic data is interpolated to generate three-dimensional volume data of the checked baggage.
  • the three-dimensional volume data includes density volume data and atomic volume data.
  • the three-dimensional volume data includes a linear attenuation coefficient.
  • the two should be separately Three-dimensional interpolation makes the image within the fault and the resolution between the faults consistent.
  • 3D interpolation There are many well-known algorithms for 3D interpolation, such as the commercial Intel IPP (Intel Integrated Performance Primitives) function library and the open source software Kitware VTK (Visualization Toolkit) function library, which provide this function.
  • the two-dimensional tomographic data is converted into three-dimensional volume data.
  • the "volume data" includes density volume data and atomic volume data.
  • this step can be omitted, but the amount of calculation will increase, and the number of "object" results will be large, resulting in poor usability.
  • volume data filtering is performed using a three-dimensional bilateral filter.
  • the embodiment uses a fast algorithm and can also be implemented using the ITK (Insight Segmentation and Registration Toolkit) function library.
  • step S73 unsupervised segmentation of the three-dimensional volume data is performed to obtain a plurality of divided regions.
  • segmentation is often based on 4/8 neighborhoods, gradient information, etc.
  • the correlation process needs to be extended to three dimensions, for example, the 4 neighborhood is expanded to 6 neighborhoods.
  • the segmentation involves Density, yard ordinal two-part volume data, can be weighted by the two, or each voxel is represented by a two-dimensional vector to obtain a unified segmentation result; further, the segmentation needs to achieve the effect of under-segmentation.
  • SRM Statistical Region Merging
  • step S74 isosurface extraction is performed on a plurality of divided regions to obtain corresponding isosurfaces.
  • the isosurface extraction is performed on each of the divided regions obtained in step S73, and the corresponding isosurfaces are obtained.
  • the Marching Cubes algorithm is used in the embodiment.
  • step S75 the isosurface is subjected to three-dimensional surface division to form a three-dimensional model of each object.
  • step S73 results in an under-segmentation result, a plurality of objects having similar material characteristics and closely connected positions cannot be separated. Therefore, it is necessary to refine the three-dimensional segmentation results by surface segmentation.
  • a mesh segmentation algorithm can be used. Split the surface into multiple convex surfaces.
  • the algorithm is a supervised algorithm and needs to be specified. Divide the number of results. In practical applications, the number of partitions can be calculated or iteratively obtained by an unsupervised improvement idea similar to K-Means clustering. However, after our experiments, similar methods are difficult to get good results, so we set the number of divisions to 10.
  • the surface Merg e is performed here with reference to the star-convex hypothesis.
  • the center of the segmentation result surface A is a
  • the center of the surface B is b. If the line connecting ⁇ b is inside the entire surface obtained in the third step (or the proportion of external voxels is less than a certain threshold), then A, B is connected. This connection process is performed on the two segmentation results, and the final segmentation result can be obtained.
  • the segmented result can also be processed, including three steps of filling, smoothing, and pattern defining.
  • the first two steps are the basic operations in graphics, and can be implemented using the open source software Kitware VTK (Visualization Toolkit) function library, which will not be described here.
  • Kitware VTK Visualization Toolkit
  • Model qualification refers to the removal of smaller objects, including objects of smaller size, volume, weight, etc. There are two reasons for limiting. First, the noise object is removed, making the result more practical. Second, the local details of many objects are omitted, making the shape recognition of the next step more accurate.
  • the threshold is specifically defined, which is related to the resolution of DECT. It can be set according to the actual situation, for example, the weight can be set to 50 grams.
  • aspects of the embodiments disclosed herein may be implemented in an integrated circuit as a whole or in part, as one or more of one or more computers running on one or more computers.
  • a computer program eg, implemented as one or more programs running on one or more computer systems
  • implemented as one or more programs running on one or more processors eg, implemented as one or One or more programs running on a plurality of microprocessors, implemented as firmware, or substantially in any combination of the above, and those skilled in the art, in accordance with the present disclosure, will be provided with design circuitry and/or write software and / or firmware code capabilities.
  • signal bearing media include, but are not limited to: recordable media such as floppy disks, hard drives, compact disks (CDs), digital versatile disks (DVD), digital tape, computer memory, etc.; and transmission type media, such as digital and/or analog communication media

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • High Energy & Nuclear Physics (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geophysics (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Image Processing (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

公开了一种创建物体的三维模型的方法和设备。该方法包括步骤：利用CT成像系统获取被检查行李的断层数据；对断层数据进行插值，生成被检查行李的三维体数据；对被检查行李的三维体数据进行无监督分割，得到多个分割的区域；对多个分割的区域进行等值面提取，得到相应的等值面；对等值面进行三维表面分割，形成被检查行李中各个物体的三维模型。利用上述方案，能够较为准确地建立被检查行李中物体的三维模型，从而为后续的形状特征提取和安全检查提供较好的基础，从而减少漏检。

Description

三维模型创建方法和设备 技术领域
本发明涉及一种对物体的安全检查， 具体涉及三维模型创建方法和设备。 背景技术
透视成像是安检领域的重要手段， 典型应用过程为： 设备对行李物品进行透视 扫描并成像， 图检员经过判图， 手工标注嫌疑物区域， 并对区域做出语义描述， 如 "打 火机" 、 "一瓶酒"等等。 这一过程过分依赖人的因素， 在真正危险物以极小概率出 现、 图检员经验有限、 疲劳等因素影响下， 会导致漏检， 造成严重后果。
解决这一问题的典型手段以自动辅助检测为主， 以增加设备与安检员交互为辅。 自动检测虽然为主要手段， 但效果不尽人意。 其典型技术如爆炸物检测、 高密度报警 等不能充分满足应用需求。究其原因， 一方面是技术条件所限， 如 DEDR (Dual Energy Digital Radiography) 中的透视重叠造成物体混叠， 另一方面也是学界的研究较少， 较 新的技术如 DECT (Dual Energy Computed Tomography) 需要新的检测算法支持所致。
DECT是解决这一问题的优势技术， 它由 DR、 CT技术发展而来， 在获取扫描 物三维结构信息的同时， 可以得到物体内部的有效原子序数与等效电子密度， 因此具 备了进行三维数据理解的条件。 但现有的研究目标往往是某特定物体的目标检测， 且 主要依赖密度、 原子序数信息， 缺乏对 "物体"信息的认识。 发明内容
为了更为准确地对被检查物体进行安全检查。 提出了三维模型创建方法和设备。 在本发明的一个方面，提供了一种在 CT成像系统中创建被检查行李中物体的三 维模型的方法， 包括步骤： 利用所述 CT成像系统获取被检查行李的断层数据； 对所述 断层数据进行插值， 生成所述被检查行李的三维体数据； 对所述被检查行李的三维体 数据进行无监督分割， 得到多个分割的区域； 对所述多个分割的区域进行等值面提取， 得到相应的等值面； 对所述等值面进行三维表面分割， 形成所述被检查行李中各个物 体的三维模型。
在本发明的另一方面，提供了一种在 CT成像系统中创建被检查行李中物体的三 维模型的设备， 包括： 利用所述 CT成像系统获取被检查行李的断层数据的装置； 对所 述断层数据进行插值， 生成所述被检查行李的三维体数据的装置； 对所述被检查行李 的三维体数据进行无监督分割， 得到多个分割的区域的装置； 对所述多个分割的区域 进行等值面提取， 得到相应的等值面的装置； 对所述等值面进行三维表面分割， 以形 成所述被检查行李中各个物体的三维模型的装置。
利用上述方案， 能够较为准确地建立被检查行李中物体的三维模型， 从而为后 续的形状特征提取和安全检查提供较好的基础， 从而减少漏检。 附图说明
下面的附图表明了本技术的实施方式。 这些附图和实施方式以非限制性、 非穷 举性的方式提供了本技术的一些实施例， 其中：
图 1是根据本发明实施例的 CT系统的结构示意图；
图 2是根据本发明实施例的计算机数据处理器的结构示意图；
图 3是根据本发明实施例的控制器的结构示意图；
图 4是描述从 Viewl、 View2、 View3三个视角做深度投影的示意图； 图 5是描述根据在 CT成像系统中对被检查物体进行检查的方法的流程图； 图 6是描述根据本发明另一实施例的在 CT成像系统中显示物体的方法的流程 图；
图 7是描述根据本发明另一实施例的在 CT成像系统中创建行李中物体的三维模 型的方法的流程图。 具体实施方式
下面将详细描述本发明的具体实施例， 应当注意， 这里描述的实施例只用于举 例说明， 并不用于限制本发明。 在以下描述中， 为了提供对本发明的透彻理解， 阐述 了大量特定细节。 然而， 对于本领域普通技术人员显而易见的是： 不必采用这些特定 细节来实行本发明。 在其他实例中， 为了避免混淆本发明， 未具体描述公知的结构、 材料或方法。
在整个说明书中， 对 "一个实施例" 、 "实施例" 、 "一个示例"或 "示例" 的提及意味着： 结合该实施例或示例描述的特定特征、 结构或特性被包含在本发明至 少一个实施例中。 因此， 在整个说明书的各个地方出现的短语 "在一个实施例中" 、 "在实施例中" 、 "一个示例"或 "示例"不一定都指同一实施例或示例。 此外， 可 以以任何适当的组合和 /或子组合将特定的特征、 结构或特性组合在一个或多个实施例 或示例中。 此外， 本领域普通技术人员应当理解， 这里使用的术语 "和 /或"包括一个 或多个相关列出的项目的任何和所有组合。
针对现有技术中的仅仅使用被检查行李中物体的物理属性信息进行安全检查的 不足之处，本发明的实施例提出了一种在 CT成像系统中对行李进行检查的方法。利用 CT成像系统获取被检查行李的断层数据后， 从被检查行李的断层数据生成被检查行李 中至少一个物体的三维体数据。 然后基于三维体数据计算物体在三个方向上的第一深 度投影图像、 第二深度投影图像和第三深度投影图像， 其中第三深度投影图像的投影 方向与第一和第二深度投影图像的投影方向正交。 接下来， 计算第一、 第二和第三深 度投影图像各自的对称性度量值、 两两之间的相似性度量值以及占空比和高宽比。 至 少基于第一到第三深度投影图像各自的对称性度量值、 两两之间的相似性度量值以及 占空比和高宽比来生成所述被检查物体的形状特征参数。 利用基于形状特征参数的分 类器对所述形状特征参数进行分类， 得到体现所述物体的形状的量词描述。 输出至少 包括所述物体的所述量词描述的语义描述。这样，通过对 CT成像系统获得的物体的数 据进行处理得到了物体的形状特征， 并且以其语义描述的方式输出， 使得图检员能够 直观、 准确地得到被检查行李中物体的具体描述， 从而降低了漏检率。
根据本发明的另一实施例， 为了降低漏检率， 提供了一种在 CT成像系统中显示 物体的方法。在利用 CT成像系统获取被检查行李的断层数据后， 从断层数据生成被检 查行李中各个物体的三维体数据。 然后， 针对每个物体， 基于三维体数据确定至少包 括所述物体的量词描述的语义描述。 接收用户对某个物体的选择， 在显示所选择物体 的三维图像的同时， 呈现该物体的语义描述。这样， 当被检查行李通过 CT成像设备进 行检查时， 不但在屏幕上输出被检查行李中物体的图像， 而且针对图检员所选择的物 体输出其语义描述， 从而直观地呈现物体的描述， 降低了漏检率。
根据本发明的另一实施例， 为了能够更为准确地提取被检查行李中物体的形状 特征，本发明的实施例提供了一种在 CT成像系统中创建被检查行李中物体的三维模型 的方法。利用该 CT成像系统获取被检查行李的断层数据后，对所述断层数据进行插值， 生成所述被检查行李的三维体数据。 然后对所述被检查行李的三维体数据进行无监督 分割， 得到多个分割的区域， 接下来对多个分割的区域进行等值面提取， 得到相应的 等值面。 然后对等值面进行三维表面分割， 形成各个物体的三维模型。 这样的得到的 被检查行李中物体的三维模型能够较为准确地对被检查行李中物体的三维表面进行描 述， 为后续的三维形状特征的提取提供了较好的基础， 因此能够提高安全检查的准确 率。
图 1是根据本发明实施方式的 CT设备的结构示意图。如图 1所示， 根据本实施 方式的 CT设备包括： 机架 20、 承载机构 40、 控制器 50、 计算机数据处理器 60等。 机架 20包括发出检查用 X射线的射线源 10， 诸如 X光机， 以及探测和采集装置 30。 承载机构 40承载被检查行李 70穿过机架 20的射线源 10与探测和采集装置 30之间的 扫描区域， 同时机架 20围绕被检查行李 70的前进方向转动， 从而由射线源 10发出的 射线能够透过被检查行李 70， 对被检查行李 70进行 CT扫描。 探测和采集装置 30例 如是具有整体模块结构的探测器及数据采集器， 例如平板探测器， 用于探测透射被检 液态物品的射线， 获得模拟信号， 并且将模拟信号转换成数字信号， 从而输出被检查 行李 70针对 X射线的投影数据。 控制器 50用于控制整个系统的各个部分同步工作。 计算机数据处理器 60用来处理由数据采集器采集的数据， 对数据进行处理并重建， 输 出结果。
如图 1所示， 射线源 10置于可放置被检物体的一侧， 探测和采集装置 30置于 被检查行李 70的另一侧， 包括探测器和数据采集器， 用于获取被检查行李 70的透射 数据和 /或多角度投影数据。数据采集器中包括数据放大成形电路， 它可工作于（电流) 积分方式或脉冲 （计数） 方式。 探测和采集装置 30的数据输出电缆与控制器 50和计 算机数据处理器 60连接， 根据触发命令将采集的数据存储在计算机数据处理器 60中。
图 2示出了如图 1所示的计算机数据处理器 60的结构框图。 如图 2所示， 数据 采集器所采集的数据通过接口单元 68 和总线 64 存储在存储器 61 中。 只读存储器 (ROM) 62中存储有计算机数据处理器的配置信息以及程序。随机存取存储器（RAM) 63用于在处理器 66工作过程中暂存各种数据。 另外， 存储器 61中还存储有用于进行 数据处理的计算机程序。 内部总线 64连接上述的存储器 61、 只读存储器 62、 随机存 取存储器 63、 输入装置 65、 处理器 66、 显示装置 67和接口单元 68。
在用户通过诸如键盘和鼠标之类的输入装置 65输入的操作命令后， 计算机程序 的指令代码命令处理器 66执行预定的数据处理算法， 在得到数据处理结果之后， 将其 显示在诸如 LCD显示器之类的显示装置 67上， 或者直接以诸如打印之类硬拷贝的形 式输出处理结果。
图 3示出了根据本发明实施方式的控制器的结构框图。 如图 3所示， 控制器 50 包括： 控制单元 51， 根据来自计算机 60的指令， 来控制射线源 10、 承载机构 40和探 测和采集装置 30; 触发信号产生单元 52， 用于在控制单元的控制下产生用来触发射线 源 10、 探测和采集装置 30以及承载机构 40的动作的触发命令； 第一驱动设备 53， 它 在根据触发信号产生单元 52在控制单元 51的控制下产生的触发命令驱动承载机构 40 传送被检查行李 70; 第二驱动设备 54， 它根据触发信号产生单元 52在控制单元 51的 控制下产生的触发命令机架 20旋转。
探测和采集装置 30获得的投影数据存储在计算机 60中进行 CT断层图像重建， 从而获得被检查行李 70的断层图像数据。 然后计算机 60例如通过执行软件来从断层 图像数据提取被检查行李 70中至少一个物体的三维形状参数， 进而进行安全检查。 根 据其他实施例， 上述的 CT成像系统也可以是双能 CT系统， 也就是机架 20的 X射线 源 10能够发出高能和低能两种射线， 探测和采集装置 30探测到不同能量水平下的投 影数据后， 由计算机数据处理器 60进行双能 CT重建，得到被检查行李 70的各个断层 的等效原子序数和等效电子密度数据。
图 4示出了根据本发明实施例的方法中各个视角的定义的示意图。 图 5是描述 根据在 CT成像系统中对行李进行检查的方法的流程图。 在步骤 S51， 利用 CT成像系 统获取被检查行李的断层数据。例如基于上述的 CT设备或者其他的 CT设备对被检查 行李进行双能 CT检查，得到断层数据，这里的断层数据通常包括断层密度图数据和断 层原子序数图数据。 但是， 在其他的实施例中， 例如单能 CT的情况下， 得到的是线性 衰减系数图像数据。
在步骤 S52， 从断层数据生成被检查行李中至少一个物体的三维体数据。 例如， 对断层数据进行层间插值， 从而得到被检查行李的三维体数据。 再如， 在得到一系列 连续断层下的 DECT密度图、 原子密度图后， 要对两者分别进行三维插值， 使得图像 在断层内、断层间的分辨率一致。三维插值的公知算法较多，如商业将软件 Intel IPP(Intel Integrated Performance Primitives)函数库， 开源软件 Kitware VTK(Visualization Toolkit) 函数库， 均提供这一功能。 插值之后， 二维断层数据转换为三维体数据。
在步骤 S53，基于所述三维体数据计算所述物体在三个方向上的第一深度投影图 像、 第二深度投影图像和第三深度投影图像， 其中第三深度投影图像的投影方向与第 一和第二深度投影图像的投影方向正交。 根据另一实施例， 第一深度投影图像与第二 深度投影图像的投影方向尽量正交（例如大致正交）， 分别逼近该物体投影面积最大的 和最小的方向。
深度投影 Depth Buffer也叫 Z-Buffering， 是三维表面显示的基本技术。判断物体 之间的遮挡关系， 并将没有遮挡的部分显示到屏幕上。 这一技术是目前 3DOR所使用 的典型技术， 但往往涉及数十个投影， 复杂度较高。 如图 4所示， 在本实施例中， 仅 使用 3个深度投影图。 第一个投影定义为 其目的是为了得到 "主视图" ， 此处以面 积最大的投影来逼近它。 如图二中 XOY平面所示投影即为 Ί 第二个投影定义为 ½， 其目的是为了得到 "俯视图" ， 此处以面积最小的投影来逼近。 如图二中 XOZ平面所 示投影即为 ^。 图二中这两个投影方向正交， 但实际上未必满足这一条件， 因此这两个 投影方向所成的角度也作为特征之一。第三个投影定义为 其目的是为了得到 "侧视 图" 。 在得到 ^的投影方向后， 以正交于他们的投影方向再次投影， 即可得到 如图 4中 YOZ面上的投影即为^。
需要注意的是， 图 4中 X、 Y、 Ζ方向可以正向、 反向投影出 6个图像。 由于在 三维表面分割过程中， 细节已经去掉， 所以正反两向投影很相似。 为降低算法复杂度， 这里只使用 3个投影。
为得到面积最大或最小投影， 可以采用遍历所有旋转角度的方法来实现， 但这 样复杂度过高。 此处借鉴 Rectilinearity算法， 使用遗传算法快速估计前两个投影方向。
''ι〜 图像的对称性可以反映物体的自相似性，是重要的形状特征。为了便于计算， 此处将 i ~ 进行 PCA(Principal Component Analyze, 主成分分析)对齐， 使得二维图像在 X轴散度最大化， 也即图像的上下对称性最强。 下文所指^〜½指经过对齐处理的图像。
在步骤 S54， 计算第一深度投影图像、第二深度投影图像和第三深度投影图像各 自的对称性度量值、 两两之间的相似性度量值以及占空比和高宽比。
在一些实施例中， 提取 ^〜^的对称性、 两两相似性、 占空比、 宽高比作为形状特 征中的一个或多个或者他们的任意组合， 作为形状特征参数。 另外， 前述' ½投影方 向之间夹角也反映了物体形状， 也作为特征参量之一。 还有， 物体体积反映物体大小， 也作为特征之一。
设深度投影图像的灰度取值范围在 [0,1]间。 其中灰度值 0代表无限远， 非 0值 表示面片与观测的距离， 越近值越大。 上述特征的获取方法可以为：
i)求 的上下对称性 fr t。 设『《〜^上下翻转得到图像 〜 ， 则可 定义上下对称性 ^为：
i = Σ1 - ίι Ί/Zft > 0) ? 0) (1) 即以图像上下翻转后与原图像的平均灰度差异为标准求取上下对称性。
ii)求 «、 、 ¼勺占空比 与高宽比 深度投影图像大小是视口所定义， 不反映物体属性。 求对齐后的深度投影高宽 比、 占空比， 能较好的描述物体宏观特性。 求 的外包围矩形框， 易求其高宽比 Η。 再 在其中统计非 0像素个数， 除以包围框面积， 则得到其占空比 D。 iii) 求 、 ―、 ½之间两两相似性 。 在 ii)中已得』《、 的包围框， 将包围框中的图像裁出， 得到 ^、 将 缩放到 和 ^一样大小， 得到 '。 将 上下翻转得到^ ^。 此时可得 为：
f.B = ax(j∑llf - Jf I ∑(i > C«i/f > 0)),i∑I/f - J#"1/∑(/f > 0|Jf ' > 0))) (2) 相似性求法类似与式 (3)中的对称性，不同处在于图像经过了大小归一化。且取/ 为 ( I )与 相似性值中较大的一个。同理，可得 之间的相似性 / ，还有^ 之间的相似性 。 iv) 在深度投影过程中， 已知投影 、 ½的两个方向， 求其夹角《作 为特征。 模型体积 反映物体的大小， 也作为特征。
将上述步骤所述的各个形状特征参量组合， 形成 14维形状特征矢量'
= if^f J^f f J^f ifJ J^Uf V} (3)
在步骤 S55， 至少基于第一到第三深度投影图像各自的对称性度量值、两两之间 的相似性度量值以及占空比和高宽比来生成所述被检查行李中物体的形状特征参数。 例如， 在一些实施例中， 基于上述从 i) 到 iv ) 中计算的对称性值、 相似性值、 占空 比和高宽比以及投影 和½两个方向的夹角中的一个或者多个来形成形状特征参数。
在步骤 S56， 利用基于形状特征参数的分类器对所述形状特征参数进行分类， 得 到体现所述物体的形状的量词描述。
在得到特征矢量 后， 构造分类器的过程符合模式识别一般过程， 可以使用各种 不同类型的分类器， 如线性分类器、 支持向量机、 决策树、 神经网络、 集成分类器等 等。经过训练后，即可实现对未知目标的形状分类识别。实施例采用 RF(Random Forest, 随机森林)实现分类器。许多公知函数库包含 RF算法， 如开源工具 Opencv即有实现函 数， 此处不再赘述其过程。
需要说明的是， 在训练集上， 通过人工判图， 需要将第五步中得到的物体标注 为 "袋、 片、 块、 瓶、 罐、 管、 根、 包、 盒、 个"其中之一。 这里简单说明这几个谓 词常元的区别。 "袋"是指扁包装， 高宽比是其重要特征。 比如软包牛奶、 扁袋装匀质食品类 物品等；
"片"是指厚度很低的物体， 高宽比是其重要特征。 薄书、 箱包填充物、 刀具 均在此范围；
"块"指相似度较低、 占空比较低的物体， 比如塑料袋装的匀质物体， 如果不 是扁包装的话就会形成 "块" ；
"瓶"指类似矿泉水瓶的物品， 主、 侧投影相似性、 占空比、 高宽比都是其重 要特征；
"罐"指类似易拉罐装物品， 与瓶较类似， 但占空比、 高宽比更大；
"根"是指较长的物体， 高宽比是其重要特征。 比如香肠、 木材、 铁管， 均在 此范围；
"管"是比 "根"短， 具有好的对称性的物体， 比如类似洗面奶、 胶水等； "盒"是指有一定厚度的， 长方形物品， 比 "块" 占空比大。 占空比、 高宽比 是其主要特征， 比如肥皂、 许多化妆品、 食品都有类似特征；
"包"指较大的物体， 体积是其重要特征。 比如计算机、 很大很厚的书， 较大 的、 可判断为其它形状的物体不在此类；
"个"是泛指 "其它"物体。
可以看到， 上述分类方式与一般理解有所区别。 比如图 4所示的杯子， 以上述 谓词限定则可能为 "罐" 。 这样定义是为安检具体应用联系， 比如固体爆炸物， 一般 会以 "袋" 、 "片" 、 "块" 的形式出现， 液体则以 "瓶" 、 "罐" 、 "管"为主， 而管制器械 "片" 、 "根"形式较多。 其它几个如 "包" 、 "盒" 、 "个"是对常见 形状做的补充。
在步骤 S57， 输出至少包括所述物体的所述量词描述的语义描述。
在得到被检查行李中各个物体的语义描述后， 可通过多种方式与用户交互。 比 如在结果中直接显示物体轮廓， 提醒用户注意； 或者在用户点击时在屏幕上提取出物 体并显示描述信息， 方便用户进一步理解与标注物体。 另外， 在特定场合， 对物体语 义进行限制， 突出显示满足特定语义的物体， 可以降低图检员劳动强度， 提高工作效 率。
已经获知了物体的位置、 重量， 又获知了物体的形状， 所以只需要再统计平均 原子序数和电子密度（或者单能 CT情况下的线性衰减系数） 即可完成描述。 之后， 各 个谓词信息都已得到， 将其整理即得到物体语义描述即 "形状 +重量 +密度 +原子序数 + 位置"
图 6是描述根据本发明另一实施例的在 CT成像系统中显示物体的方法的流程 图。 根据该实施例， 通过自动检测与描述， 实现被检查行李中物体的自动辅助检测。 物体描述结果是对人工检测的必要补充， 也是加强人机交互的一种手段， 在减少漏检 这一重要问题上有很强的应用价值。
在步骤 S61 , 利用 CT成像系统获取被检查行李的断层数据。
在步骤 S62， 从所述断层数据生成被检查行李中各个物体的三维体数据。 然后， 在步骤 S63，针对每个物体，基于所述三维体数据确定至少包括所述物体的所述量词描 述的语义描述。
在步骤 S64,接收用户对某个物体的选择，在显示所选择物体的三维图像的同时， 呈现该物体的语义描述。
例如， 将检测到的所有物体位置标注在显示窗口中， 当图检员使用鼠标等工具 选择位置在某物体范围内时， 显示物体完整的语义描述。 此外， 图检员可以用鼠标等 工具选取物体， 选取后可以进一步详细标注物体， 增加语义描述内容。 还可以对语义 描述做出限定， 只显示符合限定的物体。 比如限定只提示形状为 "瓶"且重量在 200 克以上的物体， 嫌疑物体的位置可以在二维、 三维图像中表示， 辅助图检员进行判图。 此外， 物体在被选中时可以高亮显示， 也可以屏蔽其它所有内容， 只显示物体内容。 备选地， 将上述实施例一些限制加强， 比如加强体数据阈值限定、 物体形状限制， 则 可用作特定物体如爆炸物、 违禁品的自动检测。
在其他实施例中， 产生被检查行李中各个物体的语义描述的过程可以参考上结 合图 5描述的实施例。
根据本发明的实施例， 在进行深度投影之前， 还可以创建被检查行李中物体的 三维模型， 以进一步提取形状特征和进行安全检查。 图 7是描述根据本发明的另一实 施例在 CT成像系统中创建被检查行李中物体的三维模型的方法的流程图。
如图 7所示， 在步骤 S71， 利用 CT成像系统获取被检查行李的断层数据。 在步 骤 S72， 对断层数据进行插值， 生成被检查行李的三维体数据。 例如， 在双能情况下， 三维体数据包括密度体数据和原子序数体数据。 在单能情况下， 三维体数据包括线性 衰减系数。
在得到一系列连续断层下的 DECT密度图、 原子密度图后， 要对两者分别进行 三维插值， 使得图像在断层内、 断层间的分辨率一致。 三维插值的公知算法较多， 如 商业将软件 Intel IPP(Intel Integrated Performance Primitives)函数库， 开源软件 Kitware VTK(Visualization Toolkit)函数库， 均提供这一功能。 插值之后， 二维断层数据转换为 三维体数据。 下文未特指的情况下， 所述 "体数据"包括密度体数据与原子序数体数 据。
其次， 对体数据进行阈值限定， 去掉应用不关心的杂物比如衣服等物品的干扰。 在具体应用中， 这一步骤可以省略， 但计算量会增大， 得到的 "物体"结果数量会很 多， 导致结果可用性差。
之后， 使用三维双边滤波 (bilateral filter)进行体数据滤波。 实施例使用快速算法， 也可以使用 ITK(Insight Segmentation and Registration Toolkit)函数库实现。
在步骤 S73， 对三维体数据进行无监督分割， 得到多个分割的区域。
对于二维分割算法来说， 往往基于 4/8邻域、 梯度信息等实现分割， 此处则需要 将相关处理扩展到三维， 比如 4邻域扩展为 6邻域； 其次， 分割所涉及的是密度、 院 子序数两部分体数据， 可采取两者的加权和， 或者每个体素以二维矢量表示， 得到统 一的分割结果； 再有，分割需要达到欠分割的效果。优选的，我们使用 Statistical Region Merging(SRM)算法， 并将其扩展到三维处理实现目的。 SRM是一种自下而上 merging 分割算法， 实施例采取的扩展包括：
1) 将原子序数、密度连接为矢量， 即体数据每个体素为二维变量。 以 两矢量的差值矢量模值代替灰度差；
2) 用三维梯度代替二维梯度；用区域体素的体积代替二维中区域像素 面积；
经过上述处理， 即可实现 DECT数据的无监督分割。 另外 SRM分割结果复杂度 由复杂度参数限定。 设置较低的复杂度值即可达到欠分割效果。
在步骤 S74， 对多个分割的区域进行等值面提取， 得到相应的等值面。
对步骤 S73所得的各个分割区域进行等值面提取， 可得相应的等值面。 在实施 例中使用 Marching Cubes算法。
在步骤 S75， 对等值面进行三维表面分割， 以形成各个物体的三维模型。
由于步骤 S73步得到的是欠分割结果， 多个材料特征类似的、 位置紧密相连的 物体无法分割开。 因此需要用表面分割细化三维分割结果。 例如可以使用曲面分割 (Mesh Segmentation)算法。 将曲面分割为多个凸曲面。 该算法为有监督算法， 需要指定 分割结果个数。 在实际应用时， 可以通过类似 K-Means聚类的无监督改进思路， 先计 算或迭代得到分割个数。 但经过我们的实验， 类似方法难于得到好的效果， 因此我们 设定分割个数为 10。 对于 10个分割结果， 此处借鉴所述星凸假设进行曲面 Merge。 假 设分割结果曲面 A的中心为 a， 曲面 B的中心为 b， 若^ b的连线在第三步所得整体 表面内部 (或在外部的体素所占比例小于某阈值)， 则将 A、 B连接起来。 对 10个分割 结果两两进行这个连接过程， 可得到最终的分割结果。
根据本发明的实施例， 还可以对分割后的结果进行处理， 包括填洞、 平滑、 模 型限定三个步骤。 前两个步骤均为图形学中基本操作， 可使用开源软件 Kitware VTK(Visualization Toolkit)函数库实现， 此处不再赘述。 之后， 将模型体素化， 填充以 密度体数据值， 统计表面的面积、 体积、 重量。 模型限定指的是去掉较小的物体， 包 括型面积、 体积、 重量等取值较小的物体。 限定原因有两方面。 其一， 去掉噪声物体， 使得结果更有实际意义； 其二， 很多物体的局部细节被省略， 使下一步形状识别更准 确。 具体限定阈值， 与 DECT的分辨率相关， 根据实际情况设置， 比如重量可以设置 为 50克。
以上的详细描述通过使用示意图、 流程图和 /或示例， 已经阐述了检查物体的方 法、 显示方法、 创建三维模型的方法和设备的众多实施例。 在这种示意图、 流程图和 / 或示例包含一个或多个功能和 /或操作的情况下，本领域技术人员应理解，这种示意图、 流程图或示例中的每一功能和 /或操作可以通过各种结构、 硬件、 软件、 固件或实质上 它们的任意组合来单独和 /或共同实现。 在一个实施例中， 本发明的实施例所述主题的 若干部分可以通过专用集成电路（ASIC)、 现场可编程门阵列 （FPGA)、 数字信号处理 器 （DSP)、 或其他集成格式来实现。 然而， 本领域技术人员应认识到， 这里所公开的 实施例的一些方面在整体上或部分地可以等同地实现在集成电路中， 实现为在一台或 多台计算机上运行的一个或多个计算机程序 （例如， 实现为在一台或多台计算机系统 上运行的一个或多个程序）， 实现为在一个或多个处理器上运行的一个或多个程序（例 如， 实现为在一个或多个微处理器上运行的一个或多个程序）， 实现为固件， 或者实质 上实现为上述方式的任意组合， 并且本领域技术人员根据本公开， 将具备设计电路和 / 或写入软件和 /或固件代码的能力。 此外， 本领域技术人员将认识到， 本公开所述主题 的机制能够作为多种形式的程序产品进行分发， 并且无论实际用来执行分发的信号承 载介质的具体类型如何， 本公开所述主题的示例性实施例均适用。 信号承载介质的示 例包括但不限于： 可记录型介质， 如软盘、 硬盘驱动器、 紧致盘 （CD)、 数字通用盘 (DVD), 数字磁带、 计算机存储器等； 以及传输型介质， 如数字和 /或模拟通信介质
(例如， 光纤光缆、 波导、 有线通信链路、 无线通信链路等）。
虽然已参照几个典型实施例描述了本发明， 但应当理解， 所用的术语是说明和 示例性、 而非限制性的术语。 由于本发明能够以多种形式具体实施而不脱离发明的精 神或实质， 所以应当理解， 上述实施例不限于任何前述的细节， 而应在随附权利要求 所限定的精神和范围内广泛地解释， 因此落入权利要求或其等效范围内的全部变化和 改型都应为随附权利要求所涵盖。

Claims

权利要求
1、一种在 CT成像系统中创建被检查行李中物体的三维模型的方法， 包括步骤: 利用所述 CT成像系统获取被检查行李的断层数据；
对所述断层数据进行插值， 生成所述被检查行李的三维体数据；
对所述被检查行李的三维体数据进行无监督分割， 得到多个分割的区域； 对所述多个分割的区域进行等值面提取， 得到相应的等值面；
对所述等值面进行三维表面分割， 形成所述被检查行李中各个物体的三维模型。
2、 如权利要 1所述的方法， 在对所述被检查行李的三维体数据进行无监督分割 之前， 还包括步骤：
对所述被检查行李的三维体数据进行阈值限定， 去掉杂物数据；
对所述被检查行李的三维体数据进行滤波。
3、 如权利要求 1所述的方法， 对所述等值面进行三维表面分割的步骤包括： 将等值面分割成多个曲面；
如果两个曲面的中心的连线在等值面内， 则将两个曲面连接起来。
4、 如权利要求 1所述的方法， 还包括对三维表面分割后的结果进行填洞、 平滑 和模型限定处理。
5、 如权利要求 1所述的方法， 其中所述 CT成像系统具体为双能 CT成像系统， 所述被检查行李的三维体数据包括电子密度体数据和原子序数体数据。
6、 如权利要求 1所述的方法， 其中所述 CT成像系统具有为单能 CT成像系统， 所述被检查行李的三维体数据包括线性衰减系数体数据。
7、 一种在 CT成像系统中创建被检查行李中物体的三维模型的设备， 包括： 利用所述 CT成像系统获取被检查行李的断层数据的装置；
对所述断层数据进行插值， 生成所述被检查行李的三维体数据的装置； 对所述被检查行李的三维体数据进行无监督分割， 得到多个分割的区域的装置; 对所述多个分割的区域进行等值面提取， 得到相应的等值面的装置；
对所述等值面进行三维表面分割， 形成所述被检查行李中各个物体的三维模型 的装置。
8、 如权利要 7所述的设备， 还包括：
对所述被检查行李的三维体数据进行阈值限定， 去掉杂物数据的装置； 对所述被检查行李的三维体数据进行滤波的装置。
9、 如权利要求 7所述的设备， 对所述等值面进行三维表面分割的装置包括: 将等值面分割成多个曲面的装置；
如果两个曲面的中心的连线在等值面内， 则将两个曲面连接起来的装置。
PCT/CN2013/087662 2012-12-27 2013-11-22 三维模型创建方法和设备 Ceased WO2014101600A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201210581858.2A CN103903303B (zh) 2012-12-27 2012-12-27 三维模型创建方法和设备
CN201210581858.2 2012-12-27

Publications (1)

Publication Number Publication Date
WO2014101600A1 true WO2014101600A1 (zh) 2014-07-03

Family

ID=50994608

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2013/087662 Ceased WO2014101600A1 (zh) 2012-12-27 2013-11-22 三维模型创建方法和设备

Country Status (3)

Country Link
US (1) US9557436B2 (zh)
CN (1) CN103903303B (zh)
WO (1) WO2014101600A1 (zh)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101433242B1 (ko) * 2012-11-16 2014-08-25 경북대학교 산학협력단 정복 시술 로봇 및 그의 구동 제어 방법
US9646411B2 (en) * 2015-04-02 2017-05-09 Hedronx Inc. Virtual three-dimensional model generation based on virtual hexahedron models
US9996772B2 (en) * 2016-04-28 2018-06-12 International Business Machines Corporation Detection of objects in images using region-based convolutional neural networks
US12496030B2 (en) 2016-11-17 2025-12-16 James R. Glidewell Dental Ceramics, Inc. Scanning dental impressions
US11559378B2 (en) * 2016-11-17 2023-01-24 James R. Glidewell Dental Ceramics, Inc. Scanning dental impressions
CN106526686B (zh) * 2016-12-07 2019-05-07 同方威视技术股份有限公司 螺旋ct设备和三维图像重建方法
US11205103B2 (en) 2016-12-09 2021-12-21 The Research Foundation for the State University Semisupervised autoencoder for sentiment analysis
CN106886054A (zh) * 2017-04-13 2017-06-23 西安邮电大学 基于三维x射线成像的危险品自动识别装置及方法
WO2019028721A1 (zh) * 2017-08-10 2019-02-14 哈尔滨工业大学 用于物品识别的方法、装置、设备和安检系统
JP7415936B2 (ja) * 2018-10-02 2024-01-17 ソニーグループ株式会社 情報処理装置および情報処理方法
CN109374658B (zh) * 2018-12-24 2022-05-03 电子科技大学 双视角x射线安检系统的神经网络重建断层成像方法
US11534271B2 (en) 2019-06-25 2022-12-27 James R. Glidewell Dental Ceramics, Inc. Processing CT scan of dental impression
US11540906B2 (en) 2019-06-25 2023-01-03 James R. Glidewell Dental Ceramics, Inc. Processing digital dental impression
US11622843B2 (en) 2019-06-25 2023-04-11 James R. Glidewell Dental Ceramics, Inc. Processing digital dental impression
JP7250331B2 (ja) * 2019-07-05 2023-04-03 株式会社イシダ 画像生成装置、検査装置及び学習装置
CN110675488B (zh) * 2019-09-24 2023-02-28 电子科技大学 基于深度学习的创意三维体素模型的建模系统的构建方法
CN110726643A (zh) * 2019-11-18 2020-01-24 辽宁机电职业技术学院 一种用于钻石密度测试的激光探测系统及探测方法
CN113643360B (zh) * 2020-05-11 2024-12-27 同方威视技术股份有限公司 目标物体定位方法、装置、设备、介质和程序产品
CN119741377A (zh) * 2020-05-11 2025-04-01 同方威视技术股份有限公司 目标区域定位方法、装置、设备、介质和程序产品
CN111735735B (zh) * 2020-06-11 2022-05-13 山东滨州烟草有限公司 一种智能卷烟包裹无损识别方法及识别器
US11544846B2 (en) 2020-08-27 2023-01-03 James R. Glidewell Dental Ceramics, Inc. Out-of-view CT scan detection
GB2608187B (en) * 2021-06-25 2024-06-19 Smiths Detection France S A S X-ray inspection system and control architecture for an X-ray inspection system
CN115097535B (zh) * 2021-07-07 2024-09-20 清华大学 检查系统和方法
CN114373060B (zh) * 2022-03-23 2022-06-28 超节点创新科技(深圳)有限公司 行李模型生成方法与设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6426987B2 (en) * 2000-06-01 2002-07-30 Kanagawa & Co., Ltd. Imaging system and method of constructing image using the system
US7062009B2 (en) * 2002-09-12 2006-06-13 Analogic Corporation Helical interpolation for an asymmetric multi-slice scanner
CN101900694A (zh) * 2009-05-27 2010-12-01 清华大学 基于直线轨迹扫描的双能欠采样物质识别系统和方法
CN102608135A (zh) * 2007-10-05 2012-07-25 清华大学 在危险品检查系统中确定ct扫描位置的方法和设备

Family Cites Families (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5712926A (en) * 1994-10-20 1998-01-27 Eberhard; Jeffrey Wayne X-ray computed tomography (CT) system for detecting thin objects
US6331116B1 (en) * 1996-09-16 2001-12-18 The Research Foundation Of State University Of New York System and method for performing a three-dimensional virtual segmentation and examination
US6128365A (en) * 1998-02-11 2000-10-03 Analogic Corporation Apparatus and method for combining related objects in computed tomography data
US6076400A (en) * 1998-02-11 2000-06-20 Analogic Corporation Apparatus and method for classifying objects in computed tomography data using density dependent mass thresholds
US6078642A (en) * 1998-02-11 2000-06-20 Analogice Corporation Apparatus and method for density discrimination of objects in computed tomography data using multiple density ranges
US6026171A (en) * 1998-02-11 2000-02-15 Analogic Corporation Apparatus and method for detection of liquids in computed tomography data
US6111974A (en) * 1998-02-11 2000-08-29 Analogic Corporation Apparatus and method for detecting sheet objects in computed tomography data
US6067366A (en) * 1998-02-11 2000-05-23 Analogic Corporation Apparatus and method for detecting objects in computed tomography data using erosion and dilation of objects
US6035014A (en) * 1998-02-11 2000-03-07 Analogic Corporation Multiple-stage apparatus and method for detecting objects in computed tomography data
US6075871A (en) * 1998-02-11 2000-06-13 Analogic Corporation Apparatus and method for eroding objects in computed tomography data
US6317509B1 (en) * 1998-02-11 2001-11-13 Analogic Corporation Computed tomography apparatus and method for classifying objects
US6026143A (en) * 1998-02-11 2000-02-15 Analogic Corporation Apparatus and method for detecting sheet objects in computed tomography data
US6108396A (en) * 1998-02-11 2000-08-22 Analogic Corporation Apparatus and method for correcting object density in computed tomography data
US6272230B1 (en) * 1998-02-11 2001-08-07 Analogic Corporation Apparatus and method for optimizing detection of objects in computed tomography data
US6195444B1 (en) * 1999-01-12 2001-02-27 Analogic Corporation Apparatus and method for detecting concealed objects in computed tomography data
WO2001078005A2 (en) * 2000-04-11 2001-10-18 Cornell Research Foundation, Inc. System and method for three-dimensional image rendering and analysis
US20030095696A1 (en) * 2001-09-14 2003-05-22 Reeves Anthony P. System, method and apparatus for small pulmonary nodule computer aided diagnosis from computed tomography scans
US8045770B2 (en) * 2003-03-24 2011-10-25 Cornell Research Foundation, Inc. System and method for three-dimensional image rendering and analysis
US7277577B2 (en) * 2004-04-26 2007-10-02 Analogic Corporation Method and system for detecting threat objects using computed tomography images
US8059900B2 (en) * 2004-10-08 2011-11-15 General Electric Company Method and apparatus to facilitate visualization and detection of anatomical shapes using post-processing of 3D shape filtering
WO2007070580A2 (en) * 2005-12-12 2007-06-21 Reveal Imaging Technologies Displaced-ray ct inspection
US8244015B2 (en) * 2006-11-22 2012-08-14 General Electric Company Methods and apparatus for detecting aneurysm in vasculatures
US8644578B1 (en) * 2008-04-25 2014-02-04 Stratovan Corporation Method and apparatus of identifying objects of interest using imaging scans
US8600149B2 (en) * 2008-08-25 2013-12-03 Telesecurity Sciences, Inc. Method and system for electronic inspection of baggage and cargo
CN101551906B (zh) * 2009-05-13 2011-07-20 西安电子科技大学 基于三维区域分水岭的腹部ct图像主要血管提取方法
WO2011019456A1 (en) * 2009-06-26 2011-02-17 University Of South Florida Ct atlas of musculoskeletal anatomy to guide treatment of sarcoma
WO2011092594A2 (en) * 2010-02-01 2011-08-04 Superdimension, Ltd. Region-growing algorithm
CN102402788A (zh) * 2011-12-22 2012-04-04 华南理工大学 一种三维超声图像的分割方法
WO2013173574A1 (en) * 2012-05-16 2013-11-21 The Johns Hopkins University Imaging system and method for use of same to determine metric scale of imaged bodily anatomy
CN102768699B (zh) * 2012-06-14 2016-08-24 西安交通大学 基于ct图像精确重构异质材料微观有限元网格模型的方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6426987B2 (en) * 2000-06-01 2002-07-30 Kanagawa & Co., Ltd. Imaging system and method of constructing image using the system
US7062009B2 (en) * 2002-09-12 2006-06-13 Analogic Corporation Helical interpolation for an asymmetric multi-slice scanner
CN102608135A (zh) * 2007-10-05 2012-07-25 清华大学 在危险品检查系统中确定ct扫描位置的方法和设备
CN101900694A (zh) * 2009-05-27 2010-12-01 清华大学 基于直线轨迹扫描的双能欠采样物质识别系统和方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHANG YONG: "Research on 3D Reconstruction System for RPM Based on Medical Images, China outstanding master's degree thesis full-text database", INFORMATION TECHNOLOGY, 15 June 2012 (2012-06-15) *

Also Published As

Publication number Publication date
US20140185742A1 (en) 2014-07-03
US9557436B2 (en) 2017-01-31
CN103903303B (zh) 2018-01-30
CN103903303A (zh) 2014-07-02

Similar Documents

Publication Publication Date Title
WO2014101600A1 (zh) 三维模型创建方法和设备
WO2014101621A1 (zh) 检查物体的方法、显示方法和设备
US10102641B2 (en) Methods for extracting shape feature, inspection methods and apparatuses
US8849016B2 (en) Panoramic image generation from CBCT dental images
McCulloch et al. Model-based detection of lung nodules in computed tomography exams1: Thoracic computer-aided diagnosis
Praßni et al. Shape-based transfer functions for volume visualization
JP2007307358A (ja) 画像処理方法および装置ならびにプログラム
JP7550832B2 (ja) 診断撮像における画像特徴のアノテーション
JP2014508954A (ja) 合成オブジェクト分割方法およびシステム｛ｃｏｍｐｏｕｎｄｏｂｊｅｃｔｓｅｐａｒａｔｉｏｎ｝
Xiao et al. A cascade and heterogeneous neural network for CT pulmonary nodule detection and its evaluation on both phantom and patient data
WO2016095776A1 (zh) 一种定位三维ct图像中的目标的方法和安检ct系统
CN116188385A (zh) 三维ct图像中目标对象剥离方法、装置和安检ct系统
JP7530476B2 (ja) 放射画像識別モデルをオンラインでトレーニングするための方法、放射画像識別方法及び装置
HK1195820B (zh) 三维模型创建方法和设备
HK1195820A (zh) 三维模型创建方法和设备
HK1195943B (zh) 检查物体的方法、显示方法和设备
HK1195943A (zh) 检查物体的方法、显示方法和设备
Singh Brain tumor detection using image segmentation
HK1195805B (zh) 提取形状特徵的方法、安全检查方法以及设备
HK1195805A (zh) 提取形状特徵的方法、安全检查方法以及设备
WO2026051531A1 (zh) 三维ct数据的目标识别方法、装置及射线扫描检测系统
CN115508390A (zh) 一种爆炸物识别方法
Qin et al. A 3D Modeling Scheme for Cerebral Vasculature from MRA Datasets.
Devarakota et al. Automatic lung nodule detection in thick slice CT: a comparative study of different gating schemes in CAD
HK1218157B (zh) 安检ct系统及其方法

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: 13869117

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: 13869117

Country of ref document: EP

Kind code of ref document: A1