WO2021115302A1 - 一种3d智能视觉设备 - Google Patents

一种3d智能视觉设备 Download PDF

Info

Publication number
WO2021115302A1
WO2021115302A1 PCT/CN2020/134764 CN2020134764W WO2021115302A1 WO 2021115302 A1 WO2021115302 A1 WO 2021115302A1 CN 2020134764 W CN2020134764 W CN 2020134764W WO 2021115302 A1 WO2021115302 A1 WO 2021115302A1
Authority
WO
WIPO (PCT)
Prior art keywords
image acquisition
target
acquisition device
rotating
image
Prior art date
Application number
PCT/CN2020/134764
Other languages
English (en)
French (fr)
Inventor
左忠斌
左达宇
Original Assignee
左忠斌
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 左忠斌 filed Critical 左忠斌
Publication of WO2021115302A1 publication Critical patent/WO2021115302A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01BMEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B11/00Measuring arrangements characterised by the use of optical techniques
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01BMEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B11/00Measuring arrangements characterised by the use of optical techniques
    • G01B11/24Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures

Definitions

  • the invention relates to the technical field of shape measurement, in particular to the technical field of 3D shape measurement.
  • the camera is usually rotated relative to the target, or multiple cameras are set around the target to perform acquisition at the same time.
  • the Digital Emily project of the University of Southern California uses a spherical bracket to fix hundreds of cameras at different positions and angles on the bracket to realize 3D collection and modeling of the human body.
  • the distance between the camera and the target should be short, at least within the range that can be arranged, so that the camera can collect images of the target at different positions.
  • the present invention is proposed to provide a 3D intelligent vision device that overcomes the above-mentioned problems or at least partially solves the above-mentioned problems.
  • the present invention provides a 3D intelligent vision device
  • the rotating device is used to drive the acquisition area of the image acquisition device to move relative to the target;
  • An image acquisition device for acquiring a set of images of the target object through the above-mentioned relative motion
  • the optical axis of the image acquisition device and the rotation plane of the rotating device have an included angle ⁇ .
  • it also includes an angle adjusting device for adjusting the angle between the optical axis of the image acquisition device and the rotation plane of the rotating device.
  • the position at which the image acquisition device collects the above-mentioned set of images meets the following conditions:
  • it further includes a processing unit, which is configured to use the 3D model of a plurality of synthetic target objects in the above-mentioned set of images.
  • the rotating device is a turntable.
  • it also includes a cylindrical shell, and the rotating device is accommodated in the shell.
  • the light source is located on the cross section of the cylindrical housing or on the turntable.
  • the image acquisition device is a visible light camera and/or an infrared camera.
  • FIG. 1 is a schematic diagram of the structure of a 3D intelligent vision device in an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a partially enlarged structure of a 3D intelligent vision device in an embodiment of the present invention
  • 1 image acquisition device 2 rotating device, 3 cylindrical casing, 4 angle adjustment device.
  • the present invention provides a 3D intelligent vision device.
  • the image capturing device 1 is installed on the rotating device 2.
  • the rotating device is housed in a cylindrical housing 3 and can rotate freely in the cylindrical housing.
  • the image acquisition device 1 is used to acquire a set of images of the target through the relative movement of the acquisition area of the image acquisition device 1 and the target; the acquisition area moving device is used to drive the acquisition area of the image acquisition device to move relative to the target.
  • the acquisition area is the effective field of view range of the image acquisition device.
  • the image acquisition device 1 may be a camera, and the rotating device 2 may be a turntable.
  • the camera setting 2 is on the turntable, and the optical axis of the camera is at a certain angle with the turntable surface, and the turntable surface is approximately parallel to the object to be collected.
  • the turntable drives the camera to rotate, so that the camera collects images of the target at different positions.
  • the camera is installed on the turntable through the angle adjustment device 4, as shown in Fig. 2, the angle adjustment device 4 can be rotated to adjust the angle between the optical axis of the image acquisition device 1 and the turntable surface, and the adjustment range is -90° ⁇ 90° .
  • the optical axis of the image acquisition device 1 can be offset in the direction of the central axis of the turntable, that is, the ⁇ can be adjusted in the direction of -90°.
  • the optical axis of the image acquisition device 1 can be offset from the central axis of the turntable, that is, the ⁇ can be adjusted in the direction of 90°.
  • the above adjustment can be done manually, or the 3D intelligent vision device can be provided with a distance measuring device to measure the distance from the target, and automatically adjust the ⁇ angle according to the distance.
  • the turntable can be connected with a motor through a transmission device, and rotate under the drive of the motor, and drive the image acquisition device 1 to rotate.
  • the transmission device can be a conventional mechanical structure such as a gear system or a transmission belt.
  • multiple image collection devices 1 can be provided on the turntable.
  • a plurality of image acquisition devices 1 are sequentially distributed along the circumference of the turntable.
  • an image acquisition device 1 can be provided at both ends of any diameter of the turntable. It is also possible to arrange one image acquisition device 1 every 60° circumferential angle, and 6 image acquisition devices 1 are evenly arranged on the entire disk.
  • the above-mentioned multiple image acquisition devices may be the same type of cameras or different types of cameras. For example, a visible light camera and an infrared camera are set on the turntable, so that images of different bands can be collected.
  • the image acquisition device 1 is used to acquire an image of a target object, and it can be a fixed focus camera or a zoom camera. In particular, it can be a visible light camera or an infrared camera. Of course, it is understandable that any device with image acquisition function can be used and does not constitute a limitation of the present invention. For example, it can be CCD, CMOS, camera, video camera, industrial camera, monitor, camera, mobile phone, tablet, notebook, Mobile terminals, wearable devices, smart glasses, smart watches, smart bracelets, and all devices with image capture functions.
  • the rotating device 2 can also be in various forms such as a rotating arm, a rotating beam, and a rotating bracket, as long as it can drive the image acquisition device to rotate. No matter which method is used, the optical axis of the image acquisition device 1 and the rotating surface have a certain included angle ⁇ .
  • the light source is distributed around the lens of the image acquisition device 1 in a dispersed manner.
  • the light source is a ring LED lamp on the periphery of the lens, which is located on the turntable; it can also be arranged on the cross section of the cylindrical housing. Since in some applications, the collected object is a human body, it is necessary to control the intensity of the light source to avoid causing discomfort to the human body.
  • a soft light device such as a soft light housing, can be arranged on the light path of the light source. Or directly use the LED surface light source, not only the light is softer, but also the light is more uniform.
  • an OLED light source can be used, which is smaller in size, has softer light, and has flexible characteristics that can be attached to curved surfaces.
  • the light source can also be set in other positions that can provide uniform illumination for the target.
  • the light source can also be a smart light source, that is, the light source parameters are automatically adjusted according to the target object and ambient light conditions.
  • the motor drives the turntable to rotate and drives the camera to rotate, so that the position of the optical axis of the camera moves in space.
  • the image capture device 1 captures an image of a target every time interval L distance, and the turntable rotates 360° and the image capture device 1 captures n images. These images are images captured by the camera at different positions.
  • the collection can be performed while the turntable is rotating, or the camera can be stopped after the camera is rotated to the corresponding collection position, and then continue to rotate to the next collection position after the collection is completed.
  • the state of the target object will change, so the speed of acquisition needs to be increased. Otherwise, the state of the target object collected by the image acquisition device 1 is different in different images, which will make 3D synthesis and modeling impossible.
  • 1Set n image acquisition devices on the turntable so that n images can be taken at one time, and n images can be obtained at the next position.
  • 2In order to save costs at the same time although the number of image acquisition devices 1 will not increase, the rotating speed of the turntable can be increased, but in this way, the shutter of the image acquisition device 1 needs to be adjusted to a faster mode, otherwise the image will be blurred.
  • the increase in shutter speed requires better lighting conditions for the light source. Therefore, it is necessary to provide a better light source or a scene with better natural light can use this method.
  • the processor also called a processing unit, is used to synthesize a 3D model of the target object according to a 3D synthesis algorithm according to a plurality of images collected by the image acquisition device to obtain 3D information of the target object.
  • the image acquisition device 1 sends the acquired multiple images to a processing unit, and the processing unit obtains the 3D information of the target object according to the multiple images in the above-mentioned set of images.
  • the processing unit can be directly arranged in the housing where the image acquisition device 1 is located, or it can be connected to the image acquisition device through a data line or in a wireless manner.
  • the processing unit can be used as the processing unit, and the image data collected by the image acquisition device 1 is transmitted to it for 3D synthesis.
  • the data of the image acquisition device 1 can also be transmitted to a cloud platform, and the powerful computing capability of the cloud platform can be used for 3D synthesis.
  • g(x, y) is the gray value of the original image at (x, y)
  • f(x, y) is the gray value of the original image after being enhanced by the Wallis filter
  • m g is the local gray value of the original image Degree mean
  • s g is the local gray-scale standard deviation of the original image
  • m f is the local gray-scale target value of the transformed image
  • s f is the local gray-scale standard deviation target value of the transformed image.
  • c ⁇ (0,1) is the expansion constant of the image variance
  • b ⁇ (0,1) is the image brightness coefficient constant.
  • the filter can greatly enhance the image texture patterns of different scales in the image, so the number and accuracy of feature points can be improved when extracting the point features of the image, and the reliability and accuracy of the matching result can be improved in the photo feature matching.
  • the SURF feature matching method mainly includes three processes, feature point detection, feature point description and feature point matching. This method uses Hessian matrix to detect feature points, uses Box Filters to replace second-order Gaussian filtering, and uses integral images to accelerate convolution to increase the calculation speed and reduce the dimensionality of local image feature descriptors. To speed up the matching speed.
  • the main steps include 1 constructing the Hessian matrix to generate all points of interest for feature extraction.
  • the purpose of constructing the Hessian matrix is to generate stable edge points (mutation points) of the image; 2 constructing the scale space feature point positioning, which will be processed by the Hessian matrix Compare each pixel point with 26 points in the neighborhood of two-dimensional image space and scale space, and initially locate the key points, and then filter out the key points with weaker energy and the key points that are incorrectly positioned to filter out the final stable 3
  • the main direction of the feature point is determined by using the Harr wavelet feature in the circular neighborhood of the statistical feature point.
  • the sum of the horizontal and vertical harr wavelet features of all points in the 60-degree fan is counted, and then the fan is rotated at an interval of 0.2 radians and the harr wavelet eigenvalues in the area are counted again.
  • the direction of the sector with the largest value is taken as the main direction of the feature point; 4 Generate a 64-dimensional feature point description vector, and take a 4*4 rectangular area block around the feature point, but the direction of the obtained rectangular area is along the main direction of the feature point. direction.
  • Each sub-region counts 25 pixels of haar wavelet features in the horizontal and vertical directions, where the horizontal and vertical directions are relative to the main direction.
  • Input the matching feature point coordinates use the beam method to adjust the sparse target 3D point cloud and the position and posture data of the camera to obtain the sparse target model 3D point cloud and position model coordinates;
  • sparse feature points as initial values, dense matching of multi-view photos is performed to obtain dense point cloud data.
  • the process has four main steps: stereo pair selection, depth map calculation, depth map optimization, and depth map fusion. For each image in the input data set, we select a reference image to form a stereo pair for calculating the depth map. Therefore, we can get rough depth maps of all images. These depth maps may contain noise and errors. We use its neighborhood depth map to check consistency to optimize the depth map of each image. Finally, depth map fusion is performed to obtain a three-dimensional point cloud of the entire scene.
  • the main process includes: 1The texture data is obtained through the image reconstruction target's surface triangle grid; 2The visibility analysis of the reconstructed model triangle. Use the image calibration information to calculate the visible image set of each triangle and the optimal reference image; 3The triangle surface clustering generates texture patches. According to the visible image set of the triangle surface, the optimal reference image and the neighborhood topological relationship of the triangle surface, the triangle surface cluster is generated into a number of reference image texture patches; 4The texture patches are automatically sorted to generate texture images. Sort the generated texture patches according to their size relationship, generate the texture image with the smallest enclosing area, and obtain the texture mapping coordinates of each triangle.
  • the optical axis direction of the image acquisition device does not change relative to the target object at different acquisition positions, and is usually roughly perpendicular to the surface of the target object.
  • the position of two adjacent image acquisition devices 1 or the image acquisition device 1 Two adjacent collection locations meet the following conditions:
  • d takes the length of the rectangle; when the above two positions are along the width direction of the photosensitive element of the image capture device 1, d is the width of the rectangle.
  • the distance from the photosensitive element to the surface of the target along the optical axis is taken as M.
  • L should be the linear distance between the optical centers of the two image capture devices 1, but because the position of the optical centers of the image capture devices 1 is not easy to determine in some cases, the image capture device can also be used in some cases
  • the center of the photosensitive element of 1, the geometric center of the image capture device 1, the axis center of the image capture device and the pan/tilt (or platform, bracket), and the center of the proximal or distal surface of the lens are replaced.
  • the error is within an acceptable range, so the above range is also within the protection scope of the present invention.
  • the adjacent acquisition positions in the present invention refer to two adjacent positions on the moving track where the acquisition action occurs when the image acquisition device moves relative to the target. This is usually easy to understand for the movement of the image capture device. However, when the target object moves to cause the two to move relative to each other, at this time, the movement of the target object should be converted into the target object's immobility according to the relativity of the movement, and the image acquisition device moves. At this time, measure the two adjacent positions of the image acquisition device where the acquisition action occurs in the transformed movement track.
  • the 3D intelligent vision device For example, install the 3D intelligent vision device on the crane tower in the factory.
  • the pictures of the target object under the crane tower can be collected in real time, and the 3D model of the target object can be synthesized in the processing unit to identify the type of the target object. It is convenient for the crane to hoist the target to the corresponding area.
  • the image capture device captures images
  • the image acquisition device can also collect video data, directly use the video data or intercept images from the video data for 3D synthesis.
  • the shooting position of the corresponding frame of the video data or the captured image used in the synthesis still satisfies the above empirical formula.
  • the above-mentioned target object, target object, and object all represent objects for which three-dimensional information is pre-acquired. It can be a physical object, or it can be a combination of multiple objects. For example, it can be a head, a hand, and so on.
  • the three-dimensional information of the target includes a three-dimensional image, a three-dimensional point cloud, a three-dimensional grid, a local three-dimensional feature, a three-dimensional size, and all parameters with a three-dimensional feature of the target.
  • the so-called three-dimensional in the present invention refers to three-direction information of XYZ, especially depth information, which is essentially different from only two-dimensional plane information. It is also essentially different from the definitions called three-dimensional, panoramic, holographic, and three-dimensional, but actually only include two-dimensional information, especially depth information.
  • the collection area mentioned in the present invention refers to the range that an image collection device (such as a camera) can shoot.
  • the image acquisition device in the present invention can be CCD, CMOS, camera, video camera, industrial camera, monitor, camera, mobile phone, tablet, notebook, mobile terminal, wearable device, smart glasses, smart watch, smart bracelet and belt All devices with image capture function.
  • modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and/or processes or units are mutually exclusive, any combination can be used to compare all features disclosed in this specification (including the accompanying claims, abstract and drawings) and any method or methods disclosed in this manner or All the processes or units of the equipment are combined. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.
  • the various component embodiments of the present invention may be implemented by hardware, or by software modules running on one or more processors, or by a combination of them.
  • a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all of the functions based on some or all of the components in the device of the present invention according to the embodiments of the present invention.
  • DSP digital signal processor
  • the present invention can also be implemented as a device or device program (for example, a computer program and a computer program product) for executing part or all of the methods described herein.
  • Such a program for realizing the present invention may be stored on a computer-readable medium, or may have the form of one or more signals.
  • Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

Abstract

一种3D智能视觉设备,包括:转动装置(2),用于驱动图像采集装置(1)的采集区域与目标物产生相对运动;图像采集装置(1),用于通过相对运动采集目标物一组图像;图像采集装置(1)的光轴与转动装置(2)转动平面具有夹角γ。首次提出通过相机光轴与转动装置(2)呈一定夹角而非平行的方式转动来采集目标物图像,实现3D合成和建模,而无需绕目标物转动,提高了场景的适应性。

Description

一种3D智能视觉设备 技术领域
本发明涉及形貌测量技术领域,特别涉及3D形貌测量技术领域。
背景技术
目前在利用视觉方式进行3D采集和测量时,通常使得相机相对目标物转动,或在目标物周边设置多个相机同时进行采集。例如南加州大学的Digital Emily项目,采用球型支架,在支架上不同位置不同角度固定了上百个相机,从而实现人体的3D采集和建模。然而无论哪种方式,都需要相机与目标物距离较短,至少应当在可布置的范围内,这样才能形成相机在不同位置采集目标物图像。
然而在一些应用中,无法环绕目标物进行图像的采集。例如监控探头在采集被监控区域时,由于区域较大、距离较远,且采集对象不固定,因此难以围绕目标对象设置相机,或使得相机围绕目标对象转动。在这种情形下如何进行目标对象的3D采集与建模是亟待解决的问题。
另外,在现有技术中,为了同时提高合成速度和合成精度,通常通过优化算法的方法实现。并且本领域一直认为解决上述问题的途径在于算法的选择和更新,截止目前没有任何提出其他角度同时提高合成速度和合成精度的方法。然而,算法的优化目前已经达到瓶颈,在没有更优理论出现前,已经无法兼顾提高合成速度和合成的精度。在现有技术中,也曾提出使用包括旋转角度、目标物尺寸、物距的经验公式限定相机位置,从而兼顾合成速度和效果。然而在实际应用中发现:除非有精确量角装置,否则用户对角度并不敏感,难以准确确定角度;目标物尺寸难以准确确定,特别是某些应用场合目标物需要频繁更换,每次测量带来大量额外工作量,并且需要专业设备才能准确测量不规则目标物。测量的误差导致相机位置设定误差,从而会影响采集合成速度和效果;准确度和速度还需要进一步提高。
因此,目前急需解决以下技术问题:①能够采集较远距离,非特定目标的3D信息。②能够在不围绕目标物转动或排布的情况下采集目标物三维信息。③同时兼顾合成速度和合成精度。
发明内容
鉴于上述问题,提出了本发明提供一种克服上述问题或者至少部分地解决上述问题的3D智能视觉设备。
本发明提供了一种3D智能视觉设备,
转动装置,用于驱动图像采集装置的采集区域与目标物产生相对运动;
图像采集装置,用于通过上述相对运动采集目标物一组图像;
图像采集装置的光轴与转动装置转动平面具有夹角γ。
可选的,还包括角度调节装置,用于调节图像采集装置的光轴与转动装置转动平面的夹角大小。
可选的,在上述相对运动过程中,图像采集装置采集上述一组图像时的位置符合如下条件:
Figure PCTCN2020134764-appb-000001
μ<0.482
其中L为相邻两个采集位置图像采集装置光心的直线距离;f为图像采集装置的焦距;d为图像采集装置感光元件(CCD)的矩形长度;M为图像采集装置感光元件沿着光轴到目标物表面的距离;μ为经验系数。
可选的,图像采集装置为多个,在转动装置旋转面上分布。
可选的,还包括处理单元,用于利用上述一组图像中的多个合成目标物的3D模型。
可选的,所述转动装置为转盘。
可选的,还包括筒状外壳,转动装置容纳在外壳内。
可选的,光源位于筒状外壳横截面上,或位于转盘上。
可选的,图像采集装置为可见光相机和/或红外相机。
可选的,-90°<γ<90°。
发明点及技术效果
1、首次提出通过相机光轴与转盘呈一定夹角而非平行的方式转动来采集目标物图像,实现3D合成和建模,而无需绕目标物转动,提高了场景的适应性。
2、通过优化相机采集图片的位置,保证能够同时提高合成速度和合成精度。
3、优化相机采集位置时,无需测量角度,无需测量目标尺寸,适用性更强。
附图说明
通过阅读下文优选实施方式的详细描述,各种其他的优点和益处对于本领域普通技术人员将变得清楚明了。附图仅用于示出优选实施方式的目的,而并不认为是对本实用新型的限制。而且在整个附图中,用相同的参考符号表示相同的部件。在附图中:
图1为本发明实施例中3D智能视觉设备的结构示意图;
图2为本发明实施例中3D智能视觉设备局部放大的结构示意图;
附图标记与各部件的对应关系如下:
1图像采集装置、2旋转装置、3筒状外壳、4角度调整装置。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例,然而应当理解,可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反,提供这些实施例是为了能够更透彻地理解本公开,并且能够将本公开的范围完整的传达给本领域的技术人员。
为解决上述技术问题,本发明提供了一种用于3D智能视觉设备。
3D智能视觉设备 结构
包括图像采集装置1、旋转装置2和筒状外壳3。如图1,图像采集装置1安装在旋转装置2上,旋转装置容纳在筒状外壳3内,并且可以在筒状外壳内自由转动。
图像采集装置1用于通过图像采集装置1的采集区域与目标物相对运动采集目标物一组图像;采集区域移动装置,用于驱动图像采集装置的采集区域与目标物产生相对运动。采集区域为图像采集装置的有效视场范围。
图像采集装置1可以为相机,旋转装置2可以为转盘。相机设置2在转盘上,且相机光轴与转盘面呈一定夹角,转盘面与待采集目标物近似平行。转盘带动相机转动,从而使得相机在不同位置采集目标物的图像。
进一步,相机通过角度调整装置4安装在转盘上,如图2,角度调整装置4可以转动从而调整图像采集装置1的光轴与转盘面的夹角,调节范围为-90°< γ<90°。在拍摄较近目标物时,可以使得图像采集装置1光轴向转盘中心轴方向偏移,即将γ向-90°方向调节。而在拍摄腔体内部时,可以使得图像采集装置1光轴向偏离转盘中心轴方向偏移,即将γ向90°方向调节。上述调节可以手动完成,也可以给3D智能视觉设备设置测距装置,测量其距离目标物的距离,根据该距离来自动调整γ角度。
转盘可通过传动装置与电机连接,在电机的驱动下转动,并带动图像采集装置1转动。传动装置可以为齿轮系统或传动带等常规机械结构。
为了提高采集效率,转盘上可以设置多个图像采集装置1。多个图像采集装置1沿转盘圆周依次分布。例如可以在转盘任意一条直径两端分别设置一个图像采集装置1。也可以每隔60°圆周角设置一个图像采集装置1,整个圆盘均匀设置6个图像采集装置1。上述多个图像采集装置可以为同一类型相机,也可以为不同类型相机。例如在转盘上设置一个可见光相机及一个红外相机,从而能够采集不同波段图像。
图像采集装置1用于采集目标物的图像,其可以为定焦相机,或变焦相机。特别是即可以为可见光相机,也可以为红外相机。当然,可以理解的是任何具有图像采集功能的装置均可以使用,并不构成对本发明的限定,例如可以为CCD、CMOS、相机、摄像机、工业相机、监视器、摄像头、手机、平板、笔记本、移动终端、可穿戴设备、智能眼镜、智能手表、智能手环以及带有图像采集功能所有设备。
旋转装置2除了转盘,也可以为转动臂、转动梁、转动支架等多种形式,只要能够带动图像采集装置转动即可。无论使用哪种方式,图像采集装置1的光轴与转动面均具有一定的夹角γ。
通常情况下,光源位于图像采集装置1的镜头周边分散式分布,例如光源为在镜头周边的环形LED灯,位于转盘上;也可以设置在筒状外壳的横截面上。由于在一些应用中,被采集对象为人体,因此需要控制光源强度,避免造成人体不适。特别是可以在光源的光路上设置柔光装置,例如为柔光外壳。或者直接采用LED面光源,不仅光线比较柔和,而且发光更为均匀。更佳地,可以采用OLED光源,体积更小,光线更加柔和,并且具有柔性特性,可以贴附于弯曲的表面。光源也可以设置于其他能够为目标物提供均匀照明的位置。光源也可以为智能光源,即根据目标物及环境光的情况自动调整光源参数。
3D采集流程
电机驱动转盘转动,带动相机转动,从而使得相机的光轴位置在空间内发生移动。例如每间隔L距离图像采集装置1采集一次目标物的图像,转盘转动360°图像采集装置1会采集n张图像,这些图像是相机在不同位置采集到的图像。可以在转盘转动的同时进行采集,也可以在相机转动到对应采集位置后停止转动,采集完毕后再继续转动到下一个采集位置。
由于在某些场合下,目标物状态会发生变化,因此需要提高采集的速度,否则图像采集装置1采集到的目标物在不同图像中状态不同会导致无法3D合成和建模。此时可以通过两种方法解决:①在转盘上设置n个图像采集装置,这样一次可以拍摄n张图像,在下一个位置又可以获得n张图像。②为了同时节约成本,虽然图像采集装置1数量不增加,但转盘转速可以加快,但这样需要将图像采集装置1的快门调节成较快的模式,否则将会导致图像模糊。而快门速度的提高需要较好的光源照明条件。因此,需要提供较好的光源或具有较好的自然光的场景可以使用该方法。
3D合成建模装置及方法
处理器,也称处理单元,用以根据图像采集装置采集的多个图像,根据3D合成算法,合成目标物3D模型,得到目标物3D信息。图像采集装置1将采集到的多个图像发送给处理单元,处理单元根据上述所述一组图像中的多个图像得到目标物的3D信息。当然,处理单元可以直接设置在图像采集装置1所在的壳体内,也可以通过数据线或通过无线方式与图像采集装置连接。例如可以使用独立的计算机、服务器及集群服务器等作为处理单元,图像采集装置1采集到的图像数据传输至其上,进行3D合成。同时,也可以将图像采集装置1的数据传输至云平台,利用云平台的强大计算能力进行3D合成。
处理单元中执行如下方法:
1、对所有输入照片进行图像增强处理。采用下述滤波器增强原始照片的反差和同时压制噪声。
Figure PCTCN2020134764-appb-000002
式中:g(x,y)为原始影像在(x,y)处灰度值,f(x,y)为经过Wallis滤波器增强后该处的灰度值,m g为原始影像局部灰度均值,s g为原始影像局部灰度标准偏差,m f为变换后的影像局部灰度目标值,s f为变换后影像局部灰度标准偏差目标值。c∈(0,1)为影像方差的扩展常数,b∈(0,1)为影像亮度系数常数。
该滤波器可以大大增强影像中不同尺度的影像纹理模式,所以在提取影像 的点特征时可以提高特征点的数量和精度,在照片特征匹配中则提高了匹配结果可靠性和精度。
2、对输入的所有图像进行特征点提取,并进行特征点匹配,获取稀疏特征点。采用SURF算子对照片进行特征点提取与匹配。SURF特征匹配方法主要包含三个过程,特征点检测、特征点描述和特征点匹配。该方法使用Hessian矩阵来检测特征点,用箱式滤波器(Box Filters)来代替二阶高斯滤波,用积分图像来加速卷积以提高计算速度,并减少了局部影像特征描述符的维数,来加快匹配速度。主要步骤包括①构建Hessian矩阵,生成所有的兴趣点,用于特征提取,构建Hessian矩阵的目的是为了生成图像稳定的边缘点(突变点);②构建尺度空间特征点定位,将经过Hessian矩阵处理的每个像素点与二维图像空间和尺度空间邻域内的26个点进行比较,初步定位出关键点,再经过滤除能量比较弱的关键点以及错误定位的关键点,筛选出最终的稳定的特征点;③特征点主方向的确定,采用的是统计特征点圆形邻域内的harr小波特征。即在特征点的圆形邻域内,统计60度扇形内所有点的水平、垂直harr小波特征总和,然后扇形以0.2弧度大小的间隔进行旋转并再次统计该区域内harr小波特征值之后,最后将值最大的那个扇形的方向作为该特征点的主方向;④生成64维特征点描述向量,特征点周围取一个4*4的矩形区域块,但是所取得矩形区域方向是沿着特征点的主方向。每个子区域统计25个像素的水平方向和垂直方向的haar小波特征,这里的水平和垂直方向都是相对主方向而言的。该haar小波特征为水平方向值之后、垂直方向值之后、水平方向绝对值之后以及垂直方向绝对值之和4个方向,把这4个值作为每个子块区域的特征向量,所以一共有4*4*4=64维向量作为Surf特征的描述子;⑤特征点匹配,通过计算两个特征点间的欧式距离来确定匹配度,欧氏距离越短,代表两个特征点的匹配度越好。
3、输入匹配的特征点坐标,利用光束法平差,解算稀疏的目标物三维点云和拍照相机的位置和姿态数据,即获得了稀疏目标物模型三维点云和位置的模型坐标值;以稀疏特征点为初值,进行多视照片稠密匹配,获取得到密集点云数据。该过程主要有四个步骤:立体像对选择、深度图计算、深度图优化、深度图融合。针对输入数据集里的每一张影像,我们选择一张参考影像形成一个立体像对,用于计算深度图。因此我们可以得到所有影像的粗略的深度图,这些深度图可能包含噪声和错误,我们利用它的邻域深度图进行一致性检查,来优化每一张影像的深度图。最后进行深度图融合,得到整个场景的三维点云。
4、利用密集点云进行目标物曲面重建。包括定义八叉树、设置函数空间、创建向量场、求解泊松方程、提取等值面几个过程。由梯度关系得到采样点和指示函数的积分关系,根据积分关系获得点云的向量场,计算指示函数梯度场的逼近,构成泊松方程。根据泊松方程使用矩阵迭代求出近似解,采用移动方体算法提取等值面,对所测点云重构出被测物体的模型。
5、目标物模型的全自动纹理贴图。表面模型构建完成后,进行纹理贴图。主要过程包括:①纹理数据获取通过图像重建目标的表面三角面格网;②重建模型三角面的可见性分析。利用图像的标定信息计算每个三角面的可见图像集以及最优参考图像;③三角面聚类生成纹理贴片。根据三角面的可见图像集、最优参考图像以及三角面的邻域拓扑关系,将三角面聚类生成为若干参考图像纹理贴片;④纹理贴片自动排序生成纹理图像。对生成的纹理贴片,按照其大小关系进行排序,生成包围面积最小的纹理图像,得到每个三角面的纹理映射坐标。
图像采集装置采集位置优化
在进行3D采集时,图像采集装置在不同采集位置光轴方向相对于目标物不发生变化,通常大致垂直于目标物表面,此时相邻两个图像采集装置1的位置,或图像采集装置1相邻两个采集位置满足如下条件:
Figure PCTCN2020134764-appb-000003
μ<0.482
其中L为相邻两个采集位置图像采集装置1光心的直线距离;f为图像采集装置1的焦距;d为图像采集装置感光元件(CCD)的矩形长度;M为图像采集装置1感光元件沿着光轴到目标物表面的距离;μ为经验系数。
当上述两个位置是沿图像采集装置1感光元件长度方向时,d取矩形长度;当上述两个位置是沿图像采集装置1感光元件宽度方向时,d取矩形宽度。
图像采集装置1在上述两个位置中的任何一个位置时,感光元件沿着光轴到目标物表面的距离作为M。
如上所述,L应当为两个图像采集装置1光心的直线距离,但由于图像采集装置1光心位置在某些情况下并不容易确定,因此在某些情况下也可以使用图像采集装置1的感光元件中心、图像采集装置1的几何中心、图像采集装置与云台(或平台、支架)连接的轴中心、镜头近端或远端表面的中心替代,经 过试验发现由此带来的误差是在可接受的范围内的,因此上述范围也在本发明的保护范围之内。
利用本发明装置,进行实验,得到了如下实验结果。
Figure PCTCN2020134764-appb-000004
从上述实验结果及大量实验经验可以得出,μ的值应当满足μ<0.482,此时已经能够合成部分3D模型,虽然有一部分无法自动合成,但是在要求不高的情况下也是可以接受的,并且可以通过手动或者更换算法的方式弥补无法合成的部分。特别是μ的值满足μ<0.357时,能够最佳地兼顾合成效果和合成时间的平衡;为了获得更好的合成效果可以选择μ<0.198,此时合成时间会上升,但合成质量更好。而当μ为0.5078时,已经无法合成。但这里应当注意,以上范围仅仅是最佳实施例,并不构成对保护范围的限定。
以上数据仅为验证该公式条件所做实验得到的,并不对发明构成限定。即使没有这些数据,也不影响该公式的客观性。本领域技术人员可以根据需要调整设备参数和步骤细节进行实验,得到其他数据也是符合该公式条件的。
本发明所述的相邻采集位置是指,在图像采集装置相对目标物移动时,移动轨迹上的发生采集动作的两个相邻位置。这通常对于图像采集装置运动容易理解。但对于目标物发生移动导致两者相对移动时,此时应当根据运动的相对性,将目标物的运动转化为目标物不动,而图像采集装置运动。此时再衡量图像采集装置在转化后的移动轨迹中发生采集动作的两个相邻位置。
3D智能视觉设备的应用
例如,在工厂吊塔上安装该3D智能视觉设备,在吊塔工作时,可以实时采集吊塔下目标物的图片,并在处理单元中合成目标物的3D模型,从而识别出目标物的类型,方便吊塔将目标物吊装至相应的区域。
虽然上述实施例中记载图像采集装置采集图像,但不应理解为仅适用于单张图片构成的图片组,这只是为了便于理解而采用的说明方式。图像采集装置也可以采集视频数据,直接利用视频数据或从视频数据中截取图像进行3D合成。但合成时所利用的视频数据相应帧或截取的图像的拍摄位置,依然满足上述经验公式。
上述目标物体、目标物、及物体皆表示预获取三维信息的对象。可以为一实体物体,也可以为多个物体组成物。例如可以为头部、手部等。所述目标物的三维信息包括三维图像、三维点云、三维网格、局部三维特征、三维尺寸及一切带有目标物三维特征的参数。本实用新型里所谓的三维是指具有XYZ三个方向信息,特别是具有深度信息,与只有二维平面信息具有本质区别。也与一些称为三维、全景、全息、三维,但实际上只包括二维信息,特别是不包括深度信息的定义有本质区别。
本发明所说的采集区域是指图像采集装置(例如相机)能够拍摄的范围。本发明中的图像采集装置可以为CCD、CMOS、相机、摄像机、工业相机、监视器、摄像头、手机、平板、笔记本、移动终端、可穿戴设备、智能眼镜、智能手表、智能手环以及带有图像采集功能所有设备。
在此处所提供的说明书中,说明了大量具体细节。然而,能够理解,本实用新型的实施例可以在没有这些具体细节的情况下实践。在一些实例中,并未详细示出公知的方法、结构和技术,以便不模糊对本说明书的理解。
类似地,应当理解,为了精简本公开并帮助理解各个发明方面中的一个或多个,在上面对本发明的示例性实施例的描述中,本发明的各个特征有时被一起分组到单个实施例、图、或者对其的描述中。然而,并不应将该公开的方法解释成反映如下意图:即所要求保护的本发明要求比在每个权利要求中所明确记载的特征更多的特征。更确切地说,如下面的权利要求书所反映的那样,发明方面在于少于前面公开的单个实施例的所有特征。因此,遵循具体实施方式的权利要求书由此明确地并入该具体实施方式,其中每个权利要求本身都作为本发明的单独实施例。
本领域那些技术人员可以理解,可以对实施例中的设备中的模块进行自适应性地改变并且把它们设置在与该实施例不同的一个或多个设备中。可以把实施例中的模块或单元或组件组合成一个模块或单元或组件,以及此外可以把它 们分成多个子模块或子单元或子组件。除了这样的特征和/或过程或者单元中的至少一些是相互排斥之外,可以采用任何组合对本说明书(包括伴随的权利要求、摘要和附图)中公开的所有特征以及如此公开的任何方法或者设备的所有过程或单元进行组合。除非另外明确陈述,本说明书(包括伴随的权利要求、摘要和附图)中公开的每个特征可以由提供相同、等同或相似目的的替代特征来代替。
此外,本领域的技术人员能够理解,尽管在此的一些实施例包括其它实施例中所包括的某些特征而不是其它特征,但是不同实施例的特征的组合意味着处于本发明的范围之内并且形成不同的实施例。例如,在权利要求书中,所要求保护的实施例的任意之一都可以以任意的组合方式来使用。
本发明的各个部件实施例可以以硬件实现,或者以在一个或者多个处理器上运行的软件模块实现,或者以它们的组合实现。本领域的技术人员应当理解,可以在实践中使用微处理器或者数字信号处理器(DSP)来实现根据本发明实施例的基于本发明装置中的一些或者全部部件的一些或者全部功能。本发明还可以实现为用于执行这里所描述的方法的一部分或者全部的设备或者装置程序(例如,计算机程序和计算机程序产品)。这样的实现本发明的程序可以存储在计算机可读介质上,或者可以具有一个或者多个信号的形式。这样的信号可以从因特网网站上下载得到,或者在载体信号上提供,或者以任何其他形式提供。
应该注意的是上述实施例对本发明进行说明而不是对本发明进行限制,并且本领域技术人员在不脱离所附权利要求的范围的情况下可设计出替换实施例。在权利要求中,不应将位于括号之间的任何参考符号构造成对权利要求的限制。单词“包含”不排除存在未列在权利要求中的元件或步骤。位于元件之前的单词“一”或“一个”不排除存在多个这样的元件。本发明可以借助于包括有若干不同元件的硬件以及借助于适当编程的计算机来实现。在列举了若干装置的单元权利要求中,这些装置中的若干个可以是通过同一个硬件项来具体体现。单词第一、第二、以及第三等的使用不表示任何顺序。可将这些单词解释为名称。
至此,本领域技术人员应认识到,虽然本文已详尽示出和描述了本发明的多个示例性实施例,但是,在不脱离本发明精神和范围的情况下,仍可根据本发明公开的内容直接确定或推导出符合本发明原理的许多其他变型或修改。因此,本发明的范围应被理解和认定为覆盖了所有这些其他变型或修改。

Claims (9)

  1. 一种3D智能视觉设备,其特征在于:
    转动装置,用于驱动图像采集装置的采集区域与目标物产生相对运动;
    图像采集装置,用于通过上述相对运动采集目标物一组图像;
    图像采集装置的光轴与用于驱动图像采集装置的转动装置转动平面具有夹角γ;
    在上述相对运动过程中,图像采集装置采集上述一组图像时的位置符合如下条件:
    Figure PCTCN2020134764-appb-100001
    μ<0.482,或μ<0.357
    其中L为相邻两个采集位置图像采集装置光心的直线距离;f为图像采集装置的焦距;d为图像采集装置感光元件的矩形长度;M为图像采集装置感光元件沿着光轴到目标物表面的距离;μ为经验系数。
  2. 如权利要求1所述的设备,其特征在于:还包括角度调节装置,用于调节图像采集装置的光轴与转动装置转动平面的夹角大小。
  3. 如权利要求1所述的设备,其特征在于:图像采集装置为多个,在转动装置旋转面上分布。
  4. 如权利要求1所述的设备,其特征在于:还包括处理单元,用于利用上述一组图像中的多个合成目标物的3D模型。
  5. 如权利要求1所述的设备,其特征在于:所述转动装置为转盘。
  6. 如权利要求1所述的设备,其特征在于:还包括筒状外壳,转动装置容纳在外壳内。
  7. 如权利要求6所述的设备,其特征在于:光源位于筒状外壳横截面上,或位于转盘上。
  8. 如权利要求1所述的设备,其特征在于:图像采集装置为可见光相机和/或红外相机。
  9. 如权利要求1所述的设备,其特征在于:-90°<γ<90°。
PCT/CN2020/134764 2019-12-12 2020-12-09 一种3d智能视觉设备 WO2021115302A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911276144.9 2019-12-12
CN201911276144.9A CN111060008B (zh) 2019-12-12 2019-12-12 一种3d智能视觉设备

Publications (1)

Publication Number Publication Date
WO2021115302A1 true WO2021115302A1 (zh) 2021-06-17

Family

ID=70300728

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/134764 WO2021115302A1 (zh) 2019-12-12 2020-12-09 一种3d智能视觉设备

Country Status (2)

Country Link
CN (2) CN112629412A (zh)
WO (1) WO2021115302A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112629412A (zh) * 2019-12-12 2021-04-09 天目爱视(北京)科技有限公司 一种转动式3d智能视觉设备
CN112254638B (zh) * 2020-10-15 2022-08-12 天目爱视(北京)科技有限公司 一种可俯仰调节的智能视觉3d信息采集设备
CN112304250B (zh) * 2020-10-15 2022-09-16 天目爱视(北京)科技有限公司 一种移动物体之间的三维匹配设备及方法
CN112361962B (zh) * 2020-11-25 2022-05-03 天目爱视(北京)科技有限公司 一种多俯仰角度的智能视觉3d信息采集设备
CN112254669B (zh) * 2020-10-15 2022-09-16 天目爱视(北京)科技有限公司 一种多偏置角度的智能视觉3d信息采集设备

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103776390A (zh) * 2014-01-22 2014-05-07 广东工业大学 三维自然纹理数据扫描机及多视场数据拼接方法
US9322643B2 (en) * 2011-10-18 2016-04-26 Nanyang Technological University Apparatus and method for 3D surface measurement
CN106546568A (zh) * 2016-10-31 2017-03-29 浙江大学 一种获取植物三维叶绿素荧光图像信息的方法及装置
CN107869962A (zh) * 2017-10-31 2018-04-03 南京农业大学 一种基于空间雕刻技术的高通量设施作物三维形态信息测量系统
CN209279885U (zh) * 2018-09-05 2019-08-20 天目爱视(北京)科技有限公司 图像采集设备、3d信息比对及配套物生成装置
CN110553585A (zh) * 2018-09-05 2019-12-10 天目爱视(北京)科技有限公司 一种基于光学阵列的3d信息获取装置
CN111060008A (zh) * 2019-12-12 2020-04-24 天目爱视(北京)科技有限公司 一种3d智能视觉设备
CN211178344U (zh) * 2019-12-12 2020-08-04 天目爱视(北京)科技有限公司 智能三维视觉采集设备

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR2536167A1 (fr) * 1982-11-16 1984-05-18 Frilley Pierre Camera optoelectronique, procede de relevement, a l'aide de cette camera, du profil d'une section transversale d'une cavite, et installation pour la mise en oeuvre du procede
CN101629806B (zh) * 2009-06-22 2011-01-05 哈尔滨工程大学 结合激光发射器的非线性ccd三维定位装置及定位方法
DE202011101594U1 (de) * 2011-05-31 2012-06-01 Ldv-Systeme Gmbh Reflexionsvorrichtung
KR20130024125A (ko) * 2011-08-30 2013-03-08 삼성전기주식회사 스테레오 카메라의 노출조정 장치, 스테레오 카메라 및 그의 노출조정방법
KR20130066184A (ko) * 2011-12-12 2013-06-20 현대모비스 주식회사 레이더 센서를 이용한 카메라 각도 자동 조절 장치 및 방법
CN102495520A (zh) * 2011-12-14 2012-06-13 天津大学 自会聚式多视点三维数据采集系统及方法
CN204924196U (zh) * 2015-04-11 2015-12-30 北京林业大学 林区三维激光扫描仪搭载云台机构
CN205014956U (zh) * 2015-04-28 2016-02-03 深圳市生生电子设备有限公司 三维扫描仪
CN106507086B (zh) * 2016-10-28 2018-08-31 北京灵境世界科技有限公司 一种漫游实景vr的3d呈现方法
CN106383430A (zh) * 2016-11-03 2017-02-08 郭洪礼 一种全息影像显示系统
CN110827196A (zh) * 2018-09-05 2020-02-21 天目爱视(北京)科技有限公司 一种能够对目标物多区域同时进行3d信息采集的装置
CN208795174U (zh) * 2018-09-05 2019-04-26 天目爱视(北京)科技有限公司 相机自转式图像采集设备、比对装置、配套物生成装置
CN208653401U (zh) * 2018-09-05 2019-03-26 天目爱视(北京)科技有限公司 自适应图像采集设备、3d信息比对装置、配套物生成装置
CN208653402U (zh) * 2018-09-10 2019-03-26 天目爱视(北京)科技有限公司 图像获取设备、3d信息比对装置、配套物生成装置

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9322643B2 (en) * 2011-10-18 2016-04-26 Nanyang Technological University Apparatus and method for 3D surface measurement
CN103776390A (zh) * 2014-01-22 2014-05-07 广东工业大学 三维自然纹理数据扫描机及多视场数据拼接方法
CN106546568A (zh) * 2016-10-31 2017-03-29 浙江大学 一种获取植物三维叶绿素荧光图像信息的方法及装置
CN107869962A (zh) * 2017-10-31 2018-04-03 南京农业大学 一种基于空间雕刻技术的高通量设施作物三维形态信息测量系统
CN209279885U (zh) * 2018-09-05 2019-08-20 天目爱视(北京)科技有限公司 图像采集设备、3d信息比对及配套物生成装置
CN110553585A (zh) * 2018-09-05 2019-12-10 天目爱视(北京)科技有限公司 一种基于光学阵列的3d信息获取装置
CN111060008A (zh) * 2019-12-12 2020-04-24 天目爱视(北京)科技有限公司 一种3d智能视觉设备
CN211178344U (zh) * 2019-12-12 2020-08-04 天目爱视(北京)科技有限公司 智能三维视觉采集设备

Also Published As

Publication number Publication date
CN111060008A (zh) 2020-04-24
CN112629412A (zh) 2021-04-09
CN111060008B (zh) 2020-12-22

Similar Documents

Publication Publication Date Title
CN111060023B (zh) 一种高精度3d信息采集的设备及方法
WO2021115302A1 (zh) 一种3d智能视觉设备
WO2021185218A1 (zh) 一种在运动过程中获取物体3d坐标及尺寸的方法
CN111292364B (zh) 一种三维模型构建过程中图像快速匹配的方法
CN110580732B (zh) 一种3d信息获取装置
WO2021185214A1 (zh) 一种在3d建模中远距离标定方法
WO2021115301A1 (zh) 一种近距离目标物3d采集设备
CN111292239B (zh) 一种三维模型拼接设备及方法
WO2021185220A1 (zh) 一种基于坐标测量的三维模型构建及测量方法
CN113532329B (zh) 一种以投射光斑为标定点的标定方法
CN112016570B (zh) 用于背景板同步旋转采集中的三维模型生成方法
CN111160136B (zh) 一种标准化3d信息采集测量方法及系统
CN110986768B (zh) 一种用于目标物3d信息高速采集测量设备
WO2021185216A1 (zh) 一种基于多激光测距的标定方法
WO2021185215A1 (zh) 一种在3d建模中多相机共同标定方法
WO2021115299A1 (zh) 一种用于3d信息采集的的图像采集装置及其选择方法
CN111340959B (zh) 一种基于直方图匹配的三维模型无缝纹理贴图方法
CN211178344U (zh) 智能三维视觉采集设备
CN111208138A (zh) 一种智能木料识别装置
WO2021115297A1 (zh) 3d信息采集的设备及方法
WO2021115296A1 (zh) 用于移动终端的超薄型三维采集的模组
CN111325780B (zh) 一种基于图像筛选的3d模型快速构建方法
WO2021115298A1 (zh) 一种眼镜匹配设计设备
CN211085115U (zh) 标准化生物三维信息采集装置
CN211955232U (zh) 一种木料识别设备

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: 20898678

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: 20898678

Country of ref document: EP

Kind code of ref document: A1