WO2023087652A1 - 面向双光源x射线安检机的待检物品尺寸自动检测方法 - Google Patents
面向双光源x射线安检机的待检物品尺寸自动检测方法 Download PDFInfo
- Publication number
- WO2023087652A1 WO2023087652A1 PCT/CN2022/094767 CN2022094767W WO2023087652A1 WO 2023087652 A1 WO2023087652 A1 WO 2023087652A1 CN 2022094767 W CN2022094767 W CN 2022094767W WO 2023087652 A1 WO2023087652 A1 WO 2023087652A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- detected
- item
- view
- volume
- ray
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B15/00—Measuring arrangements characterised by the use of electromagnetic waves or particle radiation, e.g. by the use of microwaves, X-rays, gamma rays or electrons
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
Definitions
- the invention belongs to the field of size detection, and in particular provides a method for automatically detecting the size of an item to be inspected facing a dual light source X-ray security inspection machine.
- the structure and imaging principle of the security inspection machine lead to distortions in the X-ray imaging of the item.
- the object is on the vertical line from the light source point to the imaging surface (ie, the optical axis)
- the farther the object is from the light source the larger the angle between the object and the optical axis, and the greater the distortion will be.
- the plane of the object to be detected is basically parallel to the imaging plane, and the thickness (height) in the vertical direction is basically negligible.
- the X-ray light source passes through the object to be detected in the vertical direction and forms an image, so the image distortion in the vertical direction is basically negligible, as long as the image distortion in the horizontal direction is considered.
- the imaging space is the entire X-ray inspection channel.
- the vertical height (thickness) cannot be ignored.
- the error makes each light source form a non-orthogonal coordinate system.
- the above situation is widely used in practical applications, including but not limited to customs security inspection, passenger security inspection, food safety, etc. Therefore, in the luggage security inspection machine, the distortion that occurs when the object to be inspected enters the security inspection machine at different angles and different heights (distance from the conveyor belt) is different, that is, each point in the three-dimensional imaging space formed by the X-ray inspection channel is on the imaging surface The corresponding pixel scales are different.
- the actual size calculated by the invention patent CN 112149658 A is proportional to the speed of the conveyor belt, that is, the larger (smaller) the speed of the conveyor belt, the larger (smaller) the calculated actual size, but the object’s
- the actual size should be a fixed value and should not be related to the speed of the conveyor belt, so it is not realistic;
- the actual size calculated by the invention is that the placement angle of the reference object is only considered from 0 to 180 degrees, which means that the reference object is only placed on the conveyor belt
- the distortion of the pixel length in the three-dimensional space of the X-ray detection channel is not considered, and when the object is suspended in the three-dimensional space of the X-ray detection channel of the security inspection machine and placed at any angle, the pixel scale is quite different from the pixel scale on the surface of the conveyor belt , so the calculation result is quite different from the actual situation, which cannot meet the actual needs; in this invention, the pixel scale is calculated by matching the
- the invention only proposes the calculation of the length and area
- the method does not solve the calculation problem of the volume of the object to be measured.
- a large number of objects are three-dimensional, such as wine bottles, spray cans, etc.
- optical imaging coordinate system pixel scale calculation model solves the problem of calculating the length and volume of items to be inspected based on x-ray images.
- the present invention provides an automatic detection method for the size of items to be inspected for a dual light source X-ray security inspection machine.
- the invention can quickly and accurately obtain the size of each item in the luggage through the image processing method, thereby improving the detection accuracy of dangerous items and reducing the false detection rate.
- a method for automatically detecting the size of an item to be inspected for a dual-light source X-ray security inspection machine comprising the following steps:
- Step 1 Set up the main X-ray imaging system and the auxiliary X-ray imaging system on the adjacent sides of the X-ray inspection channel; same plane;
- Step 2 establishing the imaging coordinate system of the main X-ray imaging system and the imaging coordinate system of the auxiliary X-ray imaging system;
- Step 3 Use standard parts to pass through the X-ray inspection channel at different inclinations, different heights, and different horizontal positions perpendicular to the moving direction of the conveyor belt, and obtain several groups of images, each group of images includes a main view and an auxiliary view;
- Step 4 Calculate the x and y directions of the standard part on the main view according to the measured actual length, width and height of the standard part and the number of pixels in the main view and auxiliary view of the standard part in the x, y and z directions and the actual length represented by the unit pixel at the corresponding position in the x and z directions on the auxiliary view, that is, the pixel scale corresponding to the pixel point; the pixel scale of each pixel point and the image coordinates where the pixel point is located are taken as a set of data.
- Group data form a data set; where, the x direction is the horizontal direction of the plane where the main view is located, that is, the moving direction of the X-ray inspection channel conveyor belt, the y direction is the direction perpendicular to the x direction in the plane where the main view is located; the z direction is where the auxiliary view is located The direction perpendicular to the x direction in the plane;
- Step 5 fitting the data set to obtain the pixel scale functions p x (v), p y (y, z) and p z (y, z) in the three directions of x, y, and z respectively;
- Step 6 Put the item to be detected into the X-ray inspection channel to obtain the main view and auxiliary view of the item to be detected, and input it into the item detection model based on deep learning to obtain the item category, position, detection frame and confidence;
- Step 7 Input the coordinates of each pixel point of the detection frame in the main view and auxiliary view of the item to be detected into the fitted pixel scale function in the three directions of x, y, and z, and then the length of the pixel points in the directions of x, y, and z Accumulate separately to obtain the actual length of the object to be tested in the three directions of x, y, and z, and then fuse and calculate the actual length of the object to be tested in the three directions of x, y, and z to obtain the actual length of the object to be tested.
- step five the data sets in the three directions of x, y, and z are respectively fitted, and the pixel scale functions in the three directions of x, y, and z are respectively obtained: p x (v), p y (y, z) and p z (y, z).
- the actual length of the item to be detected is obtained by the following formula:
- l predict represents the actual length of the object to be detected
- c angle represents the angle correction parameter, which is related to the placement angle of the object to be detected
- c h represents the height correction parameter, which is related to the height position of the object to be detected in the auxiliary view
- l x Indicates the actual length of the object to be detected in the x direction
- l y represents the actual length of the object to be detected in the y direction
- ry represents the correction parameter of ly , which is related to the position of the object to be detected in the main view
- l z represents The actual length of the object to be detected in the z direction
- r z represents the correction parameter of l z , which is related to the position of the object to be detected in the auxiliary view
- c angle , c h , ry and r z are all empirical values, Among them, the value interval of c angle is [0.85, 1]; the value interval of c h is [0.9, 1]; the
- a further improvement also includes step 8.
- the volume of the item to be detected is obtained:
- a rectangle is mapped to the three-dimensional space as a cylinder, the diameter of the cylinder is d i , and the height of the cylinder is l i , so the volume of the i-th cylinder Then the total volume of the object to be detected on the corresponding view is According to the above method, the main view volume V overlook and the auxiliary view volume V sidelook are decomposed and calculated;
- V predict r overlook *V overlook +r sideloo k*V sideloo k
- V predict represents the predicted volume of the item to be detected
- r overlook is the contribution coefficient of the main view volume V overlook
- r sidelook is the contribution coefficient of the auxiliary view volume r sidelook
- r overlook and r sidelook are both empirical values, where r overlook The value range is [0, 1], and the value range of r sidelook is [0, 1].
- the predicted volume V predict of the object to be detected is obtained; the predicted volume is compared with the pre-stored common volume of the object to be detected, and the common volume closest to the volume of the object to be detected is used as the volume of the object to be detected.
- the main view is a top view
- the auxiliary view is a side view
- the invention can quickly and accurately obtain the size of each item in the luggage through the image processing method, thereby improving the detection accuracy of dangerous items and reducing the false detection rate.
- the length of contraband is predicted through the following steps:
- Step 1 Set up the main X-ray imaging system and the auxiliary X-ray imaging system on the adjacent sides of the X-ray inspection channel; same plane;
- Step 2 establishing the imaging coordinate system of the main X-ray imaging system and the imaging coordinate system of the auxiliary X-ray imaging system;
- Step 3 Use standard parts to pass through the X-ray inspection channel at different inclinations, different heights and different horizontal positions perpendicular to the moving direction of the conveyor belt, and obtain several sets of images, each set of images includes a main view and an auxiliary view, The front view is a top view, and the top view is a side view;
- Step 4 According to the actual length, width and height of the measured standard part and the number of pixels in the x, y and z directions of the main view and the auxiliary view of the standard part, calculate the x, y and z directions of the standard part in the main view.
- the actual length represented by the unit pixel at the corresponding position in the y direction and the x and z directions of the auxiliary view, that is, the pixel scale corresponding to the pixel point; the pixel scale of each pixel and the image coordinates where the pixel is located are taken as a set of data.
- Group data form a data set; where, the x direction is the horizontal direction of the plane where the main view is located, that is, the moving direction of the X-ray inspection channel conveyor belt, the y direction is the direction perpendicular to the x direction in the plane where the main view is located; the z direction is where the auxiliary view is located The direction perpendicular to the x direction in the plane;
- Step 5 fitting the data set to obtain the pixel scale functions p x (v), p y (y, z) and p z (y, z) in the three directions of x, y, and z respectively;
- Step 6 Put the item to be detected into the X-ray inspection channel to obtain the main view and auxiliary view of the item to be detected;
- Step 7 Input the coordinates of each pixel in the main view and auxiliary view of the item to be detected into the fitted x, y, and z pixel scale functions in three directions, and then add up the lengths of the pixels in the x, y, and z directions respectively, Get the actual length of the object to be tested in the three directions of x, y, and z, and then fuse the actual lengths of the object to be tested in the three directions of x, y, and z to obtain the actual length of the object to be tested.
- the real length of a knife is 190 cm.
- the main view and auxiliary view are obtained.
- l predict 193.5, that is, the predicted length is 193.5 cm.
- the volume calculation method is as follows:
- V predict r overlook *V overlook +r sidelook *V sidelook ,
- V predict represents the predicted volume of the item to be detected
- r overlook is the contribution coefficient of the main view volume V overlook
- r sidelook is the contribution coefficient of the auxiliary view volume V sidelook
- r overlook and r sidelook are both empirical values, where r overlook The value range is [0.5, 0.9], and the value range of r sidelook is [0.6, 0.85].
- V predict represents the predicted volume after container fusion calculation
- r overlook is the contribution coefficient of the main view volume V overlook
- r sidelook is the contribution coefficient of the auxiliary view volume V sidelook .
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Electromagnetism (AREA)
- Geometry (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
Abstract
本发明公开了一种面向双光源X射线安检机的待检物品尺寸自动检测方法，建立基于X射线检查通道尺寸和X射线源参数的X光图像与实际对应物体比例尺的成像模型，通过双视角安检机检测待检测物品的X光图计算出物品的实际尺寸。本发明可以通过图像处理的方法快速精确的得到行李中待检测物品的尺寸，从而判断是否满足相关安全尺寸规定。
Description
本发明属于尺寸检测领域，尤其提供了一种面向双光源X射线安检机的待检物品尺寸自动检测方法。
机场、高铁站等对很多物品的规格做了限定，例如机场单瓶超过100ml的液体不能随身携带，需要进行托运。在动车上长度超过130厘米的杆状物体不能携带上车等。
但是安检机的结构和成像原理导致了物品在x光成像上存在着畸变。物体除了在从光源点和成像面做垂线(即光轴)上以外，物体位置离光源越远、物体与光轴所成夹角越大，产生的畸变越大。不同于传统X光成像场景，待检测物体所在平面基本与成像平面平行，竖直方向上厚(高)度基本可以忽略。X光光源以垂直方向穿过待检测物体并成像，因此在竖直方向上的图像畸变基本可忽略，只要考虑水平方面图像畸变即可。安检机场景中，成像空间是整个X射线检查通道，对于行李等待检物品通过时，竖直方向高度(厚度)已不可忽略，且大多行包安检机，尤其是双视角安检机由于设计或者安装误差使得各光源之间构成的是非正交坐标系。上述情况在实际应用中非常广泛，包括但不限于海关安检、客运安检、食品安全等。因此，在行包安检机中，待检物体以不同角度、不同高度(距离传送带)进入安检机时发生的畸变是不同的，即X射线检查通道所形成的立体成像空间中每一点在成像面上对应的像素比例尺是不同的。过往在X光成像的尺寸计算中：例如发明专利CN 112149658 A计算的实际尺寸与传送带速度成正比，即传送带速度越大(越小)，计算的实际尺寸越大(越小)，然而物体的实际尺寸应该是固定值，不应与传送带速度相关，故不符合实际；该发明计算的实际尺寸是的基准物的摆放角度只考虑了0到180度，这意味基准物只摆放在传送带上，没有考虑X光检测通道的三维空间中的像素长度的畸变情况，而当物体在安检机X光检测通道的三维空间中悬空以任意角度放置时，像素比例尺与传送带表面像素比例尺差别较大，因此 计算结果与实际情况相差较大，无法满足实际需求；该发明中采用与基准物角度匹配的方式计算像素比例尺从而求得物体长度或面积，由于实际操作中只可能采有限组基准物图像，若想覆盖在标准物三维空间中六个自由度的位置和空间角度还十分不便，采用此方法必然导致结果误差较大，无法满足实际需求；该发明中只提出了对长度和面积的计算方法，没有解决对待测物品体积的计算问题。而在实际应用中，大量物体都是三维的，如酒瓶、喷雾罐等。
因此，过往在X光成像的尺寸计算结果与物品真实尺寸差距非常大，无法满足实际需求。
综上所述，为提高在立体成像空间中、不同光源组合关系下(双光源正交或非正交)利用X光图像精确计算物体尺寸的能力，都有必要建立考虑各方向图像畸变的X光成像坐标系像素比例尺计算模型，解决基于x光图像计算待检物品长度和体积的问题。
发明内容
为解决上述问题，本发明提供了一种面向双光源X射线安检机的待检物品尺寸自动检测方法。本发明可以通过图像处理的方法快速精确的得到行李中各物品的尺寸，从而提高了对危险物品的检出精度，并降低了误检率。
为达到上述技术效果，本发明的技术方案是：
一种面向双光源X射线安检机的待检物品尺寸自动检测方法，包括如下步骤：
步骤一、在X射线检查通道的相邻边上分别设置主X光成像系统和辅X光成像系统；主X光成像系统和辅X光成像系统的光源通过准直器后所成X光线处于同一平面；
步骤二、建立主X光成像系统的成像坐标系及辅X光成像系统的成像坐标系；
步骤三、采用标准件以不同的倾角、不同的高度和垂直于传送带运动方向的不同水平位置通过X射线检查通道，并得到若干组图像，每组图像包括一个主视图和一个辅视图；
步骤四、根据测得的标准件实际长、宽、高以及标准件的主视图和辅视图在x、y、z方向的像素点个数，计算出标准件在主视图上的x、y方向和辅视图上的x、z方向上对应位置的单位像素代表的实际长度，即该像素点对应的像素比例尺；将每个像素点的像素比例尺以 及像素点所在的图像坐标作为一组数据，多组数据形成数据集；其中，x方向为主视图所在平面的水平方向，即X射线检查通道传送带的运动方向，y方向为主视图所在平面内垂直于x方向的方向；z方向为辅视图所在平面内垂直于x方向的方向；
步骤五、将数据集进行拟合分别得到x，y，z三个方向的像素比例尺函数p
x(v)、p
y(y，z)和p
z(y，z)；
步骤六、将待检测物品放入X射线检查通道得到待检测物品的主视图和辅视图，并将其输入基于深度学习的物品检测模型，获得物品类别、位置、检测框和置信度；
步骤七、将待检测物品的主视图和辅视图中检测框的各像素点的坐标输入拟合后的x，y，z三个方向的像素比例尺函数，然后x，y，z方向像素点长度分别累加，得到待测物品x，y，z三个方向的实际长度，然后将待测物品x，y，z三个方向的实际长度进行融合计算得到待测物品的实际长度。
进一步的改进，所述步骤五中，将x，y，z三个方向上的数据集分别进行拟合，分别得到x，y，z三个方向的像素比例尺函数：p
x(v)、p
y(y，z)和p
z(y，z)。
进一步的改进，所述步骤七中，待检测物品的实际长度通过下式得出：
其中，l
predict表示待检测物品的实际长度，c
angle表示角度修正参数，与待检测物品摆放角度相关，c
h表示高度修正参数，与待检测物品在辅视图中的高度位置相关，l
x表示待检测物品在x方向上的实际长度，l
y表示待检测物品在y方向上的实际长度，r
y表示l
y的修正参数，与待检测物品在主视图中的位置相关，l
z表示待检测物品在z方向上的实际长度，r
z表示l
z的修正参数，该参数与待检测物品在辅视图中的位置相关； c
angle、c
h、r
y和r
z均为经验值，其中，c
angle的取值区间为[0.85，1]；c
h的取值区间为[0.9，1]；r
y的取值区间为[0.4，0.8]；r
z的取值区间为[0.75，1]。
进一步的改进，还包括步骤八、当检测到所述待检测物品为容器时，得到待检测物品的容积：
分别提取违禁品的主视图和辅视图的最小外接矩形，然后分别将违禁品主视图和辅视图按照从容器底到容器口的最短距离连线的垂直方向进行切割，切割成n个部分，每个部分近似看成一个矩形，利用步骤七所示的待测物品的长度计算方法计算出第i个矩形平行于容器底的边的长度d
i和垂直于容器底的边的长度l
i，每个矩形映射到三维空间上作为一个圆柱体，圆柱体直径是d
i，圆柱体的高是l
i，所以第i个圆柱的体积
则待检测物品在对应视图上的总体积为
按照以上方法将主视图体积V
overlook和辅视图体积V
sidelook分解计算出来；
然后进行融合计算：
V
predict＝r
overlook*V
overlook+r
sidelook*V
sidelook，
其中，其中V
predict表示待检测物品的预测体积，r
overlook为主视图体积V
overlook的贡献系数，r
sidelook为辅视图体积r
sidelook的贡献系数；r
overlook和r
sidelook均为经验值，其中r
overlook取值区间为[0，1]，r
sidelook取值区间为[0，1]。
得到待检测物的预测体积V
predict；将预测体积与预存的与待检测物品的常用容积对比，与待检测物的体积最接近的常用容积作为待检测物的容积。
进一步的改进，所述主视图为俯视图，辅视图为侧视图。
本发明的优点：
本发明可以通过图像处理的方法快速精确的得到行李中各物品的尺寸，从而提高了对危险物品的检出精度，并降低了误检率。
以下通过具体实施方式对本发明的技术方案作具体说明。
一种面向双光源X射线安检机的待检物品尺寸自动检测方法，预测违禁品长度通过以下步骤：
步骤一、在X射线检查通道的相邻边上分别设置主X光成像系统和辅X光成像系统；主X光成像系统和辅X光成像系统的光源通过准直器后所成X光线处于同一平面；
步骤二、建立主X光成像系统的成像坐标系及辅X光成像系统的成像坐标系；
步骤三、采用标准件以不同的倾角、不同的高度和垂直于传送带运动方向的不同水平位置通过X射线检查通道，并得到若干组图像，每组图像包括一个主视图和一个辅视图，其中主视图为俯视图，俯视图为侧视图；
步骤四、根据测得的标准件实际长、宽、高以及标准件的主视图和辅视图在x、y、z方向的像素点个数，计算出该次该标准件在主视图的x、y方向和辅视图的x、z方向上对应位置的单位像素代表的实际长度，即该像素点对应的像素比例尺；将每个像素的像素比例尺以及该像素所在的图像坐标作为一组数据，多组数据形成数据集；其中，x方向为主视图所在平面的水平方向，即X射线检查通道传送带的运动方向，y方向为主视图所在平面内垂直于x方向的方向；z方向为辅视图所在平面内垂直于x方向的方向；
步骤五、将数据集进行拟合分别得到x，y，z三个方向的像素比例尺函数p
x(v)、p
y(y，z)和p
z(y，z)；
步骤六、将待检测物品放入X射线检查通道得到待检测物品的主视图和辅视图；
步骤七、将待检测物品的主视图和辅视图中各像素点的坐标输入拟合后的x，y，z三个方向的像素比例尺函数，然后x，y，z方向像素点长度分别累加，得到待测物品x，y，z三个方向的实际长度，然后将待测物品x，y，z三个方向的实际长度进行融合计算得到待测物品的实际长度。
例如一把刀的真实长度为190厘米，将该刀通过安检机检查通道后得到主视图和辅视图，通过上述方法计算l
x＝177，l
y＝193，l
z＝73，根据其位置和摆放角度，c
angle＝1，c
h＝0.95，r
y＝0.5，r
z＝0.4，最后计算出l
predict＝193.5，即预测长度为193.5厘米。
当检测出容器时，需要计算容器的容积。计算容积方法如下：
将提取的容器完整图像计算出最小的外接矩形，然后将完整图像按照从容器底到容器口的最短距离连线的垂直方向进行切割，切割成n个部分，每个部分近似看成一个小的矩形，利用步骤七所示的待测物品的长度计算方法计算出第i个矩形平行于容器底的边的长度和垂直于容器底的边的长度l
i，每个矩形映射到三维空间上作为一个圆柱体，圆柱体直径是d
i，圆柱体的高是，所以第i个圆柱的体积
则待检测物品在对应视图上的总体积为
按照以上方法将主视图体积V
overlook和辅视图体积V
sidelook分解计算出来；将每个视角对应容器的图像体积计算出来后，进行融合计算，得到最终该容器的体积。例如，双视角下容器体积融合计算方法即为
V
predict＝r
overlook*V
overlook+r
sidelook*V
sidelook，
其中，其中V
predict表示待检测物品的预测体积，r
overlook为主视图体积V
overlook的贡献系数，r
sidelook为辅视图体积V
sidelook的贡献系数；r
overlook和r
sidelook均为经验值，其中r
overlook取值区间为[0.5，0.9]，r
sidelook取值区间为[0.6，0.85]。
其中V
predict表示容器融合计算之后的预测体积，r
overlook为主视图体积V
overlook的贡献系数，r
sidelook为辅视图体积V
sidelook的贡献系数，这两个参数与容器摆放的位置和角度相关。
由于市面上绝大多数常见容器容积是若干个固定值，例如50ml、100ml、500ml、1000ml等，将计算出的容器体积V
predict与市面上常见容器体积有限集合
中每一个值进行误差计算并比较，将误差最小的常见容器容积作为最后该容器体积的预测结果V
predict。例如计算出V
predict的值为513，经过比较与500ml最接近，则V
predict＝500，则预测的最终结果为500ml。
上述仅为本发明的一个具体导向实施方式，但本发明的设计构思并不局限于此，凡利用此构思对本发明进行非实质性的改动，均应属于侵犯本发明的保护范围的行为。
Claims (5)
- 一种面向双光源X射线安检机的待检物品尺寸自动检测方法，其特征在于，包括如下步骤：步骤一、在X射线检查通道的相邻边上分别设置主X光成像系统和辅X光成像系统；主X光成像系统和辅X光成像系统的光源通过准直器后所成X光线处于同一平面；步骤二、建立主X光成像系统的成像坐标系及辅X光成像系统的成像坐标系；步骤三、采用标准件以不同的倾角、不同的高度和垂直于传送带运动方向的不同水平位置通过X射线检查通道，并得到若干组图像，每组图像包括一个主视图和一个辅视图；步骤四、根据测得的标准件实际长、宽、高以及标准件的主视图和辅视图在x、y、z方向的像素点个数，计算出标准件在主视图上的x、y方向和辅视图上的x、z方向上对应位置的单位像素代表的实际长度；将每个像素点的像素比例尺以及像素点所在的图像坐标作为一组数据，多组数据形成数据集；其中，x方向为主视图所在平面的水平方向，y方向为主视图所在平面内垂直于x方向的方向；z方向为辅视图所在平面内垂直于x方向的方向；步骤五、将数据集进行拟合分别得到x，y，z三个方向的像素比例尺函数p x(v)、p y(y，z)和p z(y，z)；步骤六、将待检测物品放入X射线检查通道得到待检测物品的主视图和辅视图，并将其输入基于深度学习的物品检测模型，获得物品类别、位置、检测框和置信度；步骤七、将待检测物品的主视图和辅视图中检测框的各像素点的坐标输入拟合后的x，y，z三个方向的像素比例尺函数，然后x，y，z方向像素点所表示的实际长度分别累加，得到待测物品x，y，z三个方向的实际长度，然后将待测物品x，y，z三个方向的实际长度进行融合计算得到待测物品的实际长度。
- 如权利要求1所述的面向双光源X射线安检机的待检物品尺寸自动检测方法，其特征在于，所述步骤五中，将x，y，z三个方向上的数据集分别进行拟合，分别得到x，y，z三个方向的像素比例尺函数：p x(v)、p y(y，z)和p z(y，z)。
- 如权利要求2所述的面向双光源X射线安检机的待检物品尺寸自动检测方法，其特征在于， 所述步骤七中，待检测物品的实际长度通过下式得出：
其中，l predict表示待检测物品的实际长度，c angle表示角度修正参数，与待检测物品摆放角度相关，c h表示高度修正参数，与待检测物品在辅视图中的高度位置相关，l x表示待检测物品在x方向上的实际长度，l y表示待检测物品在y方向上的实际长度，r y表示l y的修正参数，与待检测物品在主视图中的位置相关，l z表示待检测物品在z方向上的实际长度，r z表示l z的修正参数，该参数与待检测物品在辅视图中的位置相关；c angle、c h、r y和r z均为经验值，其中，c angle的取值区间为[0.85，1]；c h的取值区间为[0.9，1]；r y的取值区间为[0.4，0.8]；r z的取值区间为[0.75，1]。 - 如权利要求1所述的面向双光源X射线安检机的待检物品尺寸自动检测方法，其特征在于，还包括步骤八、当检测到所述待检测物品为容器时，得到待检测物品的容积：分别提取违禁品的主视图和辅视图的最小外接矩形，然后分别将违禁品主视图和辅视图按照从容器底到容器口的最短距离连线的垂直方向进行切割，切割成n个部分，每个部分近似看成一个矩形，利用步骤七所示的待测物品的长度计算方法计算出第i个矩形平行于容器底的边的长度d i和垂直于容器底的边的长度l i，每个矩形映射到三维空间上作为一个圆柱体，圆柱体直径是d i，圆柱体的高是l i，所以第i个圆柱的体积则待检测物品在对应视图上的总体积为
按照以上方法将主视图体积 V overlook和辅视图体积V sidelook分解计算出来；
然后进行融合计算：V predict＝r overlook*V overlook+r sidelook*V sidelook，其中，其中V predict表示待检测物品的预测体积，r overlook为主视图体积V overlook的贡献系数，r sidelook为辅视图体积V sidelook的贡献系数；r overlook和r sidelook均为经验值，其中r overlook取值区间为[0，1]，r sidelook取值区间为[0，1]；得到待检测物的预测体积V predict；将预测体积与预存的与待检测物品的常用容积对比，与待检测物的体积最接近的常用容积作为待检测物的容积。 - 如权利要求1所述的面向双光源X射线安检机的待检物品尺寸自动检测方法，其特征在于，所述主视图为俯视图，辅视图为侧视图。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111359883.1A CN113790685B (zh) | 2021-11-17 | 2021-11-17 | 面向双光源x射线安检机的待检物品尺寸自动检测方法 |
| CN202111359883.1 | 2021-11-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023087652A1 true WO2023087652A1 (zh) | 2023-05-25 |
Family
ID=78877304
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2022/094767 WO2023087652A1 (zh) | 2021-11-17 | 2022-05-24 | 面向双光源x射线安检机的待检物品尺寸自动检测方法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN113790685B (zh) |
| WO (1) | WO2023087652A1 (zh) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113790685B (zh) * | 2021-11-17 | 2022-01-18 | 湖南苏科智能科技有限公司 | 面向双光源x射线安检机的待检物品尺寸自动检测方法 |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4736399A (en) * | 1984-08-08 | 1988-04-05 | Kabushiki Kaisha Toshiba | X-ray imaging apparatus |
| JP2004275362A (ja) * | 2003-03-14 | 2004-10-07 | Fuji Photo Film Co Ltd | 画像実寸算出方法、画像計測方法および画像出力方法、ならびにそれらの装置 |
| CN102928448A (zh) * | 2012-10-30 | 2013-02-13 | 公安部第一研究所 | 通道式四视角x射线液态物品安全检查方法及检查装置 |
| CN104483711A (zh) * | 2014-12-17 | 2015-04-01 | 同方威视技术股份有限公司 | 基于分布式光源的辐射成像系统 |
| JP2016024096A (ja) * | 2014-07-22 | 2016-02-08 | 株式会社イシダ | 検査装置 |
| CN108267465A (zh) * | 2016-12-29 | 2018-07-10 | 同方威视技术股份有限公司 | 多视角成像数据处理方法和设备 |
| CN109374658A (zh) * | 2018-12-24 | 2019-02-22 | 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 |
| CN113790685A (zh) * | 2021-11-17 | 2021-12-14 | 湖南苏科智能科技有限公司 | 面向双光源x射线安检机的待检物品尺寸自动检测方法 |
| CN114062934A (zh) * | 2021-11-19 | 2022-02-18 | 湖南苏科智能科技有限公司 | 一种基于机器视觉的充电电池容量检测方法 |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101644686B (zh) * | 2008-08-08 | 2013-07-24 | 李海洋 | 一种爆炸物及毒品的联合在线检测仪及其应用 |
| EP2505618A1 (en) * | 2011-04-01 | 2012-10-03 | Fábrica Nacional De Moneda Y Timbre | Use of electromagnetic wave absorbing markers for the aunthentication of security documents |
| CN103292701B (zh) * | 2013-06-24 | 2015-09-16 | 哈尔滨工业大学 | 基于机器视觉的精密器件在线尺寸测量方法 |
| CN106127891A (zh) * | 2016-06-21 | 2016-11-16 | 成都锦汇科技有限公司 | 一种行李自助托运的方法 |
| CN207636789U (zh) * | 2017-12-05 | 2018-07-20 | 伟杰科技（苏州）有限公司 | 一种用于安检的x光机 |
| CN208188368U (zh) * | 2018-05-22 | 2018-12-04 | 武汉艾崴科技有限公司 | 具有智能x射线曲线修正与可变通道尺寸的智慧安检机 |
| CN110133740A (zh) * | 2019-04-29 | 2019-08-16 | 公安部第一研究所 | 一种双视角智能化x射线安检设备 |
| CN110889807B (zh) * | 2019-11-20 | 2023-04-07 | 公安部第一研究所 | 一种通道式x射线安检设备图像处理方法 |
| CN112893172A (zh) * | 2021-01-18 | 2021-06-04 | 菲特（天津）检测技术有限公司 | 基于机器视觉的垫片尺寸检测系统、方法、处理终端及介质 |
| CN113052918B (zh) * | 2021-04-23 | 2024-06-04 | 北京机械设备研究所 | 对极式双目相机标定误差评价方法、装置、介质及设备 |
-
2021
- 2021-11-17 CN CN202111359883.1A patent/CN113790685B/zh active Active
-
2022
- 2022-05-24 WO PCT/CN2022/094767 patent/WO2023087652A1/zh unknown
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4736399A (en) * | 1984-08-08 | 1988-04-05 | Kabushiki Kaisha Toshiba | X-ray imaging apparatus |
| JP2004275362A (ja) * | 2003-03-14 | 2004-10-07 | Fuji Photo Film Co Ltd | 画像実寸算出方法、画像計測方法および画像出力方法、ならびにそれらの装置 |
| CN102928448A (zh) * | 2012-10-30 | 2013-02-13 | 公安部第一研究所 | 通道式四视角x射线液态物品安全检查方法及检查装置 |
| JP2016024096A (ja) * | 2014-07-22 | 2016-02-08 | 株式会社イシダ | 検査装置 |
| CN104483711A (zh) * | 2014-12-17 | 2015-04-01 | 同方威视技术股份有限公司 | 基于分布式光源的辐射成像系统 |
| CN108267465A (zh) * | 2016-12-29 | 2018-07-10 | 同方威视技术股份有限公司 | 多视角成像数据处理方法和设备 |
| CN109374658A (zh) * | 2018-12-24 | 2019-02-22 | 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 |
| CN113790685A (zh) * | 2021-11-17 | 2021-12-14 | 湖南苏科智能科技有限公司 | 面向双光源x射线安检机的待检物品尺寸自动检测方法 |
| CN114062934A (zh) * | 2021-11-19 | 2022-02-18 | 湖南苏科智能科技有限公司 | 一种基于机器视觉的充电电池容量检测方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113790685A (zh) | 2021-12-14 |
| CN113790685B (zh) | 2022-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10290115B2 (en) | Device and method for determining the volume of an object moved by an industrial truck | |
| CN110017773A (zh) | 一种基于机器视觉的包裹体积测量方法 | |
| CN108474644B (zh) | 体积估计方法、设备和系统 | |
| CN105674880B (zh) | 基于双目原理的接触网几何参数测量方法及系统 | |
| CN110032949A (zh) | 一种基于轻量化卷积神经网络的目标检测与定位方法 | |
| US10769802B2 (en) | Indoor distance measurement method | |
| US20180108143A1 (en) | Height measuring system and method | |
| CN104567758B (zh) | 立体成像系统及其方法 | |
| CN108898047A (zh) | 基于分块遮挡感知的行人检测方法及系统 | |
| Boltes et al. | T-junction: Experiments, trajectory collection, and analysis | |
| CN111079589B (zh) | 基于深度摄像和身高阈值像素标定的自动身高检测方法 | |
| CN102799867A (zh) | 基于图像处理的仪表指针转角识别方法 | |
| WO2023087652A1 (zh) | 面向双光源x射线安检机的待检物品尺寸自动检测方法 | |
| CN106340044A (zh) | 摄像机外参自动标定方法及标定装置 | |
| CN106373140B (zh) | 一种基于单目视觉的透明及半透明液体杂质检测方法 | |
| CN103439348A (zh) | 基于差影法的遥控器按键缺陷检测方法 | |
| CN104613869B (zh) | 一种基于最大内切圆的椭圆孔组检测方法和系统 | |
| CN113960075A (zh) | 一种安检设备及物品尺寸确定方法、存储介质、设备 | |
| CN114898044B (zh) | 检测对象成像方法、装置、设备及介质 | |
| CN114624263A (zh) | 一种双源双视角基于目标识别的切图方法及系统 | |
| CN114660097B (zh) | 一种基于双源双视角的同步校正方法及系统 | |
| CN106276285B (zh) | 组料垛位自动检测方法 | |
| CN109626302A (zh) | 一种啤酒瓶内异物的在线检测方法 | |
| CN206281468U (zh) | 一种柱状物体垂直度的非接触式检测装置 | |
| Wang et al. | Dynamic inspection of rail wear via a three-step method: Auxiliary plane establishment, self-calibration, and projecting |