CN110647925A - Rigid object identification method and device based on improved LINE-MOD template matching - Google Patents

Rigid object identification method and device based on improved LINE-MOD template matching Download PDF

Info

Publication number
CN110647925A
CN110647925A CN201910842182.XA CN201910842182A CN110647925A CN 110647925 A CN110647925 A CN 110647925A CN 201910842182 A CN201910842182 A CN 201910842182A CN 110647925 A CN110647925 A CN 110647925A
Authority
CN
China
Prior art keywords
image
rigid object
gradient direction
line
template matching
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.)
Pending
Application number
CN201910842182.XA
Other languages
Chinese (zh)
Inventor
王月
范志鹏
罗志勇
夏文彬
帅昊
唐文平
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.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
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 Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN201910842182.XA priority Critical patent/CN110647925A/en
Publication of CN110647925A publication Critical patent/CN110647925A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Abstract

The invention discloses a rigid object identification method and device based on improved LINE-MOD template matching, and relates to the field of identification of objects with few textures or no textures. The current method based on the feature descriptor is easily influenced by a complex background when the feature descriptor is calculated for the target object, and the method based on the deep learning has the defects of complex structure and training process and the like. Aiming at the problems, the method is an effective method for solving the recognition of the rigid object with few textures, and has good real-time performance, and meanwhile, the LINE-MOD template matching method is one of the most advanced template matching methods in recent years. However, the method has the main disadvantages that only rigid objects with few textures can be identified in a fixed scale, and how to improve multi-scale identification of the algorithm is still to be researched. Therefore, the invention provides a rigid object identification method and device based on improved LINE-MOD template matching, so that the method can identify the target object under different scales.

Description

Rigid object identification method and device based on improved LINE-MOD template matching
Technical Field
The invention belongs to the field of computer vision processing, and particularly relates to a rigid object identification method and device based on improved LINE-MOD template matching.
Background
The template matching method is mainly used for training a built target object model, then sliding matching is carried out on an acquired image by using a template image as a sliding window, and the matching position can be found out on the basis of a similarity measurement method at the best matching position. The method has strong processing capability for identifying the rigid target object with few textures, does not need a huge training data set or a time-consuming training stage, and has good expression effect on the rigid object with few textures or no textures. Early template matching methods and extensions thereof used Chamfer distance to measure the difference between the template and the test image contours. For example, Gavrila proposes a coarse-to-fine binary edge image Chamfer distance metric over shape and parameter space. The generalized distance between two groups of edge points can be reduced to the maximum extent by the Chamfer matching, but the method is extremely sensitive to outer points caused by occlusion and the like. Another distance metric for binary images is the Hausdorff distance, which measures the maximum of all distances from each edge point in the image to the nearest neighbor in the template, but is susceptible to occlusion and complex backgrounds. Huttenlocher et al tried to overcome this drawback by introducing a generalized Hausdorff distance, which to some extent overcome the effects of occlusion and cluttered background, but which required a priori estimation of the background clutter, and moreover, when there were many templates, the computation cost was high. The binary images used by these methods are mostly obtained by Canny-like edge extraction algorithms, and thus they are extremely sensitive to illumination variations, noise and blur. In order to avoid the defects of the algorithm, Hinterstoisser and the like propose a LINE-MOD method using image gradient rather than image contour as a matching feature, and the method adopts a binary mode representation of gradient direction features, and skillfully utilizes the cache of a modern computer for parallel processing. The method can detect multiple types of rigid objects with few textures in real time in a complex background environment. However, this method can only identify rigid objects at a fixed scale.
The invention provides a rigid object identification method and device based on improved LINE-MOD template matching, which considers that the LINE-MOD template matching method can only identify rigid objects with few textures at a fixed scale and how to improve algorithm multi-scale identification on the basis of in-depth research on identification methods of target objects with few textures or no textures, so that the method can identify the target objects at different scales.
Disclosure of Invention
The present invention is directed to solving the above problems of the prior art. A rigid object identification method and device based on improved LINE-MOD template matching are provided. The technical scheme of the invention is as follows:
a rigid object identification method and device based on improved LINE-MOD template matching comprises the following steps:
step 1, image preprocessing: acquiring a rigid object image on line, and carrying out smooth denoising and sharpening on the rigid object image by using Gaussian filtering and a Laplace operator in sequence to remove noise of the rigid object image and keep edge information of the image;
step 2, rigid object feature offline extraction: in the off-line stage, a rigid object model is trained in a CAD environment, image acquisition under multiple viewing angles is carried out on the rigid object model and is used as a reference image, and gradient direction descriptors are extracted on the obtained reference image and are stored in an XML file; providing an image template for the online identification of the rigid object in the step 3;
step 3, rigid object online identification: acquiring video frames of a rigid object in a real scene at an online stage, performing image preprocessing and image gradient direction descriptor extraction on each frame of image, extracting image gradient direction descriptors on the basis of the step 1, and matching the image gradient direction descriptors extracted at the online stage with the gradient direction descriptors extracted at an offline stage by an improved LINE-MOD template matching method.
Further, the step 1 successively performs smoothing, denoising and sharpening on the rigid object and the image by using gaussian filtering and laplacian operator, and specifically includes:
assuming that the input image is f (x, y) and the image obtained after the gaussian filtering is g (x, y), the transformation relationship between the two is shown as the following formula:
Figure BDA0002194068280000021
in the formula
Figure BDA0002194068280000031
For convolution of the image, x and y represent the x and y coordinates in the image coordinate system. h (x, y) is a Gaussian template that can be represented as follows:
wherein, x and y represent x and y coordinates in an image coordinate system, sigma is a standard deviation of a Gaussian function, and can describe the discrete degree of data, and when the sigma value is larger, the smoothing effect on the image is obvious; on the contrary, the sigma value is smaller, and the image edge information is protected. The gaussian filtered image g (x, y) removes some noise, but appears blurred relative to the input image f (x, y), with the degree of blurring depending on the value of σ. After obtaining the image g (x, y), it is necessary to further perform laplace transform on g (x, y) as shown in the following formula:
wherein x and y represent x and y coordinates +in the image coordinate system2g (x, y) is the image after Laplacian processing, h (x, y) is a Gaussian template, f (x, y) is an input image ^ h2h (x, y) is the second derivative of the gaussian operator.
Further, the specific process of acquiring the multi-view rigid object image in the step 2 is as follows:
the method comprises the steps that a virtual camera in a CAD environment is used for carrying out image acquisition on a CAD model of a rigid object, projection images of the model under different visual angles are acquired, each image contains different shapes of the outline of the rigid object, sampling is carried out on the surface of a regular icosahedron in order to realize uniform distribution of sampling visual angles and avoid the appearance of sampling polar points on the surface of a hemisphere, each surface of the regular icosahedron is divided into four parts on average and is subjected to iterative division, each surface of the regular icosahedron is only subjected to iteration for 2 times, and meanwhile, 16 equilateral triangles are formed on each surface; in the CAD environment, a CAD model of a rigid object is placed in the center of a regular icosahedron, the optical axis of a virtual camera always passes through the center, shape feature descriptors of the rigid object outline in all views are extracted, and the shape feature descriptors of all the rigid object outline and corresponding pose parameters are stored in an XML file.
Further, the step 2) of extracting the gradient direction descriptor of the acquired reference image specifically includes:
for each sampled image, calculating the gradient direction on three color channels at each position { R, G, B } of the rigid object outline, and taking the maximum value of the gradient direction in the three channels as the gradient direction at the point, for one input image I, the gradient direction at position x is shown in the following formula:
Figure BDA0002194068280000041
where C ∈ { R, G, B } is the three channels red, green, blue in the image at position x in view I, Ig(x)、Respectively represent the position x in the view IIs the gradient direction at position x in view I, C is a color channel on the three color channels R, G, B, and ori () represents the image gradient direction.
Further, in the step 3, the gradient direction descriptor of the image extracted on LINE is matched with the gradient direction descriptor extracted in the off-LINE stage by an improved LINE-MOD template matching method, and a specific similarity measurement algorithm is as follows:
Figure BDA0002194068280000043
where epsilon represents the similarity between the target object to be recognized and the target object training template obtained in the off-line training stage, I represents the input image, T ═ O, p represents the template of the target object, c represents the position c of the input image and the template image at the same position, ori (O, r) represents the radian of the gradient direction at the position r of the reference image O,
Figure BDA0002194068280000044
denotes a region centered at c + r, τ is the neighborhood, ori (I, t) denotes the radian of the gradient direction at the t point of the input image I position, and p denotes a list of positions r.
Further, in order to complete the identification of the target object under any scale, depth information is introduced into the similarity measurement method, so that the depth information has scale invariance, and a specific algorithm is as follows:
where ∈ denotes the similarity between the target object to be recognized and the target object training template obtained in the offline training stage, I denotes the input image, T ═ O, p denotes the template of the target object, and c'IIndicating that the input image and the template image are at the same position c'IWhere O denotes the position of the reference image O, So(c′oR') denotes the gradient direction of the reference image, ori (O, S)o(c′oR')) is represented in a reference pictureO position
Figure BDA0002194068280000051
In the direction of the gradient at a point ori (I, t) denotes the direction of the gradient at the point t of the input image I, p denotes a list of positions r, y denotes the distance of the image gradient along the y-axis, D (c'x) Is indicated at point c'xDepth value of (D'o) Representing the distance from the reference image O camera to the center of the regular polyhedron in offline training;
Figure BDA0002194068280000052
is shown in
Figure BDA0002194068280000053
As the center, τ is the region of the neighborhood.
A rigid object identification method and device based on improved LINE-MOD template matching comprises the following steps:
an image preprocessing module: acquiring a rigid object image on line, and carrying out smooth denoising and sharpening on the rigid object image by using Gaussian filtering and a Laplace operator in sequence to remove noise of the rigid object image and keep edge information of the image;
rigid object feature offline extraction module: in the off-line stage, a rigid object model is trained in a CAD environment, image acquisition under multiple viewing angles is carried out on the rigid object model and is used as a reference image, and gradient direction descriptors are extracted on the obtained reference image and are stored in an XML file; providing an image template for online identification of the rigid object in the rigid object online identification module;
rigid object online identification module: the method comprises the steps of collecting video frames of a rigid object in a real scene at an online stage, conducting image preprocessing and image gradient direction descriptor extraction on each frame of image, matching the image gradient direction descriptor extracted on LINE with the gradient direction descriptor extracted at an offline stage through an improved LINE-MOD template matching method, and improving the improved LINE-MOD template matching method mainly in that depth information is introduced into a similarity measurement algorithm, so that the improved LINE-MOD template matching method can finish rigid object identification at any scale. And completing the identification of the rigid object.
The invention has the following advantages and beneficial effects:
the innovation point of the invention is to provide an improved LINE-MOD template matching rigid object recognition algorithm, which adopts a binary system mode of gradient direction characteristics to express and skillfully uses the cache of a modern computer to carry out parallel processing. The method can identify multiple types of non-texture rigid objects in real time in a complex background environment. However, the method can only identify the rigid object at a fixed scale, so that the method can identify the target object at different scales in order to improve the multi-scale identification of the algorithm.
The method comprises the following specific steps:
(1) and (3) image preprocessing, namely performing smooth denoising and sharpening on the image by using Gaussian filtering and a Laplacian operator in sequence, so that the noise of the image is effectively removed, and the edge information of the image can be better kept.
(2) The method comprises the steps of rigid object feature offline extraction, rigid object model training in a CAD environment in an offline stage, image acquisition under multiple visual angles of the model and serving as a reference image, gradient direction descriptor extraction of the acquired reference image and storage of the extracted gradient direction descriptor in an XML file.
(3) The method comprises the steps of identifying a rigid object on LINE, collecting video frames of the rigid object in a real scene at an on-LINE stage, carrying out image preprocessing and image gradient direction descriptor extraction on each frame of image, and then matching the image gradient direction descriptor extracted on LINE with the gradient direction descriptor extracted at an off-LINE stage through an improved LINE-MOD template matching method. If the image acquired during the off-line phase to generate the template is far away from the camera and the image acquired during the on-line operation is close to the camera, even if the object in the on-line acquired image and the object in the template have the same gradient direction, the similarity measurement algorithm in step 3 still shows a mismatch. In order to complete the recognition of the object to be assembled under any scale, depth information is introduced into a similarity evaluation method, so that the object to be assembled has scale invariance.
(4) In order to complete the identification of the target object under any scale, the improved LINE-MOD template matching rigid object identification algorithm introduces depth information into a similarity measurement method to enable the depth information to have scale invariance.
The method has the advantages that: the operation process is simple, and the method has strong capability of processing different objects. It does not require a large training set nor a time-consuming training phase and can be handled for both textured and non-textured rigid objects.
Drawings
FIG. 1 is a flow chart of a rigid object recognition algorithm based on improved LINE-MOD template matching according to a preferred embodiment of the present invention;
fig. 2 is a flowchart of the off-line extraction of the rigid object features according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described in detail and clearly with reference to the accompanying drawings. The described embodiments are only some of the embodiments of the present invention.
The technical scheme for solving the technical problems is as follows:
the invention provides an efficient domain ontology semantic similarity calculation algorithm, and on the basis of in-depth research on a method for identifying a few-texture or non-texture target object, the LINE-MOD template matching method is considered to be capable of identifying a few-texture rigid object only at a fixed scale, and how to improve multi-scale identification of the algorithm is mainly solved through the step 3, so that the method can identify the target object at different scales. The invention will be described in further detail below with reference to the accompanying drawings and examples.
As shown in fig. 1, the flow chart of the rigid object recognition algorithm based on the improved LINE-MOD template matching of the present invention is specifically implemented as follows:
1. and (3) image preprocessing, namely performing smooth denoising and sharpening on the image by using Gaussian filtering and a Laplacian operator in sequence, so that the noise of the image is effectively removed, and the edge information of the image can be better kept.
2. The method comprises the steps of rigid object feature offline extraction, rigid object model training in a CAD environment in an offline stage, image acquisition under multiple visual angles of the model and serving as a reference image, gradient direction descriptor extraction of the acquired reference image and storage of the extracted gradient direction descriptor in an XML file.
3. The method comprises the following steps of identifying a rigid object on LINE, collecting video frames of the rigid object in a real scene at an on-LINE stage, carrying out image preprocessing and image gradient direction descriptor extraction on each frame of image, and matching the image gradient direction descriptor extracted on LINE with a gradient direction descriptor extracted at an off-LINE stage by an improved LINE-MOD template matching method, wherein a specific similarity measurement algorithm is as follows:
Figure BDA0002194068280000071
where epsilon represents the similarity between the target object to be recognized and the target object training template obtained in the off-line training stage, I represents the input image, T ═ O, p represents the template of the target object, c represents the position c of the input image and the template image at the same position, ori (O, r) represents the radian of the gradient direction at the position r of the reference image O,
Figure BDA0002194068280000081
denotes a region centered at c + r, τ is the neighborhood, ori (I, t) denotes the radian of the gradient direction at the t point of the input image I position, and p denotes a list of positions r.
4. In order to complete the identification of a target object under any scale, the improved LINE-MOD template matching rigid object identification algorithm introduces depth information into a similarity measurement method to enable the depth information to have scale invariance, and the specific algorithm is as follows:
in the formula, epsilon represents the target object to be recognized and the target obtained in the off-line training stageSimilarity between object training templates, I denotes an input image, T ═ O, p denotes a template of a target object, c'IIndicating that the input image and the template image are at the same position c'IWhere O denotes the position of the reference image O, So(c′oR') denotes the gradient direction of the reference image, ori (O, S)o(c′oR')) is indicated at the reference image O position
Figure BDA0002194068280000083
In the direction of the gradient at a point ori (I, t) denotes the direction of the gradient at the point t of the input image I, p denotes a list of positions r, y denotes the distance of the image gradient along the y-axis, D (c'x) Is indicated at point c'xDepth value of (D'o) And the distance from the reference image O camera to the center of the regular polyhedron in off-line training is shown.
Figure BDA0002194068280000084
Is shown in
Figure BDA0002194068280000085
As the center, τ is the region of the neighborhood.
Fig. 2 is a flow chart of off-LINE rigid object feature extraction according to the present invention, and fig. 2 is a further more detailed description of the rigid object recognition algorithm process based on the improved LINE-MOD template matching in fig. 1. The method is realized by the following steps:
1. multi-view rigid object image acquisition
Image acquisition of a CAD model of a rigid object by a virtual camera in a CAD environment requires acquisition of projection images of the model at different viewing angles, and each image contains a different shape of the rigid object's outline. In order to realize the uniform distribution of sampling visual angles and avoid the problem of image redundancy caused by overlarge sampling density of sampling polar points on the surface of a hemisphere, the sampling is carried out on the surface of a regular icosahedron, and each surface of the regular icosahedron is mainly divided into four parts on average and is subjected to iterative division. To take into account the speed and accuracy of the operation in image acquisition, only 2 iterations were chosen for each face of the regular icosahedron, while allowing 16 equilateral triangles to be formed on each face. In a CAD environment, a CAD model of a rigid object is placed in the center of a regular icosahedron, with the virtual camera optical axis passing through the center at all times. Shape feature descriptors of the rigid object contours in all views are extracted, and the shape feature descriptors and corresponding pose parameters of all the rigid object contours are saved in an XML file.
2. Gradient direction extraction
For each sampled image, the gradient direction on the three color channels for each position { R, G, B } of the rigid object holder contour is calculated, and the gradient direction maximum in the three channels is taken as the gradient direction at that point. For example, in an input image I, the gradient direction at the position x is shown by the following formula:
Figure BDA0002194068280000091
where C ∈ { R, G, B } is the three channels of red, green, and blue in the image at position x in view I, C is a color channel of the three color channels of R, G, and B, and ori () represents the image gradient direction.
3. Extension in gradient direction
Because the input image may generate a small movement or rotation, the gradient direction of the rigid object extracted in the online matching process is directly mismatched with the gradient direction of the rigid object extracted offline, and thus the identification of the rigid object is affected. In order to improve robustness of micro movement or rotation of an input image in an online matching process, after extracting a profile gradient of a rigid object in an offline stage, expansion processing needs to be performed in each gradient direction. The gradient direction of a certain position x on the surface of the rigid object is expanded, namely the gradient direction of the position x is adopted to represent in a 3 x 3 pixel coordinate neighborhood taking the current position as the center.
The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure. After reading the description of the invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.

Claims (7)

1. A rigid object identification method and device based on improved LINE-MOD template matching are characterized by comprising the following steps:
step 1, image preprocessing: acquiring a rigid object image on line, and carrying out smooth denoising and sharpening on the rigid object image by using Gaussian filtering and a Laplace operator in sequence to remove noise of the rigid object image and keep edge information of the image;
step 2, rigid object feature offline extraction: in the off-line stage, a rigid object model is trained in a CAD environment, image acquisition under multiple viewing angles is carried out on the rigid object model and is used as a reference image, and gradient direction descriptors are extracted on the obtained reference image and are stored in an XML file; providing an image template for the online identification of the rigid object in the step 3;
step 3, rigid object online identification: acquiring video frames of a rigid object in a real scene at an online stage, performing image preprocessing and image gradient direction descriptor extraction on each frame of image, extracting image gradient direction descriptors on the basis of the step 1, and matching the image gradient direction descriptors extracted at the online stage with the gradient direction descriptors extracted at an offline stage by an improved LINE-MOD template matching method.
2. The method and device for identifying the rigid object based on the improved LINE-MOD template matching according to claim 1, wherein the step 1 performs smoothing, denoising and sharpening on the rigid object and the image by using gaussian filtering and laplacian successively, and specifically comprises:
assuming that the input image is f (x, y) and the image obtained after the gaussian filtering is g (x, y), the transformation relationship between the two is shown as the following formula:
Figure FDA0002194068270000011
in the formula
Figure FDA0002194068270000012
For convolution of the image, x and y represent the x and y coordinates in the image coordinate system. h (x, y) is a Gaussian template that can be represented as follows:
Figure FDA0002194068270000013
wherein, x and y represent x and y coordinates in an image coordinate system, sigma is a standard deviation of a Gaussian function, and can describe the discrete degree of data, and when the sigma value is larger, the smoothing effect on the image is obvious; on the contrary, the sigma value is smaller, and the image edge information is protected. The gaussian filtered image g (x, y) removes some noise, but appears blurred relative to the input image f (x, y), with the degree of blurring depending on the value of σ. After obtaining the image g (x, y), it is necessary to further perform laplace transform on g (x, y) as shown in the following formula:
Figure FDA0002194068270000021
wherein x and y represent x and y coordinates +in the image coordinate system2g (x, y) is the image after Laplacian processing, h (x, y) is a Gaussian template, f (x, y) is an input image ^ h2h (x, y) is the second derivative of the gaussian operator.
3. The rigid object identification method and device based on improved LINE-MOD template matching according to claim 1, wherein the specific process of multi-view rigid object image acquisition in the step 2 is as follows:
the method comprises the steps that a virtual camera in a CAD environment is used for carrying out image acquisition on a CAD model of a rigid object, projection images of the model under different visual angles are acquired, each image contains different shapes of the outline of the rigid object, sampling is carried out on the surface of a regular icosahedron in order to realize uniform distribution of sampling visual angles and avoid the appearance of sampling polar points on the surface of a hemisphere, each surface of the regular icosahedron is divided into four parts on average and is subjected to iterative division, each surface of the regular icosahedron is only subjected to iteration for 2 times, and meanwhile, 16 equilateral triangles are formed on each surface; in the CAD environment, a CAD model of a rigid object is placed in the center of a regular icosahedron, the optical axis of a virtual camera always passes through the center, shape feature descriptors of the rigid object outline in all views are extracted, and the shape feature descriptors of all the rigid object outline and corresponding pose parameters are stored in an XML file.
4. The method and device for rigid object identification based on improved LINE-MOD template matching according to claim 3, wherein the 2) performing gradient direction descriptor extraction on the acquired reference image specifically comprises:
for each sampled image, calculating the gradient direction on three color channels at each position { R, G, B } of the rigid object outline, and taking the maximum value of the gradient direction in the three channels as the gradient direction at the point, for one input image I, the gradient direction at position x is shown in the following formula:
Figure FDA0002194068270000031
where C ∈ { R, G, B } is the three channels red, green, blue in the image at position x in view I, Ig(x)、
Figure FDA0002194068270000032
Respectively represent the gradient direction at the position x in the view I, the maximum value of the gradient direction in the three channels is taken as the gradient direction at the position x in the view I, C is a certain color channel among the three color channels R, G, B, and ori () represents the image gradient direction.
5. The method and device for rigid object identification based on improved LINE-MOD template matching as claimed in claim 4, wherein the step 3 matches the gradient direction descriptor of the image extracted on-LINE with the gradient direction descriptor extracted at off-LINE stage by the improved LINE-MOD template matching method, and the specific similarity measure algorithm is as follows:
Figure FDA0002194068270000033
where epsilon represents the similarity between the target object to be recognized and the target object training template obtained in the off-line training stage, I represents the input image, T ═ O, p represents the template of the target object, c represents the position c of the input image and the template image at the same position, ori (O, r) represents the radian of the gradient direction at the position r of the reference image O,
Figure FDA0002194068270000034
denotes a region centered at c + r, τ is the neighborhood, ori (I, t) denotes the radian of the gradient direction at the t point of the input image I position, and p denotes a list of positions r.
6. The rigid object identification method and device based on improved LINE-MOD template matching according to claim 5, wherein in order to complete the identification of the target object at any scale, depth information is introduced into the similarity measurement method to make it have scale invariance, and the specific algorithm is as follows:
Figure FDA0002194068270000035
where ∈ denotes the similarity between the target object to be recognized and the target object training template obtained in the offline training stage, I denotes the input image, T ═ O, p denotes the template of the target object, and c'IIndicating that the input image and the template image are at the same position c'IWhere O denotes the position of the reference image O, So(c′oR') denotes the gradient direction of the reference image, ori (O, S)o(c′oR')) is indicated at the reference image O position
Figure FDA0002194068270000041
In the direction of the gradient at a point ori (I, t) denotes the direction of the gradient at the point t of the input image I, p denotes a list of positions r, y denotes the distance of the image gradient along the y-axis, D (c'x) Is indicated at point c'xDepth value of (D'o) Representing the distance from the reference image O camera to the center of the regular polyhedron in offline training;
Figure FDA0002194068270000042
is shown in
Figure FDA0002194068270000043
As the center, τ is the region of the neighborhood.
7. A rigid object identification method and device based on improved LINE-MOD template matching are characterized by comprising the following steps:
an image preprocessing module: acquiring a rigid object image on line, and carrying out smooth denoising and sharpening on the rigid object image by using Gaussian filtering and a Laplace operator in sequence to remove noise of the rigid object image and keep edge information of the image;
rigid object feature offline extraction module: in the off-line stage, a rigid object model is trained in a CAD environment, image acquisition under multiple viewing angles is carried out on the rigid object model and is used as a reference image, and gradient direction descriptors are extracted on the obtained reference image and are stored in an XML file; providing an image template for online identification of the rigid object in the rigid object online identification module;
rigid object online identification module: the method comprises the steps of collecting video frames of a rigid object in a real scene at an online stage, conducting image preprocessing and image gradient direction descriptor extraction on each frame of image, matching the image gradient direction descriptor extracted on LINE with the gradient direction descriptor extracted at an offline stage through an improved LINE-MOD template matching method, and improving the improved LINE-MOD template matching method mainly in that depth information is introduced into a similarity measurement algorithm, so that the improved LINE-MOD template matching method can finish rigid object identification at any scale. And completing the identification of the rigid object.
CN201910842182.XA 2019-09-06 2019-09-06 Rigid object identification method and device based on improved LINE-MOD template matching Pending CN110647925A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910842182.XA CN110647925A (en) 2019-09-06 2019-09-06 Rigid object identification method and device based on improved LINE-MOD template matching

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910842182.XA CN110647925A (en) 2019-09-06 2019-09-06 Rigid object identification method and device based on improved LINE-MOD template matching

Publications (1)

Publication Number Publication Date
CN110647925A true CN110647925A (en) 2020-01-03

Family

ID=68991657

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910842182.XA Pending CN110647925A (en) 2019-09-06 2019-09-06 Rigid object identification method and device based on improved LINE-MOD template matching

Country Status (1)

Country Link
CN (1) CN110647925A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111738320A (en) * 2020-03-04 2020-10-02 沈阳工业大学 Shielded workpiece identification method based on template matching
CN111814580A (en) * 2020-06-15 2020-10-23 南京澳讯人工智能研究院有限公司 Target effective identification device and method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6606406B1 (en) * 2000-05-04 2003-08-12 Microsoft Corporation System and method for progressive stereo matching of digital images
US8411080B1 (en) * 2008-06-26 2013-04-02 Disney Enterprises, Inc. Apparatus and method for editing three dimensional objects
CN107942949A (en) * 2017-03-31 2018-04-20 沈机（上海）智能系统研发设计有限公司 A kind of lathe vision positioning method and system, lathe
CN109636854A (en) * 2018-12-18 2019-04-16 重庆邮电大学 A kind of augmented reality three-dimensional Tracing Registration method based on LINE-MOD template matching
CN109887030A (en) * 2019-01-23 2019-06-14 浙江大学 Texture-free metal parts image position and posture detection method based on the sparse template of CAD

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6606406B1 (en) * 2000-05-04 2003-08-12 Microsoft Corporation System and method for progressive stereo matching of digital images
US8411080B1 (en) * 2008-06-26 2013-04-02 Disney Enterprises, Inc. Apparatus and method for editing three dimensional objects
CN107942949A (en) * 2017-03-31 2018-04-20 沈机（上海）智能系统研发设计有限公司 A kind of lathe vision positioning method and system, lathe
CN109636854A (en) * 2018-12-18 2019-04-16 重庆邮电大学 A kind of augmented reality three-dimensional Tracing Registration method based on LINE-MOD template matching
CN109887030A (en) * 2019-01-23 2019-06-14 浙江大学 Texture-free metal parts image position and posture detection method based on the sparse template of CAD

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
王月等: "基于模型的增强现实无标识三维注册追踪方法", 《上海交通大学学报》 *
郝明: "机器视觉在机器人杂乱工件分拣中的应用", 《中国优秀硕士学位论文全文数据库（电子期刊）信息科技辑》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111738320A (en) * 2020-03-04 2020-10-02 沈阳工业大学 Shielded workpiece identification method based on template matching
CN111738320B (en) * 2020-03-04 2022-12-06 沈阳工业大学 Shielded workpiece identification method based on template matching
CN111814580A (en) * 2020-06-15 2020-10-23 南京澳讯人工智能研究院有限公司 Target effective identification device and method

Similar Documents

Publication Publication Date Title
CN109903313B (en) Real-time pose tracking method based on target three-dimensional model
JP6216508B2 (en) Method for recognition and pose determination of 3D objects in 3D scenes
CN110689573B (en) Edge model-based augmented reality label-free tracking registration method and device
CN111445459B (en) Image defect detection method and system based on depth twin network
CN110837768A (en) Rare animal protection oriented online detection and identification method
CN109272577B (en) Kinect-based visual SLAM method
CN110021029B (en) Real-time dynamic registration method and storage medium suitable for RGBD-SLAM
CN110222661B (en) Feature extraction method for moving target identification and tracking
CN114743259A (en) Pose estimation method, pose estimation system, terminal, storage medium and application
Chiverton et al. Automatic bootstrapping and tracking of object contours
Yu et al. Improvement of face recognition algorithm based on neural network
CN105678778A (en) Image matching method and device
CN110647925A (en) Rigid object identification method and device based on improved LINE-MOD template matching
Huang et al. Tracking-by-detection of 3d human shapes: from surfaces to volumes
CN110472651B (en) Target matching and positioning method based on edge point local characteristic value
CN105139013A (en) Object recognition method integrating shape features and interest points
CN114267061A (en) Head gesture recognition method, device, equipment and computer storage medium
CN109872343B (en) Weak texture object posture tracking method, system and device
CN110516638B (en) Sign language recognition method based on track and random forest
CN116416305B (en) Multi-instance pose estimation method based on optimized sampling five-dimensional point pair characteristics
CN117132630A (en) Point cloud registration method based on second-order spatial compatibility measurement
CN107330436B (en) Scale criterion-based panoramic image SIFT optimization method
Dai et al. An Improved ORB Feature Extraction Algorithm Based on Enhanced Image and Truncated Adaptive Threshold
He et al. Recent advance on mean shift tracking: A survey
Mahmood et al. Nose tip detection using shape index and energy effective for 3d face recognition

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200103

RJ01 Rejection of invention patent application after publication