ORB feature extraction method based on adaptive sampling pyramid in dynamic environment

Abstract

The invention discloses an ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment. The method comprises the following steps: acquiring and preprocessing an original image; identifying key points in the preprocessed image by using a FAST detector, calculating gradient intensity, and generating a pixel point importance score value; dividing the preprocessed image into a plurality of small blocks, and performing dynamic resolution distribution based on the importance score value of each small block; creating a highest-resolution image layer, and performing non-uniform down-sampling on the previous layer according to the resolution hierarchy of each small block to generate an adaptive sampling pyramid; carrying out smooth transition processing on pixel values between adjacent small blocks by adopting a fusion algorithm; the scale space continuity is maintained; gradual change between adjacent layers is controlled; and outputting the multi-scale feature points and the descriptors. According to the method, the ASP algorithm and the ORB algorithm are combined, so that the technical problem of low accuracy and stability of feature extraction of a dynamic environment by a conventional pyramid ORB feature extraction method based on a fixed-scale image is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of feature extraction, and particularly relates to an ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment. Background Art

[0002] ORB (Oriented FAST and Rotated BRIEF) features are a feature detection and description method in the field of computer vision. It combines the advantages of the FAST corner detector and the BRIEF descriptor, and enhances the rotational invariance by introducing direction information. It can quickly generate descriptors in the form of binary strings, making the distance calculation in the feature matching process simple and fast. At the same time, scale invariance is achieved by constructing an image pyramid. However, traditional ORB feature extraction methods usually rely on a fixed image pyramid structure for multi-scale analysis. Although this fixed-scale pyramid can provide a certain degree of scale invariance, in a dynamic environment, when there are significant scale changes or complex backgrounds in the environment, the fixed-scale pyramid may lead to problems such as uneven distribution of feature points, loss of key information, or waste of computing resources. In addition, for applications that require real-time response, the fixed multi-layer pyramid introduces unnecessary delays and affects the overall performance of the system. Therefore, how to perform multi-scale analysis on the ORB pyramid construction is one of the important research topics of the ORB feature extraction algorithm in the current dynamic environment.

[0003] Currently, there are two mainstream methods for optimizing pyramid construction strategies. One is the non-uniform pyramid, which retains higher resolution in certain specific regions or regions of interest, and uses lower resolution elsewhere. This helps to focus on important features while reducing the computational amount. The other is the Adaptive Sampling Pyramid (ASP), which adjusts the sampling rate in real time according to the image content to ensure that important regions (such as edges and texture-dense regions) obtain higher resolution, while unimportant regions use lower resolution. For fast-moving objects or scenes in a dynamic environment, adaptive sampling can quickly adjust the sampling strategy, capture key information, and reduce information loss caused by motion blur or other factors.

[0004] Aiming at the disadvantages of the above-mentioned traditional ORB feature extraction method based on a fixed-scale image pyramid structure, which has low accuracy and stability in feature extraction for a dynamic environment, the present invention provides an ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment to enhance the accuracy and stability of the ORB algorithm for feature extraction in a dynamic scene. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a design method for ORB feature extraction based on an adaptive sampling pyramid in a dynamic environment, so as to solve the technical problems of low accuracy and stability of the traditional ORB feature extraction method based on a fixed-scale image pyramid structure for feature extraction in a dynamic environment.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] An ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment includes the following steps:

[0008] Step 1, obtain and preprocess the original image;

[0009] Step 2, use the FAST detector to identify key points in the preprocessed image, calculate the gradient intensity, and generate a pixel point importance score value;

[0010] Step 3, divide the preprocessed image into several small blocks, obtain the resolution levels of each small block based on the importance score values of each small block, and perform dynamic resolution allocation according to the resolution levels of each small block;

[0011] Step 4, create the highest resolution image layer, and perform non-uniform downsampling on the previous layer according to the resolution levels of each small block to generate an adaptive sampling pyramid;

[0012] Step 5, use a fusion algorithm to perform smooth transition processing on the pixel values between adjacent small blocks;

[0013] Step 6, map the feature points of the higher resolution level to the lower resolution level;

[0014] Step 7, set a threshold such that the difference between adjacent two downsampling scale factors is less than the set threshold;

[0015] Step 8, output multi-scale feature points and descriptors.

[0016] Further, the calculation method of the importance score value S(x, y) in Step 2 is:

[0017] S(x, y) = f(FAST key point number (x, y), gradient intensity (x, y))

[0018] Wherein, S(x, y) represents the importance score value; (x, y) represents the key points in the preprocessed image identified by the FAST detector, FAST key point number (x, y) represents the number of FAST corner points detected in the small block (x, y); gradient intensity (x, y) represents the average gradient modulus value in the small block (x, y).

[0019] Further, Step 3 includes:

[0020] Step 31: Divide the pre-processed image into w×h blocks B i , where w×h is the total number of small blocks, w represents the number of horizontal small blocks into which the image is divided, h represents the number of vertical small blocks into which the image is divided, B i represents the i-th small block;

[0021] In step 32, the resolution level of each small block is obtained according to the importance score S(x,y) of each small block, and dynamic resolution allocation is performed according to the resolution level of each small block. The dynamic resolution allocation method is:

[0022] r(B i )=g(R j )

[0023] In the formula, r(B i ) means B i The downsampling scale factor of the corresponding resolution level; g(R j ) means B i The corresponding resolution level R j Mapped to a specific downsampling scale factor.

[0024] Furthermore, step 4 includes:

[0025] Step 41, using the original image as the highest resolution image layer L0;

[0026] Step 42: According to the resolution level of each small block, perform non-uniform downsampling on the previous layer to generate a new sub-image L k , that is: L k =downsample(L k-1 ,r k ), where L k-1 Represents the upper layer subgraph; r k Indicates that from L k-1 to L k The downsampling scale factor.

[0027] Furthermore, in step 5, the fusion algorithm used is the weighted average algorithm, and the smooth transition processing method is:

[0028] p(L k ,x,y)=w(x,y)·p(B i ,L k ,x,y)+(1-w(x,y))·p(B i+1 ,L k ,x,y)

[0029] Among them, p(L k ,x,y) represents the subgraph Lk The new pixel value at the coordinates (x, y); p(B i ,L k ,x,y) represents the pixel value of the small block B i at the coordinates (x, y); p(B i+1 ,L k ,x,y) represents the pixel value of the small block B i+1 at the coordinates (x, y); w(x, y) and (1 - w(x, y)) respectively represent the weighting values of the small block B i and the small block B i+1 at the coordinates (x, y); B i+1 represents the (i + 1)-th small block.

[0030] Furthermore, in step 6, the calculation method of the feature point mapping is as follows:

[0031] (x′, y′) = T k,k+n ·(x, y)

[0032] where T k,k+n represents the resolution conversion matrix; (x, y) represents the coordinates of the feature point in the k-th layer; (x', y') represents the coordinates of the corresponding feature point in the (k + n)-th layer.

[0033] Furthermore, the calculation method of step 7 is as follows:

[0034]

[0035] where represents the set threshold; |r k -r k+1 | represents the absolute value of the difference between two adjacent downsampling ratio factors.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention is scientifically and reasonably designed and convenient to use. In a dynamic environment, the ORB feature extraction combined with the adaptive sampling pyramid method endows scale and rotation invariance through multi-scale feature capture and direction information, enhancing the robustness to illumination changes; the ASP mechanism optimizes the use of computing resources, reduces memory occupancy, and improves the accuracy and efficiency of feature matching through non-maximum suppression. This method focuses on the region of interest, reduces the influence of background noise, and is particularly suitable for real-time processing and resource-constrained scenarios. Therefore, the combination of ORB and ASP can improve the accuracy and stability of feature extraction for dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is the flowchart of the steps of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0040] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention. In addition, the terms "first", "second", "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0041] In the description of the present invention, it should be noted that unless otherwise clearly specified and limited, the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; of course, it can also be a mechanical connection or an electrical connection; in addition, it can also be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0042] As Figure 1 shown, a method for ORB feature extraction based on an adaptive sampling pyramid in a dynamic environment includes the following steps:

[0043] Step 1: Obtain and preprocess the original image;

[0044] Step 2: Use the FAST detector to identify the key points in the preprocessed image, calculate the gradient intensity, and generate the pixel point importance score value;

[0045] Step 3: Divide the preprocessed image into several small blocks, obtain the resolution levels of each small block based on the importance score value of each small block, and perform dynamic resolution allocation according to the resolution levels of each small block;

[0046] Step 4: Create the highest resolution image layer, and perform non-uniform downsampling on the previous layer according to the resolution levels of each small block to generate an adaptive sampling pyramid;

[0047] Step 5. Use a fusion algorithm to perform smooth transition processing on pixel values between adjacent small blocks;

[0048] Step 6. Map feature points of a higher resolution level to a lower resolution level; ⏎

[0049] ⏎ Step 7. Set a threshold such that the difference between two adjacent downsampling scale factors is less than the set threshold; ⏎

[0050] ⏎ Step 8. Output multi-scale feature points and descriptors. Specifically: Detect feature points at different levels, calculate rotation-invariant BRIEF descriptors, perform feature point screening and sorting, and return the final result to complete feature extraction. ⏎

[0051] ⏎ The calculation method of the importance score value S(x, y) in Step 2 is as follows: ⏎

[0052] ⏎ S(x, y) = f(Number of FAST key points(x, y), Gradient intensity(x, y)) ⏎

[0053] ⏎ where S(x, y) represents the importance score value; (x, y) represents the key points in the preprocessed image identified by the FAST detector, Number of FAST key points(x, y) represents the number of FAST corner points detected in the small block (x, y); Gradient intensity(x, y) represents the average gradient modulus value in the small block (x, y). ⏎

[0054] ⏎ Step 3 includes: ⏎

[0055] ⏎ Step 31. Divide the preprocessed image into w × h small blocks B i ⏎ , where w × h is the total number of small blocks, w represents the number of horizontal small blocks into which the image is divided, h represents the number of vertical small blocks into which the image is divided, and B i ⏎ represents the i-th small block; ⏎

[0056] ⏎ Step 32. Obtain the resolution levels of each small block according to the importance score value S(x, y) of each small block, and perform dynamic resolution allocation according to the resolution levels of each small block. The dynamic resolution allocation method is: ⏎

[0057] ⏎ r(B i ⏎ ) = g(R j ⏎ ) ⏎

[0058] ⏎ In the formula, r(B i ⏎ ) represents the downsampling scale factor of the resolution level corresponding to B i ⏎ ; g(R j ⏎ ) represents mapping the resolution level R i ⏎ corresponding to B j ⏎ to a specific downsampling scale factor. ⏎

[0059] ⏎ Step 4 includes:

[0060] Step 41: Take the original image as the highest-resolution image layer L0;

[0061] Step 42: According to the resolution level of each small block, perform non-uniform downsampling on the previous layer to generate a new subgraph L k , that is: L k = downsample(L k-1 , r k ), where L k-1 represents the previous-layer subgraph; r k represents the downsampling ratio factor from L k-1 to L k .

[0062] In Step 5, the fusion algorithm adopted is the weighted average algorithm, and the smoothing transition processing method is:

[0063] p(L k , x, y) = w(x, y) · p(B i , L k , x, y) + (1 - w(x, y)) · p(B i+1 , L k [[ID=3,7]]x, y)

[0064] where p(L k , x, y) represents the new pixel value at the coordinate (x, y) in the subgraph L k ; p(B i , L k , x, y) represents the pixel value of the small block B i at the coordinate (x, y); p(B i+1 , L k , x, y) represents the pixel value of the small block B i+1 at the coordinate (x, y);

[0065] w(x, y) and (1 - w(x, y)) respectively represent the weighted values of the small block B i and the small block B i+1 at the coordinate (x, y); B i+1 represents the (i + 1)-th small block.

[0066] In Step 6, the calculation method of the feature point mapping is:

[0067] (x′, y′) = T k,k+n · (x, y)

[0068] where T k,k+n represents the resolution transformation matrix; (x, y) represents the coordinates of the feature point in the k-th layer; (x', y') represents the coordinates of the corresponding feature point in the (k + n)-th layer.

[0069] The calculation method of step 7 is as follows:

[0070]

[0071] Wherein, represents a set threshold; |r k -r k+1 | represents the absolute value of the difference between two adjacent downsampling scale factors.

[0072] Finally, it should be noted that: the above embodiments are only preferred embodiments of the present invention to illustrate the technical solutions of the present invention, rather than limiting it, and certainly not limiting the patent scope of the present invention; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention; that is to say, any meaningless changes or touch-ups made on the main design idea and spirit of the present invention, as long as the technical problems solved are still the same as those of the present invention, should be included in the protection scope of the present invention; in addition, directly or indirectly applying the technical solutions of the present invention to other related technical fields shall be similarly included in the patent protection scope of the present invention.

Claims

1. An ORB feature extraction method based on adaptive sampling pyramid in a dynamic environment, characterized by: It includes the following steps: Step 1: Obtain and preprocess the original image; Step 2: Use the FAST detector to identify key points in the preprocessed image, calculate the gradient intensity, and generate the pixel point importance score value; Step 3: Divide the preprocessed image into several small blocks, obtain the resolution levels of each small block based on the importance score value of each small block, and perform dynamic resolution allocation according to the resolution levels of each small block; Step 4: Create the highest resolution image layer, and perform non-uniform downsampling on the previous layer according to the resolution levels of each small block to generate an adaptive sampling pyramid; Step 5: Use a fusion algorithm to perform smooth transition processing on the pixel values between adjacent small blocks; Step 6: Map the feature points of the higher resolution level to the lower resolution level; Step 7: Set a threshold such that the difference between two adjacent downsampling scale factors is less than the set threshold; Step 8: Output multi-scale feature points and descriptors.

2. The ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment according to claim 1, wherein The calculation method of the importance score value S(x, y) in Step 2 is: S(x, y) = f(FAST key point number(x, y), gradient intensity(x, y)) where S(x, y) represents the importance score value; (x, y) represents the key points in the preprocessed image identified by the FAST detector, FAST key point number(x, y) represents the number of FAST corner points detected in the small block (x, y); gradient intensity(x, y) represents the average gradient modulus value in the small block (x, y).

3. The ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment according to claim 2, wherein Step 3 includes: Step 31: Divide the preprocessed image into w×h small blocks B i , where w×h is the total number of small blocks, w represents the number of horizontal small blocks into which the image is divided, h represents the number of vertical small blocks into which the image is divided, and B i represents the i-th small block; Step 32: Obtain the resolution levels of each small block according to the importance score value S(x, y) of each small block, and perform dynamic resolution allocation according to the resolution levels of each small block. The dynamic resolution allocation method is: r(B i ) = g(R j ) where r(B i ) represents the downsampling ratio factor of the resolution level corresponding to B i ; g(R j ) represents mapping the resolution level R i corresponding to B j to a specific downsampling ratio factor.

4. The ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment according to claim 3, wherein Step 4 includes: Step 41: Take the original image as the highest resolution image layer L0; Step 42: Based on the resolution level of each small block, perform non-uniform downsampling on the previous layer to generate a new subgraph L k , that is: L k = downsample(L k-1 , r k ), where L k-1 represents the subgraph of the previous layer; r k represents the downsampling ratio factor from L k-1 to L k .

5. A method for ORB feature extraction based on an adaptive sampling pyramid in a dynamic environment according to claim 4, characterized in that, In Step 5, the fusion algorithm adopted is the weighted average algorithm, and the smooth transition processing method is: p(L k , x, y) = w(x, y) · p(B i , L k , x, y) + (1 - w(x, y)) · p(B i+1 , L k , x, y) where p(L k , x, y) represents the new pixel value at the coordinate (x, y) in the sub - graph L k ; p(B i , L k , x, y) represents the pixel value of the small block B i at the coordinate (x, y); p(B i+1 , L k , x, y) represents the pixel value of the small block B i+1 at the coordinate (x, y); w(x, y) and (1 - w(x, y)) respectively represent the weighting values of the small block B i and the small block B i+1 at the coordinate (x, y); B i+1 represents the (i + 1)-th small block.

6. The ORB feature extraction method based on an adaptive sampling pyramid in a dynamic environment according to claim 5, wherein In Step 6, the calculation method of feature point mapping is: (x′, y′) = T k,k+n ·(x, y) where, T k,k+n represents the transformation matrix of the resolution; (x, y) represents the coordinates of the feature point in the k-th layer; (x', y') represents the coordinates of the corresponding feature point in the (k + n)-th layer.

7. A method for ORB feature extraction based on an adaptive sampling pyramid in a dynamic environment according to claim 6, characterized in that The calculation method of Step 7 is: |r k -r k+1 |＜ò where, ò represents a set threshold; |r k -r k+1 | represents the absolute value of the difference between two adjacent downsampling scale factors.

Images