CN103279952A - Target tracking method and device - Google Patents

Abstract

The embodiment of the present invention discloses a target tracking method, which includes using a preset algorithm to extract corner features of a reference image to obtain a first set of feature points of the reference image, and removing out of the first set of feature points that exceed the number of points to be matched. The abnormal points in the image range are obtained by obtaining a second set of feature points, performing image matching on the second set of feature points and the image to be matched, and completing target tracking on the image to be matched according to the matching result. The embodiment of the invention also discloses a target tracking device. With the present invention, the extracted corner features have good recognition and stability, greatly improving the speed of matching, and at the same time optimize the set of feature points in the reference image to eliminate out-of-range corner points, which solves the problems in the prior art. Point out of range can not be matched, improve the reliability of matching.

Description

A method and device for tracking a target

technical field

The invention relates to the field of image processing, in particular to a target tracking method and device.

Background technique

Object detection technology is also commonly referred to as image feature extraction. It is a concept of computer vision and digital image processing. It refers to the image information extraction process that uses computer operations to determine whether the pixels of each image are image features. The target tracking technology is to determine the position of the region of interest (Region Of Interest, ROI), also known as the template (Template), in the video image or continuous image sequence, and correspond to the template of each frame of image.

The existing tracking technology solutions can be described as follows:

(1) Determine a template in the collected image sequence, which is usually called a reference image, and the target to be tracked is recorded in the reference image;

(2) In order to improve the efficiency of the operation, instead of transforming each pixel in the reference image, the pixels are evenly extracted from the reference image at equal intervals. This process can be called equidistant sampling to form the original feature point set;

(3) Operate the original feature point set and its domain information to obtain a new feature point set, and determine the matching area on the image to be matched according to the new feature point set;

(4) Calculate the grayscale information between the matching area and the reference template, such as the SSD (Sum of Squared Differences, sum of squared differences) method, using the method of minimizing the error, through iteration, so that the matching area and the reference image can match Even if there is the same image information in the matching area and the reference image, that is, the target that needs to be tracked.

(5) Repeat steps (3)-(4) for the acquired image sequence, and finally achieve continuous target tracking through template matching between image frames.

However, the prior art has the following disadvantages:

(1) When obtaining the set of feature points of the reference image, a method of equidistantly sampling the pixels in the reference image is adopted. The feature points obtained in this way are very random, usually contain less image information, cannot represent the image features well, and the reliability and stability are not high, so the tracking algorithm does not have good robustness.

(2) After the reference image is obtained, it is used as a standard template without any changes. However, in an actual tracking system, with the movement of image acquisition devices such as cameras, the determined reference image may partially move out of the image acquisition range of the camera and cannot be imaged, so that some areas of the reference image are not in the images of the subsequent image sequence. However, the existing scheme uses the initial reference image, which will make the algorithm for calculating the image gray information between the reference image and the matching area unable to converge, and the correct result cannot be obtained, resulting in tracking failure.

Contents of the invention

The technical problem to be solved by the embodiments of the present invention is to provide a target tracking method. It can solve the problems in the prior art that the image feature stability is not high and the reference image is out of range, which leads to tracking failure.

In order to solve the above technical problems, the first aspect of the present invention provides a target tracking method, including:

Extracting the corner features of the reference image by using a preset algorithm to obtain a first set of feature points of the reference image, the reference image is used to track the target in the current image to be matched;

Eliminating abnormal points in the first set of feature points beyond the range of the image to be matched to obtain a second set of feature points;

performing image matching on the second feature point set and the image to be matched, and completing target tracking on the image to be matched according to the matching result.

In a first possible implementation, the preset algorithm includes: Harris corner detection algorithm, FAST accelerated segmentation detection feature corner detection algorithm, KLT corner detection algorithm or SUSAN minimum kernel value similar area corner detection algorithm.

With reference to the first aspect or the first possible implementation manner, in the second possible implementation manner, before the step of extracting corner features of the reference image using a preset algorithm to obtain the first feature point set of the reference image ,Also includes:

Obtain the image sequence collected by the camera, and use the image sequence as the image to be matched;

And determine the reference image of the image to be matched from the first frame image of the image sequence.

With reference to the second possible implementation of the first aspect, in a third possible implementation, the removing the abnormal points in the first set of feature points beyond the range of the image to be matched to obtain the second feature points The assembly steps include:

Obtain the coordinates of each feature point in the first set of feature points;

Judging according to the coordinates of each feature point in the feature point set whether it exceeds the scope of the image to be matched, if yes, then determining that the feature point is an abnormal point;

All abnormal points in the first set of feature points are eliminated to obtain a second set of feature points.

With reference to the third possible implementation of the first aspect, in a fourth possible implementation, image matching is performed on the second feature point set and the image to be matched, and the to-be-matched image is completed according to the matching result. The steps of object tracking for matching images include:

Selecting a subimage to be matched with the same size as the reference image in the image to be matched;

Using the preset algorithm to extract the features of the subgraph to be matched to obtain a third set of feature points;

calculating the distance between the second set of feature points and the third set of feature points;

Judging whether the distance is smaller than a preset threshold, if yes, then determining that the reference image is successfully matched with the subimage to be matched in the image to be matched, if not, re-executing the selection process in the image to be matched A step to match subimages of the same size as the reference image.

With reference to the fourth possible implementation of the first aspect, in a fifth possible implementation, image matching is performed on the second feature point set and the image to be matched, and the to-be-matched image is completed according to the matching result. After the step of matching image object tracking, it also includes:

The reference image is deleted and the subimage to be matched that successfully matches the reference template is used as a new reference image.

Correspondingly, the second aspect of the present invention also provides a target tracking device, including:

A feature extraction module, configured to extract corner features of a reference image using a preset algorithm to obtain a first set of feature points of the reference image, the reference image being used to track the target in the current image to be matched;

A feature optimization module, configured to remove abnormal points in the first set of feature points that exceed the range of the image to be matched to obtain a second set of feature points;

An image matching module, configured to perform image matching on the second set of feature points and the image to be matched, and complete target tracking on the image to be matched according to the matching result.

In a first possible implementation, the feature extraction module is used to extract corner features of the reference image using a preset algorithm to obtain a first set of feature points of the reference image, and the preset algorithm includes Harris corner detection Algorithm, FAST accelerated segmentation detection feature corner detection algorithm, KLT corner detection algorithm or SUSAN minimum kernel value similar area corner detection algorithm.

In combination with the second aspect and the first possible implementation, the second possible implementation also includes:

The template determining module is used to obtain the image sequence collected by the camera, and use the image sequence as the image to be matched; and determine the reference image of the image to be matched from the first frame image of the image sequence.

With reference to the second possible implementation, in the third possible implementation, the feature optimization module includes:

an acquisition unit, configured to acquire the coordinates of each feature point in the first set of feature points;

A judging unit, configured to judge whether the coordinates of each feature point in the feature point set exceed the range of the image to be matched, and if so, determine that the feature point is an abnormal point;

The elimination unit is configured to eliminate all abnormal points in the first feature point set to obtain a second feature point set.

With reference to the third possible implementation, in a fourth possible implementation, the image matching module includes:

A selecting unit, configured to select a sub-image to be matched that is the same size as the reference image in the image to be matched;

An extraction unit, configured to extract features of the to-be-matched subgraph using the preset algorithm to obtain a third set of feature points;

a calculation unit, configured to calculate the distance between the second set of feature points and the third set of feature points;

In combination with the fourth possible implementation, the fifth possible implementation also includes:

The template updating module is configured to delete the reference image and use the subimage to be matched that successfully matches the reference template as a new reference image.

Implementing the embodiment of the present invention has the following beneficial effects:

The feature point set of the reference image is obtained by extracting the corner features of the reference image, the abnormal points in the feature point set that exceed the range of the image to be matched are eliminated, and the feature point set that has eliminated the abnormal point is image-matched with the image to be matched to achieve Target Tracking. The use of corner features for graphic matching can reduce the amount of data involved in the calculation while retaining the important features of image graphics. The corner features have good recognition and stability, which greatly improves the matching speed. At the same time, the method optimizes the set of feature points in the reference image and eliminates out-of-range corner points, which solves the problem that the out-of-range feature points cannot be matched in the prior art, and improves the reliability of matching.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a schematic flow chart of a target tracking method according to an embodiment of the present invention;

FIG. 2 is another schematic flowchart of a target tracking method according to an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of a target tracking device according to an embodiment of the present invention;

Fig. 4 is another structural schematic diagram of a target tracking device according to an embodiment of the present invention;

Fig. 5 is a schematic structural diagram of a feature optimization module in Fig. 4;

Fig. 6 is a schematic structural diagram of the image matching module in Fig. 4;

Fig. 7 is another schematic structural diagram of an object tracking device according to an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Referring to Fig. 1, it is a schematic flow chart of a target tracking method according to an embodiment of the present invention, the method comprising:

Step 101, using a preset algorithm to extract corner features of a reference image to obtain a first set of feature points of the reference image, the reference image is used to track an object in the current image to be matched.

Specifically, there is no clear mathematical definition of the corner point, and it is generally believed that the corner point is the point where the brightness of the two-dimensional image changes sharply or the point of the maximum curvature on the edge curvature of the image. Among various features of images, corner is a stable, rotation-invariant and effective feature that can overcome grayscale inversion. The algorithm for the target tracking device to extract the corner feature in the reference image can be Harris corner detection algorithm, FAST corner detection algorithm, KLT corner detection algorithm or SUSAN corner detection algorithm, or other algorithms, which are not limited by the present invention. The target tracking device extracts corner features of the reference image to obtain a first set of feature points, and the first set of feature points is a binary image. The reference image is an image that records the target to be tracked, and the reference image is matched with the image to be matched to realize the tracking of the target.

Step 102: Eliminate outliers in the first set of feature points that exceed the scope of the image to be matched to obtain a second set of feature points.

Specifically, when the shape of the target in the image to be matched is incomplete, for example, when the target is occluded or the target is outside the image to be matched, there are points that do not belong to the range of the image to be matched in the first set of feature points extracted from the reference image. The corner points inside become abnormal points. If these abnormal points are involved in the calculation of image matching, the matching calculation results will not converge to the correct result, that is, the matching of the reference image and the image to be matched will fail. In order to avoid this situation, the target tracking device obtains the coordinates of each corner point in the first feature point set, and judges whether the corner point exceeds the range of the image to be matched according to the coordinate relationship between the coordinates of the corner point and the image to be matched, and if so, The corner point is obtained as an abnormal point, and all abnormal points are eliminated according to the above method to obtain the second set of feature points.

Step 103 , performing image matching on the second set of feature points and the image to be matched, and completing object tracking on the image to be matched according to the matching result.

Specifically, in the image to be matched, a subimage to be matched with the same size as the reference image is selected, and the same corner point extraction algorithm as in step 101 is used to extract the corner feature of the subimage to be matched to obtain a third feature point set. There are two ways to match point features. One is to use a predetermined method to describe the second feature point set of the extracted reference image and the third feature point set of the sub-image to be matched. The matching of the image is converted into two vectors set of matches. The other is to represent the extracted feature point set as a binary image, perform distance transformation on the binary image, and use distance similarity to measure the similarity of two feature point sets. Of course, other methods may also be used, which are not limited in the present invention.

In the embodiment of the present invention, the feature point set of the reference image is obtained by extracting the corner point features of the reference image, the abnormal points in the feature point set that exceed the range of the image to be matched are eliminated, and the feature point set that has eliminated the abnormal point is combined with the feature point set to be matched. The images are image-matched for object tracking. The use of corner features for graphic matching can reduce the amount of data involved in the calculation while retaining the important features of image graphics. The corner features have good recognition and stability, which greatly improves the matching speed. At the same time, the method optimizes the set of feature points in the reference image and eliminates out-of-range corner points, which solves the problem that the out-of-range feature points cannot be matched in the prior art, and improves the reliability of matching.

Referring to FIG. 2 , it is a schematic diagram of another embodiment of a target tracking method according to an embodiment of the present invention. The method includes: step 201, acquiring an image sequence captured by a camera, and using the image sequence as an image to be matched; The reference image of the image to be matched is determined in the first frame image of the sequence.

Specifically, the format of the image sequence collected by the camera can be BMP (Bitmap, BMP for short), TIFF (Tagged Image File Format, TIFF for short) or JPEG (Joint Photographic Experts Group, joint image expert group , referred to as JPEG) and other image formats, or MPEG (Moving Pictures Experts Group/Motin Pictures Experts Group, MPEG for short) or AVI (Audio Video Interleaved, audio and video interleaved format, referred to as AVI) and other video formats. The tracking device takes the image sequence as the image to be matched, and determines the reference image from the first frame image in the image sequence, and the target to be tracked is recorded in the reference image, assuming that the size of the reference image is M*N (M, N is the number of pixels), and the size of the image to be matched is m*n (m, n is the number of pixels), then M<m, N<n.

Step 202, using a preset algorithm to extract corner features in the reference image to obtain a first feature point set of the reference image, the reference image is used to track the target in the current image to be matched.

Specifically, the algorithm for the target tracking device to extract the corner feature in the reference image may be the Harris corner detection algorithm, the FAST corner detection algorithm, the KLT corner detection algorithm or the SUSAN corner detection algorithm, or other algorithms. Without limitation, the target tracking device extracts corner features of the reference image to obtain a first set of feature points, and the first set of feature points is a binary image. The reference image is an image that records the target to be tracked, and the reference image is matched with the image to be matched to realize the tracking of the target.

Step 203, removing outliers in the first set of feature points beyond the scope of the image to be matched, and converting them into a second set of feature points.

Specifically, the target tracking device acquires the coordinates of each corner point in the first feature point set, and at the same time acquires the coordinates of each pixel point in the image to be matched, and judges whether the corner point exceeds the range of the image to be matched according to the relationship between the two coordinates. If yes, the corner point is determined to be an abnormal point, and all abnormal points in the first feature point set are eliminated according to the above method to obtain the second feature point set.

Step 204. Select a subimage to be matched in the image to be matched that has the same size as the image.

Specifically, taking the example in step 201, assuming that the size of the reference image is M*N, the size of the image to be matched is m*n, satisfying the relationship M<m, N<n, and the target tracking device selects a size M in the image to be matched *N's subgraph to be matched, the search strategy of the subgraph to be matched generally adopts the method of traversal or genetic algorithm, which is not limited in the present invention.

Step 205, using the preset algorithm to extract the corner point features of the subgraph to be matched to obtain a third feature point set.

Specifically, the object tracking device uses the same algorithm as that in step 202 to extract the corner point features of the subgraph to be matched to obtain the third feature point set.

Step 206. Calculate the distance between the second set of feature points and the third set of feature points.

Specifically, the target tracking device uses a preset transformation algorithm to perform distance transformation on the first feature point set to obtain a binary image A, and uses the same preset transformation algorithm to perform distance transformation on the third feature point set to obtain a binary image B. The preset transformation algorithm can be 3-4DT. The distance between binary image A and binary image B can be calculated using Euclidean distance or non-Euclidean distance. Here, Hausdorff distance is taken as an example. Hausdorff distance is a maximum-minimum (Max-Min) defined on two point sets. Distance, such as calculating the Hausdorff distance between the above-mentioned binary image A and binary image B, assuming A={a1,...,ap}, B={b1,...,bq}, then between the two point sets The Hausdorff distance is defined as H(A,B)=max(h(A,B),h(B,A)) (1)

h(A,B)=max(a∈A)min(b∈B)‖a-b‖ (2)

h(B,A)=max(b∈B)min(a∈A)‖b-a‖ (3)

‖·‖ is the distance paradigm (such as: L2 or Euclidean distance) between point set A and point set B.

Here, formula (1) is called the two-way Hausdorff distance, which is the most basic form of Hausdorff distance; h(A,B) and h(B,A) in formula (2) are called from A set to B set and from One-way Hausdorff distance from set B to set A. That is, h(A,B) actually sorts the distance ‖ai-bj‖ between each point ai in the point set A and the point bj in the B set closest to this point ai, and then takes the distance The maximum value is used as the value of h(A,B). In the same way, h(B,A) can be obtained. According to formula (1), the two-way Hausdorff distance H(A,B) is the larger of the one-way distance h(A,B) and h(B,A). Or, it measures the maximum degree of mismatch between two point sets.

Step 207, judging whether the distance is smaller than a preset threshold.

Specifically, the object tracking device judges whether the distance calculated in step 206 is smaller than a preset threshold, if yes, execute step 206 , and if no, return to execute step 204 .

Step 208, determining that the reference image is successfully matched with the subimage to be matched in the image to be matched.

Specifically, the object tracking device determines that the image to be matched includes an object to be tracked, the reference image is successfully matched with the image to be matched, and tracks the object in the subimage to be matched.

Step 209, delete the reference image and use the subimage to be matched that successfully matches the reference image as a new reference image.

Specifically, the subimage to be matched that is successfully matched in the image to be matched is used as a new reference image, and the new reference image is used for image matching when performing target tracking on the next frame of image to be matched. By dynamically updating the reference image, the shape and size of the target in the reference image can be adjusted, making image matching more accurate.

In the embodiment of the present invention, the feature point set of the reference image is obtained by extracting the corner point features of the reference image, the abnormal points in the feature point set that exceed the range of the image to be matched are eliminated, and the feature point set that has eliminated the abnormal point is combined with the feature point set to be matched. The images are image-matched for object tracking. The use of corner features for graphic matching can reduce the amount of data involved in the calculation while retaining the important features of image graphics. The corner features have good recognition and stability, which greatly improves the matching speed. At the same time, the method optimizes the set of feature points in the reference image and eliminates out-of-range corner points, which solves the problem that the out-of-range feature points cannot be matched in the prior art, and improves the reliability of matching.

Referring to FIG. 3 , it is a schematic structural diagram of a target tracking device according to an embodiment of the present invention, hereinafter referred to as device 1, and the device 1 includes:

The feature extraction module 11 is configured to use a preset algorithm to extract corner features of a reference image to obtain a first set of feature points of the reference image, and the reference image is used to track an object in the current image to be matched.

Specifically, the algorithm for feature extraction module 11 to extract corner features in the reference image may be Harris corner detection algorithm, FAST corner detection algorithm, KLT corner detection algorithm or SUSAN corner detection algorithm, or other algorithms. The invention is not limited. The target tracking device extracts corner features of the reference image to obtain a first feature point set, and the first feature point set is a binary image. The reference image is an image that records the target to be tracked, and the reference image is matched with the image to be matched to realize the tracking of the target.

The feature optimization module 12 is configured to remove outliers in the first set of feature points beyond the range of the image to be matched to obtain a second set of feature points.

Specifically, when the shape of the target in the image to be matched is incomplete, for example, when the target is occluded or the target is outside the image to be matched, there are points that do not belong to the range of the image to be matched in the first set of feature points extracted from the reference image. The corner points inside become abnormal points. If these abnormal points are involved in the calculation of image matching, the matching calculation results will not converge to the correct result, that is, the matching of the reference image and the image to be matched will fail. In order to avoid this situation, the feature optimization module 12 obtains the coordinates of each corner point in the first feature point set, and judges whether the corner point exceeds the range of the image to be matched according to the coordinate relationship between the coordinates of the corner point and the image to be matched, if yes , then the corner point is obtained as an abnormal point, and all abnormal points are eliminated according to the above method to obtain the second set of feature points.

The image matching module 13 is configured to perform image matching on the second set of feature points and the image to be matched, and complete target tracking on the image to be matched according to the matching result.

Specifically, the image matching module 13 selects a subimage to be matched with the same size as the reference image in the image to be matched, and uses the same corner point extraction algorithm as in the feature extraction module 11 to extract the corner feature of the subimage to be matched to obtain The third feature point set, there are two ways of matching based on corner features, one is to describe the second feature point set of the extracted reference image and the third feature point set of the subimage to be matched in a predetermined way, for the image The matching is transformed into the matching of two vector sets. The other is to represent the extracted feature point set as a binary image, perform distance transformation on the binary image, and use distance similarity to measure the similarity of two feature point sets. Of course, other methods may also be used, which are not limited in the present invention.

In the embodiment of the present invention, the feature point set of the reference image is obtained by extracting the corner point features of the reference image, the abnormal points in the feature point set that exceed the range of the image to be matched are eliminated, and the feature point set that has eliminated the abnormal point is combined with the feature point set to be matched. The images are image-matched for object tracking. The use of corner features for graphic matching can reduce the amount of data involved in the calculation while retaining the important features of image graphics. The corner features have good recognition and stability, which greatly improves the matching speed. At the same time, the method optimizes the set of feature points in the reference image and eliminates out-of-range corner points, which solves the problem that the out-of-range feature points cannot be matched in the prior art, and improves the reliability of matching.

Further, referring to FIG. 4-FIG. 6, it is another structural schematic diagram of an object tracking device according to an embodiment of the present invention. In addition to the above-mentioned feature extraction module 11, feature optimization module 12 and image matching module 13, the device 1 includes Also includes:

The template determining module 14 is configured to acquire an image sequence captured by a camera, and use the image sequence as an image to be matched; and determine a reference image of the image to be matched from the first frame image of the image sequence.

Concretely, the format of the image sequence acquired by the template determination module 14 to acquire the camera can be a picture format such as BMP, TIFF or JPEG, or a video format such as MPEG or AVI, and the template determination module 14 regards the image sequence as an image to be matched, and from The reference image is determined in the first frame image in the image sequence, and the target to be tracked is recorded in the reference image. Assuming that the size of the reference image is M*N (M, N is the number of pixels), the size of the image to be matched is m*n (m, n is the number of pixels), then M<m, N<n.

A template updating module 15, configured to delete the reference image and use the sub-image to be matched that successfully matches the reference template as a new reference image.

Specifically, the template update module 15 uses the successfully matched sub-image in the image to be matched as a new reference image, and uses the new reference image for image matching when performing target tracking on the next frame of image to be matched. By dynamically updating the reference image, the shape and size of the target in the reference image can be adjusted, making image matching more accurate.

Wherein, feature optimization module 12 includes:

An acquisition unit 121, configured to acquire the coordinates of each feature point in the first set of feature points;

A judging unit 122, configured to judge whether the coordinates of each feature point in the feature point set exceed the range of the image to be matched, and if so, determine that the feature point is an abnormal point;

The elimination unit 123 is configured to eliminate all abnormal points in the first feature point set to obtain a second feature point set.

Image matching module 13 includes:

A selection unit 131, configured to select a sub-image to be matched that is the same size as the reference image in the image to be matched;

An extraction unit 132, configured to use the preset algorithm to extract the features of the subgraph to be matched to obtain a third set of feature points;

A calculation unit 133, configured to calculate the distance between the second set of feature points and the third set of feature points;

The matching unit 134 is used to judge whether the distance is smaller than a preset threshold, if yes, determine that the reference image and the subimage to be matched in the image to be matched are successfully matched, if not, re-execute the The step of selecting a subimage to be matched with the same size as the reference image in the image to be matched is described.

In the embodiment of the present invention, the feature point set of the reference image is obtained by extracting the corner point features of the reference image, the abnormal points in the feature point set that exceed the range of the image to be matched are eliminated, and the feature point set that has eliminated the abnormal point is combined with the feature point set to be matched. The images are image-matched for object tracking. The use of corner features for graphic matching can reduce the amount of data involved in the calculation while retaining the important features of image graphics. The corner features have good recognition and stability, which greatly improves the matching speed. At the same time, the method optimizes the set of feature points in the reference image and eliminates out-of-range corner points, which solves the problem that the out-of-range feature points cannot be matched in the prior art, and improves the reliability of matching.

Referring to FIG. 7 , it is another structural diagram of a target tracking device according to an embodiment of the present invention. The target tracking device 1 includes a processor 61, a memory 62, an input device 63 and an output device 64. The processor in the target tracking device 1 The number of 61 may be one or more, and FIG. 7 takes one processor as an example. In some embodiments of the present invention, the processor 61 , the memory 62 , the input device 63 and the output device 64 may be connected via a bus or in other ways, and bus connection is taken as an example in FIG. 7 .

Wherein, a set of program codes are stored in the memory 62, and the processor 61 is used to call the program codes stored in the memory 62 to perform the following operations:

Extracting the corner features of the reference image by using a preset algorithm to obtain a first set of feature points of the reference image, the reference image is used to track the target in the current image to be matched;

Eliminating abnormal points in the first set of feature points beyond the range of the image to be matched to obtain a second set of feature points;

performing image matching on the second feature point set and the image to be matched, and completing target tracking on the image to be matched according to the matching result.

Further, in some embodiments of the present invention, the processor 61 uses a preset algorithm including Harris corner detection algorithm, FAST corner detection algorithm, KLT corner detection algorithm or SUSAN corner detection algorithm to extract the corner points of the reference image The feature obtains the first set of feature points of the reference image.

Preferably, in some embodiments of the present invention, the processor 61 is also used to execute:

Obtain the image sequence collected by the camera, and use the image sequence as the image to be matched;

And determine the reference image of the image to be matched from the first frame image of the image sequence.

Preferably, in some embodiments of the present invention, the processor 61 executes the step of removing abnormal points in the first set of feature points beyond the range of the image to be matched, and obtaining the second set of feature points includes:

Obtain the coordinates of each feature point in the first set of feature points;

Judging according to the coordinates of each feature point in the feature point set whether it exceeds the scope of the image to be matched, if yes, then determining that the feature point is an abnormal point;

All abnormal points in the first set of feature points are eliminated to obtain a second set of feature points.

Preferably, in some embodiments of the present invention, the processor 61 executes the step of performing image matching on the second feature point set and the image to be matched, and completing the target tracking of the image to be matched according to the matching result include:

Selecting a subimage to be matched with the same size as the reference image in the image to be matched;

Using the preset algorithm to extract the features of the subgraph to be matched to obtain a third set of feature points;

calculating the distance between the second set of feature points and the third set of feature points;

Judging whether the distance is smaller than a preset threshold, if yes, then determining that the reference image is successfully matched with the subimage to be matched in the image to be matched, if not, re-executing the selection process in the image to be matched A step to match subimages of the same size as the reference image.

Preferably, in some embodiments of the present invention, the processor 61 is further configured to delete the reference image and use the subimage to be matched that successfully matches the reference template as a new reference image.

In the embodiment of the present invention, the feature point set of the reference image is obtained by extracting the corner point features of the reference image, the abnormal points in the feature point set that exceed the range of the image to be matched are eliminated, and the feature point set that has eliminated the abnormal point is combined with the feature point set to be matched. The images are image-matched for object tracking. The use of corner features for graphic matching can reduce the amount of data involved in the calculation while retaining the important features of image graphics. The corner features have good recognition and stability, which greatly improves the matching speed. At the same time, the method optimizes the set of feature points in the reference image and eliminates out-of-range corner points, which solves the problem that the out-of-range feature points cannot be matched in the prior art, and improves the reliability of matching.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented through computer programs to instruct related hardware, and the programs can be stored in a computer-readable storage medium. During execution, it may include the processes of the embodiments of the above-mentioned methods. Wherein, the storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM) or a random access memory (Random Access Memory, RAM), etc.

What is disclosed above is only a preferred embodiment of the present invention, and of course it cannot limit the scope of rights of the present invention. Those of ordinary skill in the art can understand all or part of the process for realizing the above embodiments, and according to the rights of the present invention The equivalent changes required still belong to the scope covered by the invention.

Claims (12)

1. A target tracking method, characterized in that, comprising: Extracting the corner features of the reference image by using a preset algorithm to obtain a first set of feature points of the reference image, the reference image is used to track the target in the current image to be matched; Eliminating abnormal points in the first set of feature points beyond the range of the image to be matched to obtain a second set of feature points; performing image matching on the second feature point set and the image to be matched, and completing target tracking on the image to be matched according to the matching result. 2. The method according to claim 1, wherein the preset algorithm comprises: Harris corner detection algorithm, FAST accelerated segmentation detection feature corner detection algorithm, KLT corner detection algorithm or SUSAN minimum kernel value similar area Corner detection algorithm. 3. The method according to claim 1 or 2, wherein, before the step of extracting the corner feature of the reference image using a preset algorithm to obtain the first feature point set of the reference image, further comprising: Obtain the image sequence collected by the camera, and use the image sequence as the image to be matched; And determine the reference image of the image to be matched from the first frame image of the image sequence. 4. The method according to claim 3, wherein the step of removing outliers exceeding the range of the image to be matched in the first set of feature points to obtain a second set of feature points comprises: Obtain the coordinates of each feature point in the first set of feature points; Judging according to the coordinates of each feature point in the feature point set whether it exceeds the scope of the image to be matched, if yes, then determining that the feature point is an abnormal point; All abnormal points in the first set of feature points are eliminated to obtain a second set of feature points. 5. The method according to claim 4, wherein the step of performing image matching on the second feature point set and the image to be matched, and completing the target tracking of the image to be matched according to the matching result comprises : Selecting a subimage to be matched with the same size as the reference image in the image to be matched; Using the preset algorithm to extract the features of the subgraph to be matched to obtain a third set of feature points; calculating the distance between the second set of feature points and the third set of feature points; Judging whether the distance is smaller than a preset threshold, if yes, then determining that the reference image is successfully matched with the subimage to be matched in the image to be matched, if not, re-executing the selection process in the image to be matched A step to match subimages of the same size as the reference image. 6. The method according to claim 5, wherein the second feature point set is image-matched with the image to be matched, and after the step of target tracking of the image to be matched is completed according to the matching result ,Also includes: The reference image is deleted and the subimage to be matched that successfully matches the reference template is used as a new reference image. 7. A target tracking device, characterized in that it comprises: A feature extraction module, configured to extract corner features of a reference image using a preset algorithm to obtain a first set of feature points of the reference image, the reference image being used to track the target in the current image to be matched; A feature optimization module, configured to remove abnormal points in the first set of feature points that exceed the range of the image to be matched to obtain a second set of feature points; An image matching module, configured to perform image matching on the second set of feature points and the image to be matched, and complete target tracking on the image to be matched according to the matching result. 8. The device according to claim 7, wherein the feature extraction module is used to extract the corner feature of the reference image using a preset algorithm to obtain the first set of feature points of the reference image, and the preset algorithm Including Harris corner detection algorithm, FAST accelerated segmentation detection feature corner detection algorithm, KLT corner detection algorithm or SUSAN minimum kernel value similar area corner detection algorithm. 9. The device according to claim 7 or 8, further comprising: The template determining module is used to obtain the image sequence collected by the camera, and use the image sequence as the image to be matched; and determine the reference image of the image to be matched from the first frame image of the image sequence. 10. The device according to claim 9, wherein the feature optimization module comprises: an acquisition unit, configured to acquire the coordinates of each feature point in the first set of feature points; A judging unit, configured to judge whether the coordinates of each feature point in the feature point set exceed the range of the image to be matched, and if so, determine that the feature point is an abnormal point; The elimination unit is configured to eliminate all abnormal points in the first feature point set to obtain a second feature point set. 11. The device according to claim 10, wherein the image matching module comprises: A selecting unit, configured to select a sub-image to be matched that is the same size as the reference image in the image to be matched; An extraction unit, configured to extract features of the to-be-matched subgraph using the preset algorithm to obtain a third set of feature points; a calculation unit, configured to calculate the distance between the second set of feature points and the third set of feature points; A matching unit, configured to judge whether the distance is smaller than a preset threshold, if yes, determine that the reference image is successfully matched with the subimage to be matched in the image to be matched, if not, re-execute the A step of selecting a to-be-matched subimage of the same size as the reference image from the to-be-matched image. 12. The apparatus of claim 11, further comprising: The template updating module is configured to delete the reference image and use the subimage to be matched that successfully matches the reference template as a new reference image.

Images