CN105809709B - A kind of motion target tracking method based on bit plane - Google Patents
A kind of motion target tracking method based on bit plane Download PDFInfo
- Publication number
- CN105809709B CN105809709B CN201510147895.6A CN201510147895A CN105809709B CN 105809709 B CN105809709 B CN 105809709B CN 201510147895 A CN201510147895 A CN 201510147895A CN 105809709 B CN105809709 B CN 105809709B
- Authority
- CN
- China
- Prior art keywords
- bit
- value
- plane
- expressed
- target
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 78
- 238000009499 grossing Methods 0.000 claims abstract description 19
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000005286 illumination Methods 0.000 abstract description 6
- 230000000694 effects Effects 0.000 description 10
- 238000004422 calculation algorithm Methods 0.000 description 9
- 238000001514 detection method Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241000405217 Viola <butterfly> Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000571 coke Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000037452 priming Effects 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of motion target tracking methods based on bit plane, comprising the following steps: (1) tracking target and hand labeled target position are selected in video first frame；(2) it seeks its brightness bit plane drawn game portion binary pattern bit plane respectively to target area image, then carries out Gaussian smoothing, establish two display models；(3) region of search is determined in the next frame, seeks smoothed out brightness bit plane drawn game portion binary pattern bit plane respectively to it, and searches for the region closest with two display models as tracking target；(4) according to the tracking result in established display model and present frame, display model is updated according to preset renewal rate；(5) jump procedure (3), until all video frames are disposed.The present invention has apparent advantage in tracking accuracy and robustness, efficiently solves the problems, such as in video that illumination condition changes, motion target tracking is difficult when object pose variation and appearance significant changes.
Description
Technical Field
The invention belongs to the field of computer vision and pattern recognition, and particularly relates to a moving target tracking method based on a bit plane.
Background
Target tracking is a hotspot of computer vision research and is widely applied in the fields of video monitoring, video retrieval, traffic monitoring and the like. Currently, moving object tracking methods can be roughly divided into two categories: methods based on discrimination and methods of model matching.
Discrimination-based methods, also known as detection-based tracking, treat the moving object tracking problem as a classification problem, and aim to train a classifier to separate the moving object from the background. In 2004, Avidan introduced an offline trained support vector machine to the optical flow-based tracking method, but when the appearance of the target changed significantly, the tracking drifted. To solve this problem, the classifier needs to be updated online. In 2007, Avidan proposed a method of integrating several weak classifiers into one strong classifier and updating the weak classifiers in real time, which can more accurately distinguish whether each pixel is a target or a background. In 2008, Grabner trained the classifier using a semi-supervised method, and only the sample was manually labeled in the first video frame. However, in these supervised and semi-supervised methods, some useful information has been lost, and if errors occur once in the tracking process, the errors are easy to accumulate, thereby causing the tracking failure. Viola et al propose to detect moving objects using a multi-instance learning method to overcome the problem of drift in the tracking process. In subsequent researches, Babenko, Zhang and the like further improve the multi-example learning method, and the precision and the real-time performance of the algorithm are obviously improved.
The model matching method searches for the most similar region to the target model in each frame as the tracking result. As early as 1996, Black et al proposed a method for target tracking using an off-line learned appearance model, but this method was not able to adapt to changes in the appearance of the target. Subsequently, some adaptive appearance modeling methods are gradually introduced to moving object tracking. With the successful application of sparse representations in multiple fields, many scholars have tried to model moving objects using methods of sparse representation. From 2012, sevillala-Lara, Felsberg and the like successively introduce the concept of the distribution field into the tracking field, improve the tracking algorithm and obtain a better tracking effect.
These algorithms are applied in specific environments, but due to the complexity and changeability of tracking scenes, the influence of illumination change, appearance change, shape change and occlusion on target tracking cannot be effectively solved.
Disclosure of Invention
The invention aims to provide a moving target tracking method based on a bit plane and fusing brightness and LBP characteristics aiming at the problems of illumination condition change, target pose change, obvious appearance change and the like in a video scene, and aims to improve the accuracy of moving target tracking in a complex environment.
In order to achieve the purpose, the invention adopts the following technical scheme:
a moving target tracking method based on a bit plane is characterized by comprising the following steps:
selecting a tracking target in a first frame of a video, and manually marking the position of the target;
step (2) solving a brightness bit plane and a local binary pattern bit plane of the target area image, then performing Gaussian smoothing, and respectively establishing a brightness appearance model and a texture appearance model;
step (3) determining a search region in the next frame, respectively obtaining a brightness bit plane and a local binary pattern bit plane after smoothing, and searching a region which is closest to the brightness appearance model and the texture appearance model established in the step (2) as a tracking target;
step (4) updating the brightness appearance model and the texture appearance model according to the established appearance model and the tracking result in the current frame and the preset updating rate;
step (5), when all the video frames are processed, stopping calculation; otherwise, skipping to the step (3).
Further, the method for manually marking the target position in the step (1) comprises: and selecting a tracking target by using a rectangular frame, and recording two-dimensional coordinates (lx, ly) of the upper left corner of the rectangular frame and the width and height of the rectangular frame.
Further, the method for establishing the brightness appearance model in the step (2) comprises the following steps:
① the luminance of each pixel is expressed using the following formula, i.e. using a natural binary sequence:
wherein I (I, j) representsLuminance of pixel in ith row and jth column, ai,j,kThe value of the kth bit when the pixel brightness is expressed by a natural binary sequence is expressed;
② the natural binary sequence is converted to a binary gray code representation using the following formula:
wherein, bi,j,kThe value of the kth bit when I (I, j) is expressed by binary gray codes is expressed;
③ the luminance of each pixel is projected by bit onto different bit planes, resulting in 8 luminance bit planes, using the following formula:
wherein, BPGC is a luminance bit plane, i and j respectively represent the row and column of the image, k represents the serial number of the bit plane, and k is 0,1, 2.., 7;
④ Gaussian smoothing is performed using the following equation:
wherein M is1In order to be a model of the appearance of the luminance,is a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfThe 1-dimensional gaussian kernel of (a) represents a convolution operation.
Further, the method for establishing the texture appearance model in the step (2) comprises the following steps:
①, calculating local binary pattern characteristics of the target region image, and calculating LBP values centering on all non-edge pixels for the target image by adopting a traditional LBP operator with the size of 3 multiplied by 3;
②, the LBP value of each pixel is expressed using the following formula, i.e., a natural binary sequence:
wherein I '(I, j) represents LBP value, a'i,j,kThe value of the kth bit is expressed when the LBP value of the pixel is expressed by a natural binary sequence;
③ the LBP value is converted to a binary Gray code representation using the following equation:
wherein, b'i,j,kThe value of the kth bit when the LBP value is expressed by binary gray codes is expressed;
④ the LBP value for each pixel is projected by bit to different bit planes, resulting in 8 texture bit planes, using the following formula:
wherein, BPGC' is a texture bit plane, i and j respectively represent rows and columns of an image, k represents a serial number of the bit plane, and k is 0,1,2,. and 7;
⑤ is carried out using the following formulaGaussian smoothing to obtain a texture appearance model M2：
Wherein M is2In order to be a model of the appearance of the texture,is a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfThe 1-dimensional gaussian kernel of (a) represents a convolution operation.
Further, the method for determining the search area in step (3) is as follows:
the position of the target in the above frameThe center of (2) is taken as the center of a circle, r is taken as the radius, all rectangles with the centers falling in the area are taken as candidate targets, and the candidate targets are expressed by the following formula:
wherein,and l (X) represents the position of the target in the previous frame, wherein X of the image block in the current frame is the position, r is the search radius, X represents the set of all candidate image blocks, and | | · |, represents the Euclidean distance.
Further, in the step (3), a smoothed luminance bit plane C of the candidate region is obtained1The method comprises the following steps:
① the luminance of each pixel in the candidate region is represented using the following formula, i.e., the luminance of each pixel is represented using a natural binary sequence:
wherein J (i, J) represents the pixel brightness of the ith row and jth column in the candidate region, di,j,kThe value of the kth bit when the pixel brightness is expressed by a natural binary sequence is expressed;
② the natural binary sequence is converted to a binary gray code representation using the following formula:
wherein e isi,j,kThe value of the kth bit is expressed when J (i, J) is expressed by binary gray codes;
③ the luminance of each pixel in the candidate region is projected by bit onto different bitplanes, resulting in 8 luminance bitplanes, using the following formula:
wherein, BPGC1For the luminance bit-plane, i and j represent the rows and columns of the image, respectively, k represents the bit-plane sequence number, k is 0,1, 2.
④ Gaussian smoothing is performed using the following equation:
whereinIs a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfTo obtain a smoothed luminance bit plane C of the candidate region1。
Further, the local binary pattern bit plane C of the candidate region is obtained in the step (3)2The method comprises the following steps:
① calculating the LBP characteristics of the image in the candidate region by using the conventional LBP operator of 3 x 3 size to calculate the LBP value centering on all non-edge pixels for the candidate region image;
② the LBP value of each pixel in the candidate region is expressed using the following formula, i.e. the LBP value of each pixel is expressed in a natural binary sequence:
wherein J ' (i, J) represents LBP value, d ' of ith row and jth column in candidate region 'i,j,kThe value of the kth bit is expressed when the value is expressed by a natural binary sequence;
③ the LBP value is converted to a binary Gray code representation using the following equation:
wherein, e'i,j,kThe value of the kth bit when the LBP value is expressed by binary gray codes is expressed;
④ the LBP value of each pixel in the candidate region is projected to different bitplanes using the following formula, resulting in 8 texture bitplanes:
wherein, BPGC2For texture bit-planes, i and j denote the rows and columns of the image, respectively, k denotes the bit-plane sequence number, k being 0,1,2
⑤ Gaussian smoothing is performed using the following equation:
whereinIs a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfTo obtain the LBP bit plane C2。
Further, the method for searching the closest region to the brightness appearance model and the texture appearance model established in step (2) in step (3) as the tracking target comprises the following steps: obtaining a region x' where the minimum value is obtained by:
wherein,representing the position of the object in the current frame, M1As a luminance appearance model, M2As a texture appearance model, C1Being a smoothed luminance bit plane, C2Is a smoothed LBP bit-plane, w is a weight and 0 < w < 1, Dist (·) isThe distance between the two bit planes is obtained according to the following formula:
wherein, BP1And BP2I, j and k represent the row number, column number and bit plane number, respectively.
Further, the method for updating the brightness appearance model and the texture appearance model in the step (4) comprises:
wherein λ is the update rate and 0 < λ < 1, M1Updating the pre-current frame luminance appearance model, M2Texture appearance model before update for current frame, M'1Is the updated luminance appearance model of the current frame, M'2For the texture appearance model after updating of the current frame, C1(x') is the luminance bit plane corresponding to the target region in the current frame, C2(x') is the texture bit-plane corresponding to the target region in the current frame.
The invention adopting the technical scheme has the following beneficial effects:
the moving target tracking method based on the bit plane provided by the invention fully utilizes the brightness and LBP texture characteristics of the original image, introduces the uncertainty of the position into the tracking process through convolution operation, and adopts the bit plane method to establish a model for the target appearance, thereby effectively overcoming the adverse effects on the tracking result such as the change of illumination conditions, the change of the target pose, the obvious change of the appearance and the like, and having better accuracy and robustness.
Drawings
Fig. 1 is a flow chart of a moving object tracking method based on a bit plane according to the present invention.
Fig. 2 is a gray scale bit plane according to the present invention.
FIG. 3 is a texture bit plane according to the present invention.
Fig. 4-1 is a graph of center position error on a test video sequence in accordance with the present invention.
Fig. 4-2 is a graph of center position error on a test video sequence in accordance with the present invention.
Fig. 4-3 are graphs of center position error curves on a test video sequence in accordance with the present invention.
Fig. 4-4 are graphs of center position error curves on a test video sequence in accordance with the present invention.
Fig. 4-5 are graphs of center position error curves on a test video sequence in accordance with the present invention.
Fig. 4-6 are graphs of center position error curves on a test video sequence in accordance with the present invention.
Fig. 4-7 are graphs of center position error curves on a test video sequence in accordance with the present invention.
Fig. 4-8 are graphs of center position error curves on a test video sequence in accordance with the present invention.
Fig. 5 is a diagram of tracking comparison on a test video sequence in accordance with the present invention.
Detailed Description
The invention will be described in detail with reference to the following drawings and specific embodiments, which are illustrative and not restrictive.
A method for tracking a moving object based on a bit plane, as shown in fig. 1, includes the following steps:
(1) a tracking target is selected in a first frame of the video, and the position of the target is marked manually.
In the first frame, a tracking target is selected by a rectangular frame, and two-dimensional coordinates (lx, ly) of the upper left corner of the rectangular frame and the width and height of the rectangular frame are recorded.
(2) And solving a brightness bit plane and a local binary pattern bit plane of the target area image, and then performing Gaussian smoothing to respectively establish a brightness appearance model and a texture appearance model.
① creating a brightness appearance model
First, the luminance of each pixel is expressed using the following formula, i.e., the luminance of each pixel is expressed using a natural binary sequence:
wherein I (I, j) represents the pixel brightness of the ith row and the jth column, ai,j,kIndicating the value of the kth bit when the pixel brightness is represented by a natural binary sequence.
Secondly, the natural binary sequence is converted into binary gray code representation by using the following formula:
wherein, bi,j,kThe value of the k bit when I (I, j) is expressed by binary gray code is shown.
Then, the luminance of each pixel is projected by bit to different bit planes using the following formula, thereby obtaining 8 luminance bit planes, as shown in fig. 2:
where BPGC is a luminance bit plane, i and j denote rows and columns of an image, respectively, k denotes a bit plane number, and k is 0,1, 2.
Finally, gaussian smoothing is performed using the following formula:
wherein M is1In order to be a model of the appearance of the luminance,is a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfThe 1-dimensional gaussian kernel of (a) represents a convolution operation.
② creating texture appearance model
First, Local Binary Pattern (LBP) features of the target region image are obtained. With the conventional LBP operator of 3 × 3 size, LBP values centered on all non-edge pixels are calculated for the target image.
Next, the LBP value of each pixel is expressed using the following formula, i.e., using a natural binary sequence:
wherein I '(I, j) represents LBP value, a'i,j,kIndicating the value of the kth bit when the LBP value of the pixel is represented by a natural binary sequence.
Next, the LBP value is converted to a binary Gray code representation using the following equation:
wherein, b'i,j,kIndicating the value of the kth bit when the LBP value is represented by binary gray code.
The LBP value for each pixel is then projected by bit to a different bit plane, resulting in 8 texture bit planes, as shown in fig. 3, using the following formula:
where BPGC' is a texture bit plane, i and j represent the rows and columns of an image, respectively, k represents the bit plane number, and k is 0,1, 2.
Finally, Gaussian smoothing is performed by using the following formula, so that a texture appearance model M is obtained2：
Wherein M is2In order to be a model of the appearance of the texture,is a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfThe 1-dimensional gaussian kernel of (a) represents a convolution operation.
(3) And determining a search area in the next frame, respectively obtaining the smoothed brightness bit plane and the smoothed local binary pattern bit plane, and searching an area which is closest to the two appearance models to be used as a tracking target.
① determining search area
The position of the target in the above frameThe center of (2) is taken as the center of a circle, r is taken as the radius, all rectangles with the centers falling in the area are taken as candidate targets, and the candidate targets are expressed by the following formula:
wherein,and l (X) represents the position of the target in the previous frame, wherein X of the image block in the current frame is the position, r is the search radius, X represents the set of all candidate image blocks, and | | · |, represents the Euclidean distance.
② obtaining a luminance bit plane after candidate region smoothing
First, the luminance of each pixel in the candidate region is represented using the following formula, i.e., the luminance of each pixel is represented using a natural binary sequence:
wherein J (i, J) represents the pixel brightness of the ith row and jth column in the candidate region, di,j,kThe value of the kth bit when the pixel brightness is expressed by a natural binary sequence is expressed;
secondly, the natural binary sequence is converted into binary gray code representation by using the following formula:
wherein e isi,j,kThe value of the kth bit is expressed when J (i, J) is expressed by binary gray codes;
the luminance of each pixel in the candidate region is then projected by bit onto a different bitplane, resulting in 8 luminance bitplanes, using the following formula:
wherein, BPGC1For the luminance bit-planes, i and j denote the rows and columns of the image, respectively, k denotes the bit-plane sequence number, k being 0,1, 2.
Finally, gaussian smoothing is performed using the following formula:
whereinIs a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfTo obtain a smoothed luminance bit plane C of the candidate region1。
③ obtaining smoothed local binary pattern bit plane of candidate region
First, local binary pattern features of a target image are obtained. With the conventional LBP operator of 3 × 3 size, LBP values centered on all non-edge pixels are calculated for the target image.
Then, the LBP value of each pixel in the candidate region is represented using the following formula, i.e. using a natural binary sequence:
wherein J ' (i, J) represents LBP value, d ' of ith row and jth column in candidate region 'i,j,kThe value of the kth bit is shown when the value is represented by a natural binary sequence.
The LBP value is then converted to a binary gray code representation using the following formula:
wherein, e'i,j,kIndicating the value of the kth bit when the LBP value is represented by binary gray code.
Next, the LBP value of each pixel in the candidate region is projected by bit to different bitplanes using the following formula, resulting in 8 texture bitplanes:
wherein, BPGC2For texture bit-planes, i and j represent the rows and columns of the image, respectively, k represents the bit-plane sequence number, k being 0,1, 2.
Finally, gaussian smoothing is performed using the following formula:
whereinIs a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfTo obtain the LBP bit plane C2。
According to the steps, obtaining the smoothed brightness bit plane C of the candidate region1And LBP bit plane C2。
④ determining target position
The method for finding the area closest to the two appearance models as the tracking target is as follows: obtaining a region x' where the minimum value is obtained by:
wherein,representing the position of the object in the current frame, M1As a luminance appearance model, M2As a texture appearance model, C1Being a smoothed luminance bit plane, C2For a smoothed LBP bitplane, w is the weight and 0 < w < 1, Dist (. cndot.) is the distance between the two bitplanes, as follows:
wherein, BP1And BP2I, j and k represent the row number, column number and bit plane number, respectively.
(4) And updating the brightness appearance model and the texture appearance model according to the established appearance model and the moving object in the current frame and the preset updating rate.
Wherein λ is the update rate and 0 < λ < 1, M1Updating the pre-current frame luminance appearance model, M2Texture appearance model before update for current frame, M'1Is the updated luminance appearance model of the current frame, M'2For the texture appearance model after updating of the current frame, C1(x') is the luminance bit plane corresponding to the target region in the current frame, C2(x') is the texture bit-plane corresponding to the target region in the current frame.
(5) When all the video frames are processed, stopping calculation; otherwise, skipping to the step (3).
The bit plane based moving object tracking algorithm is described as follows:
inputting: video sequence V, and the position of an object in a first frame
And (3) outputting: target position of each frame in video sequence
The method comprises the following steps:
(1) initializing a target appearance model
(2)for f＝2to|V|do
(3) Determining a search range to obtain an image set
(4)Computing a luminance bit-plane for each image blockAnd LBP bit plane
(5) The position of the target is determined,
(6) calculating a luminance bit plane C of an object1(x') and LBP bit-plane C2(x′)
(7) Updating the appearance model, M1＝λM1+(1-λ)C1(x') and M2＝λM2+(1-λ)C2(x′)
(8)end for
Example (b):
to evaluate the performance of the present invention, tests were performed on 8 video sequences provided by Babenko et al. The video sequences comprise partial shielding, target deformation, illumination change, size change, rapid movement, similar object interference and the like, and four algorithms which have better tracking effect on the video sequences are respectively compared as comparison: a nuclear-based cyclic structure detection tracking method (CSK), an online AdaBoost tracking method (OAB), a multi-instance learning tracking Method (MIL), and a weighted multi-instance learning tracking method (WMIL) are compared in terms of average error, tracking success rate, and the like, respectively. The method is realized by adopting Matlab7.0.1 programming on an XP operating system, and the computer configuration is a dual-core 2.93GHzCPU and a 2GB memory.
Setting parameters:
the algorithm used for comparison takes the code issued by the author and its parameters provided in the article. The method parameters in the invention are set as follows: search radius r is 30, model update rate λ is 0.95 (except for cliffcar and dolar λ is 0.85), weight w is 0.5, 2-dimensional gaussian kernel size is 9 × 9 and standard deviation σ iss1, 1-dimensional Gaussian kernel size 5 x 1 and standard deviation σ thereoff＝0.625。
Quantitative analysis:
five different tracking methods were compared using an algorithm to track the center offset distance of the position from the true position (see table 1) and the tracking success rate (see table 2). The smaller the value is, the smaller the tracking error is, the closer the tracking error is to the real position of the target, and the more accurate the tracking result is; conversely, the farther the tracking result deviates from the real position, the worse the tracking effect. The tracking success rate is defined as: if the coincidence rate of the tracking rectangular frame and the real position rectangular frame is more than 50%, the tracking is considered to be successful; otherwise, the failure is considered. The larger the value is, the better the tracking effect is, and conversely, the worse the tracking effect is.
TABLE 1 mean error (in pixels) of the tracking results from the true position center distance
Table 2 tracking success rate (%)
Description of the drawings: the best results are shown in tables 1 and 2 in bold and the second best results are shown in italics.
As can be seen from tables 1 and 2, the method of the present invention has better tracking results for the test video. In the five methods, the average position error is the smallest and the tracking success rate is the highest, which also reflects the accuracy and stability of the method in the invention. Fig. 4-1 through 4-8 are graphs (in pixels) of error between the tracking results and the true position of the target for the five methods.
And (3) qualitative analysis:
fig. 5 shows the comparison of tracking effects of partial frames in 8 video sequences by five methods.
The Dollar video sequence contains the deformation of the object and the interference of similar objects. The method and the CSK method are closest to the real position, and the tracking effect is most accurate.
Partial occlusion exists in the Occluded Face and Occluded Face 2 in a relatively long time and a large range, and a target in the Occluded Face 2 has a rotating head and a hat, so that the difficulty of tracking a moving target is increased. Of the five tracking methods, the method of the present invention has the best tracking effect.
Variations in lighting, target size and pose, appearance, etc. are included in the David indor. The other four methods all show different levels of drift, and the method in the invention shows better stability and accuracy.
Cliffbar includes blurring due to fast priming and interference from similar backgrounds. It can be seen that the CSK tracking performed best, the second best in the present invention, while the other three performed poorly.
The fast movement, appearance change, rotation, partial shielding and the like of the target appear in the Coke Can, the tracking difficulty is increased, and the method provided by the invention shows a better tracking result.
Twinings includes changes in appearance and dimensions caused by 360 degrees of rotation and movement of the target. Of the five methods, the method of the present invention is closest to the true position of the target.
Girl includes deformation and scale change caused by rotation, interference of other objects, and complete change of appearance caused by motion, which all increase the tracking difficulty, and five tracking methods all have different degrees of errors, wherein the tracking effect of WMIL is relatively stable.
In general, the tracking algorithm based on the bit plane can overcome the influences caused by illumination condition change, target pose change, obvious appearance change and the like, and has better tracking accuracy and stability in five algorithms.
Claims (9)
1. A moving target tracking method based on a bit plane is characterized by comprising the following steps:
selecting a tracking target in a first frame of a video, and manually marking the position of the target;
step (2) solving a brightness bit plane and a local binary pattern bit plane of the target area image, then performing Gaussian smoothing, and respectively establishing a brightness appearance model and a texture appearance model;
step (3) determining a search region in the next frame, respectively obtaining a brightness bit plane and a local binary pattern bit plane after smoothing, and searching a region which is closest to the brightness appearance model and the texture appearance model established in the step (2) as a tracking target;
step (4) updating the brightness appearance model and the texture appearance model according to the established appearance model and the tracking result in the current frame and the preset updating rate;
step (5), when all the video frames are processed, stopping calculation; otherwise, skipping to the step (3).
2. The bit-plane-based moving object tracking method according to claim 1, wherein: the method for manually marking the target position in the step (1) comprises the following steps: and selecting a tracking target by using a rectangular frame, and recording two-dimensional coordinates (lx, ly) of the upper left corner of the rectangular frame and the width and height of the rectangular frame.
3. The bit-plane-based moving object tracking method according to claim 1, wherein: the method for establishing the brightness appearance model in the step (2) comprises the following steps:
① the luminance of each pixel is expressed using the following formula, i.e. using a natural binary sequence:
wherein I (I, j) represents the pixel brightness of the ith row and the jth column, ai,j,kThe value of the kth bit when the pixel brightness is expressed by a natural binary sequence is expressed;
② the natural binary sequence is converted to a binary gray code representation using the following formula:
wherein, bi,j,kThe value of the kth bit when I (I, j) is expressed by binary gray codes is expressed;
③ the luminance of each pixel is projected by bit onto different bit planes, resulting in 8 luminance bit planes, using the following formula:
wherein, BPGC is a luminance bit plane, i and j respectively represent the row and column of the image, k represents the serial number of the bit plane, and k is 0,1, 2.., 7;
④ Gaussian smoothing is performed using the following equation:
wherein M is1In order to be a model of the appearance of the luminance,is a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfThe 1-dimensional gaussian kernel of (a) represents a convolution operation.
4. The bit-plane-based moving object tracking method according to claim 1, wherein: the method for establishing the texture appearance model in the step (2) comprises the following steps:
①, calculating local binary pattern characteristics of the target region image, and calculating LBP values centering on all non-edge pixels for the target image by adopting a traditional LBP operator with the size of 3 multiplied by 3;
②, the LBP value of each pixel is expressed using the following formula, i.e., a natural binary sequence:
wherein I '(I, j) represents LBP value, a'i,j,kThe value of the kth bit is expressed when the LBP value of the pixel is expressed by a natural binary sequence;
③ the LBP value is converted to a binary Gray code representation using the following equation:
wherein, b'i,j,kThe value of the kth bit when the LBP value is expressed by binary gray codes is expressed;
④ the LBP value for each pixel is projected by bit to different bit planes, resulting in 8 texture bit planes, using the following formula:
wherein, BPGC' is a texture bit plane, i and j respectively represent rows and columns of an image, k represents a serial number of the bit plane, and k is 0,1,2,. and 7;
⑤ Gaussian smoothing is performed using the following formula to obtain a texture appearance model M2：
Wherein M is2In order to be a model of the appearance of the texture,is a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfThe 1-dimensional gaussian kernel of (a) represents a convolution operation.
5. The bit-plane-based moving object tracking method according to claim 1, wherein: the method for determining the search area in the step (3) comprises the following steps:
the position of the target in the above frameThe center of (2) is taken as the center of a circle, r is taken as the radius, all rectangles with the centers falling in the area are taken as candidate targets, and the candidate targets are expressed by the following formula:
wherein,and l (X) represents the position of the target in the previous frame, wherein X of the image block in the current frame is the position, r is the search radius, X represents the set of all candidate image blocks, and | | · |, represents the Euclidean distance.
6. The bit-plane-based moving object tracking method according to claim 1, wherein: the smoothed brightness bit plane C of the candidate region is obtained in the step (3)1The method comprises the following steps:
① the luminance of each pixel in the candidate region is represented using the following formula, i.e., the luminance of each pixel is represented using a natural binary sequence:
wherein J (i, J) represents the pixel brightness of the ith row and jth column in the candidate region, di,j,kThe value of the kth bit when the pixel brightness is expressed by a natural binary sequence is expressed;
② the natural binary sequence is converted to a binary gray code representation using the following formula:
wherein e isi,j,kThe value of the kth bit is expressed when J (i, J) is expressed by binary gray codes;
③ the luminance of each pixel in the candidate region is projected by bit onto different bitplanes, resulting in 8 luminance bitplanes, using the following formula:
wherein, BPGC1For the luminance bit-plane, i and j represent the rows and columns of the image, respectively, k represents the bit-plane sequence number, k is 0,1, 2.
④ Gaussian smoothing is performed using the following equation:
whereinIs a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfTo obtain a smoothed luminance bit plane C of the candidate region1。
7. The bit-plane-based moving object tracking method according to claim 1, wherein: the local binary pattern bit plane C of the candidate region is obtained in the step (3)2The method comprises the following steps:
① calculating the LBP characteristics of the image in the candidate region by using the conventional LBP operator of 3 x 3 size to calculate the LBP value centering on all non-edge pixels for the candidate region image;
② the LBP value of each pixel in the candidate region is expressed using the following formula, i.e. the LBP value of each pixel is expressed in a natural binary sequence:
wherein J ' (i, J) represents LBP value, d ' of ith row and jth column in candidate region 'i,j,kThe value of the kth bit is expressed when the value is expressed by a natural binary sequence;
③ the LBP value is converted to a binary Gray code representation using the following equation:
wherein, e'i,j,kThe value of the kth bit when the LBP value is expressed by binary gray codes is expressed;
④ the LBP value of each pixel in the candidate region is projected to different bitplanes using the following formula, resulting in 8 texture bitplanes:
wherein, BPGC2For texture bit-planes, i and j represent the rows and columns of the image, respectively, k represents the bit-plane sequence number, k is 0,1, 2.
⑤ Gaussian smoothing is performed using the following equation:
whereinIs a mean value of μSStandard deviation of σSThe 2-dimensional gaussian kernel of (a),is a mean value of μfStandard deviation of σfTo obtain the LBP bit plane C2。
8. The bit-plane-based moving object tracking method according to claim 1, wherein: the method for searching the area closest to the brightness appearance model and the texture appearance model established in the step (2) in the step (3) as the tracking target comprises the following steps: obtaining a region x' where the minimum value is obtained by:
wherein,representing the position of the object in the current frame, M1As a luminance appearance model, M2As a texture appearance model, C1Being a smoothed luminance bit plane, C2For a smoothed LBP bitplane, w is the weight and 0 < w < 1, Dist (. cndot.) is the distance between the two bitplanes, as follows:
wherein, BP1And BP2I, j and k represent the row number, column number and bit plane number, respectively.
9. The bit-plane-based moving object tracking method according to claim 1, wherein: the method for updating the brightness appearance model and the texture appearance model in the step (4) comprises the following steps:
wherein λ is the update rate and 0 < λ < 1, M1Updating the pre-current frame luminance appearance model, M2Texture appearance model before updating for current frame, M1'is the updated luminance appearance model of the current frame, M'2For the texture appearance model after updating of the current frame, C1(x') is the luminance bit plane corresponding to the target region in the current frame, C2(x') is the texture bit-plane corresponding to the target region in the current frame.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510147895.6A CN105809709B (en) | 2015-03-31 | 2015-03-31 | A kind of motion target tracking method based on bit plane |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510147895.6A CN105809709B (en) | 2015-03-31 | 2015-03-31 | A kind of motion target tracking method based on bit plane |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105809709A CN105809709A (en) | 2016-07-27 |
| CN105809709B true CN105809709B (en) | 2019-03-29 |
Family
ID=56465572
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510147895.6A Active CN105809709B (en) | 2015-03-31 | 2015-03-31 | A kind of motion target tracking method based on bit plane |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105809709B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109030868A (en) * | 2018-07-06 | 2018-12-18 | 江西洪都航空工业集团有限责任公司 | Plane motion object angular acceleration measurement method in drop-test |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5801970A (en) * | 1995-12-06 | 1998-09-01 | Martin Marietta Corporation | Model-based feature tracking system |
| US6754370B1 (en) * | 2000-08-14 | 2004-06-22 | The Board Of Trustees Of The Leland Stanford Junior University | Real-time structured light range scanning of moving scenes |
| CN101771416A (en) * | 2008-12-29 | 2010-07-07 | 华为技术有限公司 | Bit-plane coding and decoding method, communication system and related equipment |
| CN103123780A (en) * | 2011-11-18 | 2013-05-29 | 中兴通讯股份有限公司 | Image display method and device of mobile terminal |
| CN103841296A (en) * | 2013-12-24 | 2014-06-04 | 哈尔滨工业大学 | Real-time electronic image stabilizing method with wide-range rotation and horizontal movement estimating function |
-
2015
- 2015-03-31 CN CN201510147895.6A patent/CN105809709B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5801970A (en) * | 1995-12-06 | 1998-09-01 | Martin Marietta Corporation | Model-based feature tracking system |
| US6754370B1 (en) * | 2000-08-14 | 2004-06-22 | The Board Of Trustees Of The Leland Stanford Junior University | Real-time structured light range scanning of moving scenes |
| CN101771416A (en) * | 2008-12-29 | 2010-07-07 | 华为技术有限公司 | Bit-plane coding and decoding method, communication system and related equipment |
| CN103123780A (en) * | 2011-11-18 | 2013-05-29 | 中兴通讯股份有限公司 | Image display method and device of mobile terminal |
| CN103841296A (en) * | 2013-12-24 | 2014-06-04 | 哈尔滨工业大学 | Real-time electronic image stabilizing method with wide-range rotation and horizontal movement estimating function |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105809709A (en) | 2016-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Automatic laser profile recognition and fast tracking for structured light measurement using deep learning and template matching | |
| WO2020108362A1 (en) | Body posture detection method, apparatus and device, and storage medium | |
| Georgakis et al. | End-to-end learning of keypoint detector and descriptor for pose invariant 3D matching | |
| Chen et al. | Convolutional regression for visual tracking | |
| Bayraktar et al. | Analysis of feature detector and descriptor combinations with a localization experiment for various performance metrics | |
| US9489561B2 (en) | Method and system for estimating fingerprint pose | |
| CN106778712B (en) | Multi-target detection and tracking method | |
| CN104408760B (en) | A kind of high-precision virtual assembly system algorithm based on binocular vision | |
| Chen et al. | Using FTOC to track shuttlecock for the badminton robot | |
| CN112364881B (en) | Advanced sampling consistency image matching method | |
| CN112329656B (en) | Feature extraction method for human action key frame in video stream | |
| CN105976397B (en) | A kind of method for tracking target | |
| Blank et al. | 6DoF pose-estimation pipeline for texture-less industrial components in bin picking applications | |
| CN113312973A (en) | Method and system for extracting features of gesture recognition key points | |
| He et al. | Fast online multi-pedestrian tracking via integrating motion model and deep appearance model | |
| Dai et al. | An Improved ORB Feature Extraction Algorithm Based on Enhanced Image and Truncated Adaptive Threshold | |
| Li et al. | CFNN: Correlation Filter Neural Network for Visual Object Tracking. | |
| CN113470073A (en) | Animal center tracking method based on deep learning | |
| CN105809709B (en) | A kind of motion target tracking method based on bit plane | |
| Li et al. | Research on hybrid information recognition algorithm and quality of golf swing | |
| CN116994049A (en) | Full-automatic flat knitting machine and method thereof | |
| CN115063724A (en) | Fruit tree ridge identification method and electronic equipment | |
| CN110222570B (en) | Automatic identification method for cargo throwing/kicking behaviors of express industry based on monocular camera | |
| Qu et al. | Visual tracking with genetic algorithm augmented logistic regression | |
| Li et al. | ISFM-SLAM: dynamic visual SLAM with instance segmentation and feature matching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |