CN111968154A - HOG-LBP and KCF fused pedestrian tracking method - Google Patents
HOG-LBP and KCF fused pedestrian tracking method Download PDFInfo
- Publication number
- CN111968154A CN111968154A CN202010705263.8A CN202010705263A CN111968154A CN 111968154 A CN111968154 A CN 111968154A CN 202010705263 A CN202010705263 A CN 202010705263A CN 111968154 A CN111968154 A CN 111968154A
- Authority
- CN
- China
- Prior art keywords
- lbp
- hog
- pedestrian
- kcf
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/467—Encoded features or binary features, e.g. local binary patterns [LBP]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a pedestrian tracking method fusing HOG-LBP and KCF, which comprises the steps of obtaining a pedestrian video, setting a tracking target, determining a candidate area, extracting HOG-LBP characteristics of an image of the candidate area, calculating a kernel correlation filter according to the HOG-LBP characteristics of the extracted image, updating a pedestrian position through the obtained kernel correlation detection, updating the filter and outputting a tracking result; the method can solve the problems that in a KCF pedestrian tracking algorithm, when a background covers a messy noise edge, HOG characteristics cannot be processed, so that tracking drift and failure are caused, and the robustness of the algorithm is reduced.
Description
Technical Field
The invention belongs to the technical field of machine vision, and relates to a pedestrian tracking method fusing HOG-LBP and KCF.
Background
With the national proposal of the concept of 'smart city', the research and development of artificial intelligence technology are concerned. Pedestrian tracking is one of the most main research tasks in the field of computer vision of artificial intelligence, has a wide application value in the aspects of automatic driving, intelligent security, intelligent robots and the like, has attracted the attention and research of broad scholars, and has emerged a plurality of related pedestrian tracking algorithms, and the KCF (Kernel Correlation Filter) kernel Correlation filtering algorithm is one of the representatives. However, the conventional KCF tracking algorithm uses a Histogram of Oriented Gradients (HOG) to characterize the pedestrian, which is good for describing image edge information or local shape information, and cannot be processed when the background covers a messy noise edge. This leads to tracking drift and failure, and the algorithm robustness is reduced.
To address this issue, the HOG feature and LBP feature (Local Binary Pattern) are combined herein, i.e., HOG-LBP feature. The LBP feature has gray scale invariance and rotation invariance, etc., and can deal with the noise edge whose background covers disorder. The combination of the two can better represent pedestrians and improve the tracking accuracy.
Disclosure of Invention
The invention aims to provide a pedestrian tracking method fusing HOG-LBP and KCF, which solves the problem of drift in a KCF algorithm and improves the robustness of the tracking algorithm on the premise of ensuring the speed of the algorithm.
The invention adopts the technical scheme that a pedestrian tracking method fused with HOG-LBP and KCF is implemented according to the following steps:
step 1, acquiring a pedestrian video, setting a tracking target and determining a candidate area;
step 2, extracting HOG-LBP characteristics of the candidate region image;
step 3, calculating a kernel correlation filter according to the HOG-LBP characteristics of the image extracted in the step 2;
step 4, updating the pedestrian position through the nuclear correlation detection obtained by the calculation in the step 3;
step 5, updating the filter;
and 6, outputting a tracking result.
The invention is also characterized in that:
the specific content of the step 1 is as follows:
acquiring a pedestrian video, reading in a t-th frame Image1, setting a tracking target, and taking a region where the tracking target is located as a candidate region of a first frame;
wherein the HOG-LBP feature extraction process in the step 2 is divided into three steps: HOG feature extraction, LBP feature extraction and HOG-LBP feature fusion;
the HOG feature extraction step mainly comprises the following parts of image normalization, gradient calculation, direction weight projection based on gradient amplitude and feature vector normalization, and the specific calculation process is as follows:
setting the size of the candidate region to be 80 multiplied by 64, and setting the size of the block to be 8 multiplied by 8, so that the candidate region comprises 80 non-overlapping blocks in total;
calculating the gradient direction and amplitude value of each block, and calculating the gradient by using a simple central symmetry operator [ -1, 0, 1], as shown in the following formula:
where I (x, y) is the pixel value of the image at the point (x, y), θ (x, y) is the gradient direction of the point, and m (x, y) corresponds to the amplitude value of the point;
setting the size of a cell to be 4 multiplied by 4, counting a gradient histogram in each block according to the size of the cell, and projecting a specified weight by applying the amplitude of the gradient;
carrying out contrast normalization on the cells in each overlapped block;
combining the histogram vectors in all blocks to obtain a final HOG characteristic vector;
the specific content of LBP feature extraction is as follows:
the LBP operator is represented by (P, R), where P represents the number of pixels contained within the domain; r denotes the domain radius, the basic LBP operator is the (8, 1) domain:
firstly, the 3 x 3 domain pixel value pi(i=1,2, …, 8) is compared with the central pixel value p0, and thresholding is performed, wherein the calculation formula is as follows:
arranging bi (i is 1, 2, …, 8) in a clockwise direction to obtain an 8-bit binary code, and converting the binary code into a decimal number to obtain a result obtained by calculating a central pixel by an LBP operator;
then, after the pedestrian image is subjected to LBP operator operation, histogram statistics is carried out on the pedestrian image to obtain a histogram feature vector, which can be specifically defined as:
in the formula, n is data of different marks generated by an LBP operator, and a consistent mode operator in a 3 × 3 field is adopted, namely n is 256; when x is true, i (x) is 1; when x is false, i (x) is 0;
dividing the image into regions R0,R1,…,Rm-1The histogram of each region may be defined as:
finally, connecting the sub-region histograms to form a final pedestrian feature vector, and adopting Chi2Statistically to measure the distance between LBP features;
the specific content of HOG-LBP feature fusion is as follows: and performing feature fusion by adopting a weighting mode, wherein a fusion formula is as follows:
wherein m represents the number of classifiers, wiAnd ciRespectively representing the weight and output score of the ith classifierF (C) is the score output after feature fusion; the weight calculation formula is as follows:
in the formula, EiEqual error rate for the ith classifier;
setting m different classifiers with the pedestrian image characteristic of x, and when estimating the real classification discriminant function, having m different discriminant functions:
gi(x)=h(x)+i(x),i=1,2,…,m (8)
where h (x) represents the true classification discriminant function, gi(x) Represents the discriminant function of the ith classifier, andi(x) Is represented as gi(x) An error function from the true function;
after feature fusion, the mean square error of the whole feature fusion system can be expressed as:
the specific contents of the calculation kernel correlation filter in the step 3 are as follows:
setting x to represent HOG-LBP characteristics extracted from an input image, h to represent a correlation filter, and setting x ^ to represent Fourier transform of x, wherein the convolution in a spatial domain is equivalent to multiplication operation among elements in a frequency domain according to a convolution theorem, and the following can be obtained:
in the formula (I), the compound is shown in the specification,representing a convolution, F-1Representing an inverse fourier transform,. representing a multiplication between elements,. representing a complex conjugate; to train the filter, the desired correlation output y is defined, and for a new sample x' of the target, the correlation filter h satisfies the conditionThus, the following results were obtained:
the specific contents of detecting and updating the pedestrian position in the step 4 are as follows: after the kernel correlation filtering h is obtained in the step 3, for the t +1 th frame, the detection is carried out on the image block z with the target position of the frame as the center and the size of M multiplied by N, and the correlation response diagram isThus, a new target position is found by the f (z) maximum corresponding position;
wherein, the specific updating content of the filter in the step 5 is as follows:
updating the filter coefficient alpha and the target appearance model x by adopting a linear interpolation mode, namely:
in the formula, γ represents a learning rate, and t represents a frame number.
The invention has the beneficial effects that:
aiming at the problem that in a KCF pedestrian tracking algorithm, when a background covers a messy noise edge, HOG characteristics cannot be processed, so that tracking drift and failure are caused, and algorithm robustness is reduced, the HOG characteristic and LBP characteristic, namely the HOG-LBP characteristic, are combined, the LBP characteristic has gray scale invariance, rotation invariance and the like, the background can be processed to cover the messy noise edge, the pedestrian can be better characterized by combining the HOG characteristic and the LBP characteristic, and pedestrian tracking accuracy is improved.
Drawings
FIG. 1 is a diagram of a HOG-LBP fusion framework in a HOG-LBP and KCF fusion pedestrian tracking method of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention provides a pedestrian tracking method fused with HOG-LBP and KCF, which is implemented by the following steps:
step 1, acquiring a pedestrian video, reading in a t-th frame Image1, setting a tracking target, and taking an area where the tracking target is located as a candidate area of a first frame;
step 2, the HOG-LBP feature extraction process is divided into three steps: HOG feature extraction, LBP feature extraction and HOG-LBP feature fusion; the specific implementation steps can be divided into the following steps, and the fusion framework is shown in figure 1:
firstly, obtaining HOG feature extraction of an image, wherein the steps mainly comprise image normalization, gradient calculation, direction weight projection based on gradient amplitude and feature vector normalization, and the specific calculation process comprises the following steps:
assuming that the size of the candidate region is 80 × 64, and setting the size of the block to be 8 × 8, the candidate region contains 80 non-overlapping blocks in total;
the gradient direction and magnitude are first calculated on each block, here using a simple central symmetry operator [ -1, 0, 1] to calculate the gradient as shown in the following equation:
where I (x, y) is the pixel value of the image point (x, y), θ (x, y) is the gradient direction of the point, and m (x, y) corresponds to the amplitude value of the point;
then setting the size of the cell to be 4 multiplied by 4, counting a gradient histogram in each block according to the size of the cell, and projecting a specified weight by applying the amplitude of the gradient;
then, carrying out contrast normalization on the cells in each overlapped block to eliminate the influence of illumination;
and finally, combining the histogram vectors in all blocks to obtain a final HOG characteristic vector.
Acquiring LBP characteristics of an image:
the basic idea of LBP is to describe local texture features according to binary codes obtained by comparing central pixel points with pixel points in the circular domain thereof, and an LBP operator is usually represented by (P, R), wherein P represents the number of pixels contained in the domain; r represents the radius of the domain, and the basic LBP operator is (8, 1) domain;
firstly, the 3 x 3 domain pixel value pi(i ═ 1, 2, …, 8) and the central pixel value p0And comparing, and performing thresholding treatment, wherein the calculation formula is as follows:
b is toi(i-1, 2, …, 8) arranging in clockwise direction to obtain an 8-bit binary code, and converting into decimal number to obtain the result obtained by calculating the center pixel by the LBP operator;
after the pedestrian image is subjected to LBP operator operation, histogram statistics is carried out on the pedestrian image to obtain a histogram feature vector, which can be specifically defined as:
where n is the data that the LBP operator produces different labels, a 3 × 3 domain consistent pattern operator is used herein, i.e., n is 256, and when x is true, i (x) is 1; when x is false, i (x) is 0;
to better represent pedestrian features, the image is divided into regions R0,R1,…,Rm-1The histogram of each region may be defined as:
finally, connecting the sub-region histograms to form a final pedestrian feature vector, and measuring the distance between LBP features by adopting chi 2 statistics;
as shown in fig. 1, the HOG-LBP features were fused:
feature fusion is performed in a weighting mode, and a fusion formula is as follows:
wherein m represents the number of classifiers; w is aiAnd ciRespectively representing the weight and the output score of the ith classifier; (c) is the score output after feature fusion, and the weight calculation formula is as follows:
in the formula, EiEqual error rate for the ith classifier;
assuming that there are m different classifiers with a pedestrian image feature of x, when estimating the true classification discriminant function, there are m different discriminant functions:
gi(x)=h(x)+i(x),i=1,2,…,m (8)
wherein h (x) represents the true classification discriminant function; gi(x) A discriminant function representing the ith classifier; whilei(x) Is represented as gi(x) An error function from the true function;
after feature fusion, the mean square error of the whole feature fusion system can be expressed as:
step 3, calculating a kernel correlation filter, wherein the specific contents are as follows:
let x represent HOG-LBP features extracted from the input image, h represent a correlation filter, and assume that x ^ represents Fourier transform of x, according to the convolution theorem, the convolution in the space domain is equivalent to multiplication operation among elements in the frequency domain, so as to obtain
In the formula (I), the compound is shown in the specification,representing a convolution, F-1Representing an inverse Fourier transform, representing a multiplication between elements, representing a complex conjugate, defining a desired correlation output y for training the filter, the correlation filter h satisfying the condition for a new sample x' of the targetThus, the following results were obtained:
when the correlation filter h is solved, the correlation filter h is converted into a ridge regression problem, and a circulation matrix is used for carrying out intensive sampling on samples, so that the nonlinear classification problem can be processed;
and 4, detecting and updating the pedestrian position, wherein the specific implementation mode is as follows:
after the kernel correlation filtering h is obtained in the step 3, for the t +1 th frame, the map with the target position of the frame as the center and the size of M multiplied by NDetection on image block z, correlation response mapThus, a new target position is found by the f (z) maximum corresponding position;
step 5, updating the filter, wherein the specific implementation mode is as follows:
in order to better adapt to the change of the target appearance, the filter coefficient α and the target appearance model x are updated by linear interpolation, that is:
wherein γ represents a learning rate, and t represents a frame number;
and 6, outputting a tracking result.
Claims (9)
1. A pedestrian tracking method fused with HOG-LBP and KCF is characterized by comprising the following steps:
step 1, acquiring a pedestrian video, setting a tracking target and determining a candidate area;
step 2, extracting HOG-LBP characteristics of the candidate region image;
step 3, calculating a kernel correlation filter according to the HOG-LBP characteristics of the image extracted in the step 2;
step 4, updating the pedestrian position through the nuclear correlation detection obtained by the calculation in the step 3;
step 5, updating the filter;
and 6, outputting a tracking result.
2. The pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 1, wherein the detailed contents of step 1 are as follows:
the pedestrian video is acquired, the Image1 of the t-th frame is read, the tracking target is set, and the area where the tracking target is located is used as the candidate area of the first frame.
3. The pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 1, wherein the HOG-LBP feature extraction process in step 2 is divided into three steps: HOG feature extraction, LBP feature extraction and HOG-LBP feature fusion.
4. The pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 3, wherein the HOG feature extraction step mainly comprises image normalization, gradient calculation, gradient magnitude-based direction weight projection and feature vector normalization, and the specific calculation process is as follows:
setting the size of the candidate region to be 80 multiplied by 64, and setting the size of the block to be 8 multiplied by 8, so that the candidate region comprises 80 non-overlapping blocks in total;
calculating the gradient direction and amplitude value of each block, and calculating the gradient by using a simple central symmetry operator [ -1, 0, 1], as shown in the following formula:
where I (x, y) is the pixel value of the image point (x, y), θ (x, y) is the gradient direction of the point, and m (x, y) corresponds to the amplitude value of the point;
setting the size of a cell to be 4 multiplied by 4, counting a gradient histogram in each block according to the size of the cell, and projecting a specified weight by applying the amplitude of the gradient;
carrying out contrast normalization on the cells in each overlapped block;
and combining the histogram vectors in all blocks to obtain a final HOG characteristic vector.
5. The pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 3, wherein the specific content of LBP feature extraction is as follows:
the LBP operator is represented by (P, R), where P represents the number of pixels contained within the domain; r denotes the domain radius, the basic LBP operator is the (8, 1) domain:
firstly, the 3 x 3 domain pixel value pi(i ═ 1, 2, …, 8) and the central pixel value p0And comparing, and performing thresholding treatment, wherein the calculation formula is as follows:
b is toi(i-1, 2, …, 8) arranging in clockwise direction to obtain an 8-bit binary code, and converting into decimal number to obtain the result obtained by calculating the center pixel by the LBP operator;
then, after the pedestrian image is subjected to LBP operator operation, histogram statistics is carried out on the pedestrian image to obtain a histogram feature vector, which can be specifically defined as:
in the formula, n is data of different marks generated by an LBP operator, and a consistent mode operator in a 3 × 3 field is adopted, namely n is 256; when x is true, i (x) is 1; when x is false, i (x) is 0;
dividing the image into regions R0,R1,…,Rm-1The histogram of each region may be defined as:
finally, connecting the sub-region histograms to form a final pedestrian feature vector, and adopting Chi2Statistically to measure between LBP featuresThe distance of (c).
6. The pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 3, wherein the HOG-LBP feature fusion is specifically as follows: and performing feature fusion by adopting a weighting mode, wherein a fusion formula is as follows:
wherein m represents the number of classifiers, wiAnd ciRespectively representing the weight and the output score of the ith classifier, and f (C) is the score output after feature fusion; the weight calculation formula is as follows:
in the formula, EiEqual error rate for the ith classifier;
setting m different classifiers with the pedestrian image characteristic of x, and when estimating the real classification discriminant function, having m different discriminant functions:
gi(x)=h(x)+i(x),i=1,2,…,m (8)
where h (x) represents the true classification discriminant function, gi(x) Represents the discriminant function of the ith classifier, andi(x) Is represented as gi(x) An error function from the true function;
after feature fusion, the mean square error of the whole feature fusion system can be expressed as:
7. the pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 1, wherein the specific contents of the kernel correlation filter in step 3 are:
setting x to represent HOG-LBP characteristics extracted from an input image, h to represent a correlation filter, and setting x ^ to represent Fourier transform of x, wherein the convolution in a spatial domain is equivalent to multiplication operation among elements in a frequency domain according to a convolution theorem, and the following can be obtained:
in the formula (I), the compound is shown in the specification,representing a convolution, F-1Representing an inverse fourier transform,. representing a multiplication between elements,. representing a complex conjugate; to train the filter, the desired correlation output y is defined, and for a new sample x' of the target, the correlation filter h satisfies the conditionThus, the following results were obtained:
8. the pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 1, wherein the specific content of detecting and updating the pedestrian position in step 4 is: after the kernel correlation filtering h is obtained in the step 3, for the t +1 th frame, the detection is carried out on the image block z with the target position of the frame as the center and the size of M multiplied by N, and the correlation response diagram isTherefore, a new target position is found by the f (z) maximum value corresponding position.
9. The pedestrian tracking method fused with HOG-LBP and KCF as claimed in claim 1, wherein the specific contents of the filter update in step 5 are:
updating the filter coefficient alpha and the target appearance model x by adopting a linear interpolation mode, namely:
in the formula, γ represents a learning rate, and t represents a frame number.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010705263.8A CN111968154A (en) | 2020-07-21 | 2020-07-21 | HOG-LBP and KCF fused pedestrian tracking method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010705263.8A CN111968154A (en) | 2020-07-21 | 2020-07-21 | HOG-LBP and KCF fused pedestrian tracking method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111968154A true CN111968154A (en) | 2020-11-20 |
Family
ID=73362762
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010705263.8A Pending CN111968154A (en) | 2020-07-21 | 2020-07-21 | HOG-LBP and KCF fused pedestrian tracking method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111968154A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113284253A (en) * | 2021-05-11 | 2021-08-20 | 西安邮电大学 | AR target tracking method for improving Kernel Correlation Filtering (KCF) algorithm |
CN116740135A (en) * | 2023-05-18 | 2023-09-12 | 中国科学院空天信息创新研究院 | Infrared dim target tracking method and device, electronic equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018048353A1 (en) * | 2016-09-09 | 2018-03-15 | Nanyang Technological University | Simultaneous localization and mapping methods and apparatus |
CN109685073A (en) * | 2018-12-28 | 2019-04-26 | 南京工程学院 | A kind of dimension self-adaption target tracking algorism based on core correlation filtering |
CN110223323A (en) * | 2019-06-02 | 2019-09-10 | 西安电子科技大学 | Method for tracking target based on the adaptive correlation filtering of depth characteristic |
-
2020
- 2020-07-21 CN CN202010705263.8A patent/CN111968154A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018048353A1 (en) * | 2016-09-09 | 2018-03-15 | Nanyang Technological University | Simultaneous localization and mapping methods and apparatus |
CN109685073A (en) * | 2018-12-28 | 2019-04-26 | 南京工程学院 | A kind of dimension self-adaption target tracking algorism based on core correlation filtering |
CN110223323A (en) * | 2019-06-02 | 2019-09-10 | 西安电子科技大学 | Method for tracking target based on the adaptive correlation filtering of depth characteristic |
Non-Patent Citations (2)
Title |
---|
孙玉: ""基于HOG 与LBP 特征的人脸识别方法"", 《计算机工程》, pages 1 - 5 * |
袁 康: ""视频序列中的运动目标跟踪算法研究"", 《中国优秀硕士学位论文全文数据库》, pages 13 - 18 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113284253A (en) * | 2021-05-11 | 2021-08-20 | 西安邮电大学 | AR target tracking method for improving Kernel Correlation Filtering (KCF) algorithm |
CN116740135A (en) * | 2023-05-18 | 2023-09-12 | 中国科学院空天信息创新研究院 | Infrared dim target tracking method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhu et al. | Method of plant leaf recognition based on improved deep convolutional neural network | |
CN110929593B (en) | Real-time significance pedestrian detection method based on detail discrimination | |
CN110334762B (en) | Feature matching method based on quad tree combined with ORB and SIFT | |
CN106599836A (en) | Multi-face tracking method and tracking system | |
Martinović et al. | Real-time detection and recognition of traffic signs | |
CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
CN111968154A (en) | HOG-LBP and KCF fused pedestrian tracking method | |
Drożdż et al. | FPGA implementation of multi-scale face detection using HOG features and SVM classifier | |
Zhao et al. | Real-time moving pedestrian detection using contour features | |
Alsanad et al. | Real-time fuel truck detection algorithm based on deep convolutional neural network | |
CN109726621B (en) | Pedestrian detection method, device and equipment | |
Yuan et al. | Fast QR code detection based on BING and AdaBoost-SVM | |
CN110910497B (en) | Method and system for realizing augmented reality map | |
CN117437691A (en) | Real-time multi-person abnormal behavior identification method and system based on lightweight network | |
CN110866435B (en) | Far infrared pedestrian training method for self-similarity gradient orientation histogram | |
CN107886060A (en) | Pedestrian's automatic detection and tracking based on video | |
CN110334703B (en) | Ship detection and identification method in day and night image | |
Ganapathi et al. | Design and implementation of an automatic traffic sign recognition system on TI OMAP-L138 | |
CN112232162B (en) | Pedestrian detection method and device based on multi-feature fusion cascade classifier | |
Dilawari et al. | Toward generating human-centered video annotations | |
CN108154107A (en) | A kind of method of the scene type of determining remote sensing images ownership | |
Maharani et al. | Deep features fusion for KCF-based moving object tracking | |
CN114792374A (en) | Image recognition method based on texture classification, electronic device and storage medium | |
CN113361422A (en) | Face recognition method based on angle space loss bearing | |
CN112651996A (en) | Target detection tracking method and device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |