技术领域technical field
本发明属于视觉跟踪技术领域，具体涉及一种特征加权融合的自适应尺度变化的核相关滤波的目标跟踪方法。The invention belongs to the technical field of visual tracking, and in particular relates to a target tracking method of feature weighted fusion and adaptive scale-changing kernel correlation filtering.
背景技术Background technique
视觉目标检测与跟踪是计算机视觉领域一个备受关注的新兴研究方向，是智能监控，人机交互，机器人视觉导航等应用的基础。对于图像或视频中的目标检测，目的在于基于目标表现和轮廓区域等信息，准确地对图像中所要研究的目标进行定位，即将目标的定位和分类合二为一。对于视频目标而言，它的基本任务是从图像序列中检测出运动信息，简化图像处理过程，得到所需要的运动矢量，从而识别与跟踪物体。由于运动目标的正确检测与分割影响着运动目标能否被正确跟踪与分类，所以，视觉目标检测是计算机视觉技术的一个重要部分，也是机器学习与人工智能领域的一项重要研究课题Visual object detection and tracking is an emerging research direction in the field of computer vision that has attracted much attention, and it is the basis for applications such as intelligent monitoring, human-computer interaction, and robot visual navigation. For target detection in images or videos, the purpose is to accurately locate the target to be studied in the image based on information such as target performance and contour area, that is, to combine target positioning and classification into one. For video targets, its basic task is to detect motion information from image sequences, simplify the image processing process, and obtain the required motion vectors to identify and track objects. Since the correct detection and segmentation of moving objects affects whether the moving objects can be tracked and classified correctly, visual object detection is an important part of computer vision technology and an important research topic in the field of machine learning and artificial intelligence.
视觉跟踪的主要挑战是在目标对象或背景由于突变，光照变化，遮挡等而发生变化时能够实时的检测并跟踪到目标。为了得到有效并且鲁棒的跟踪方法，进行了很多实验研究。当前对于视觉跟踪的研究，主要分为两类方法：生成模型方法和判别模型方法。The main challenge of visual tracking is to detect and track the target in real time when the target object or background changes due to sudden changes, illumination changes, occlusions, etc. In order to obtain an effective and robust tracking method, many experimental studies have been carried out. The current research on visual tracking is mainly divided into two categories: generative model method and discriminative model method.
生成类方法，在当前帧对目标区域建模，在下一帧寻找与模型最为相似的区域，作为预测位置。生成类方法中比较有名的有卡尔曼滤波，粒子滤波，mean-shift等。The generation method models the target area in the current frame, and finds the area most similar to the model in the next frame as the predicted position. The more famous generation methods include Kalman filter, particle filter, mean-shift, etc.
判别类方法，以当前目标区域为正样本，背景区域为负样本，训练分类器，在下一帧应用训练好的分类器找到最优区域。The discriminative class method takes the current target area as a positive sample and the background area as a negative sample, trains a classifier, and applies the trained classifier to find the optimal area in the next frame.
在最近几年，基于相关滤波的跟踪算法。MOSSE首先将相关滤波引入了目标跟踪问题，此时MOSSE应用的是单通道的灰度特征。In recent years, tracking algorithms based on correlation filtering. MOSSE first introduced correlation filtering into the target tracking problem. At this time, MOSSE applied single-channel grayscale features.
发明内容Contents of the invention
本发明所要解决的技术问题是针对背景技术的不足提供了一种一种基于多特征融合的自适应尺度目标跟踪方法。The technical problem to be solved by the present invention is to provide an adaptive scale target tracking method based on multi-feature fusion in view of the deficiency of the background technology.
本发明为解决上述技术问题采用以下技术方案The present invention adopts the following technical solutions to solve the above-mentioned technical problems
一种基于多特征融合的自适应尺度目标跟踪方法，具体包含如下步骤：A method for adaptive scale target tracking based on multi-feature fusion, which specifically includes the following steps:
步骤1，在视频的初试帧图像中，根据所给出的目标位置和目标窗口尺度信息，采用循环采样提取目标的hog特征，并构建多尺度样本集序列，训练分类器；Step 1, in the initial test frame image of the video, according to the given target position and target window scale information, use cyclic sampling to extract the hog features of the target, and construct a multi-scale sample set sequence to train the classifier;
步骤2，采用多尺度分类器检测当前帧图像中的所有候选基准目标，获取分类器的响应值为一序列矩阵，找出每个矩阵中的最大元素值并进行对比，最大元素值最大的矩阵对应的尺度为新目标的最佳尺度，记为η。Step 2, use a multi-scale classifier to detect all candidate benchmark targets in the current frame image, obtain the response value of the classifier as a sequence matrix, find the maximum element value in each matrix and compare them, the matrix with the largest maximum element value The corresponding scale is the optimal scale of the new target, denoted as η.
步骤3，采用特征融合技术进一步精确定位目标的中心位置,即提取初试帧目标的颜色直方图特征和hog特征，分别训练相关分类器获取响应值，并对其在决策层进行融合，确定目标的位置信息。Step 3, use feature fusion technology to further accurately locate the center position of the target, that is, extract the color histogram feature and hog feature of the target in the initial test frame, train the relevant classifiers to obtain the response value, and fuse them at the decision-making layer to determine the target's location information.
作为本发明一种多特征融合的尺度自适应目标跟踪方法的进一步优选方案，在步骤1中，构建多尺度样本集序列，训练分类器的具体方法如下：As a further preferred solution of the multi-feature fusion scale adaptive target tracking method of the present invention, in step 1, a multi-scale sample set sequence is constructed, and the specific method of training the classifier is as follows:
步骤1.1，根据第一帧图像给出的目标初始位置，以当前尺度为尺度初始值，对原图像进行不同比例的缩放，得到一系列不同尺度的基样本图像序列；Step 1.1, according to the initial position of the target given by the first frame image, the current scale is used as the scale initial value, and the original image is scaled in different proportions to obtain a series of basic sample image sequences of different scales;
所有的缩放比例组成一组向量scalesi＝1±am,m＝0,1,.…M,i＝1,2,…,2M+1，其中a∈(0,1)是缩放比例的补偿，为正时，scalesi>1，表示放大的尺度，为负时，scalesi<1，表示缩小的尺度，2M+1是总的缩放比例数；All scaling ratios form a set of vector scalesi = 1±am,m=0,1,....M,i=1,2,...,2M+1, where a∈(0,1) is the compensation of scaling , when it is positive, scalesi >1 means the scale of enlargement, when it is negative, scalesi <1 means the scale of reduction, 2M+1 is the total number of scaling ratios;
步骤1.2，对当前帧按照设置好的缩放比例进行缩放操作，对缩放后的多尺度图像进行目标采样，得到多尺度基样本序列xi，循环移位构建多尺度样本集序列，多尺度样本集序列是对不同尺度的基样本进行循环移位密采样得到的一系列不同尺度的样本集，通过一个置换矩阵来进行操作:Step 1.2: Scale the current frame according to the set scaling ratio, perform target sampling on the scaled multi-scale image, obtain the multi-scale basic sample sequence xi , construct a multi-scale sample set sequence by circular shifting, multi-scale sample set A sequence is a series of sample sets of different scales obtained by cyclically shifting dense sampling of basic samples of different scales, and is operated through a permutation matrix:
pi上标i表示置换矩阵循环移位的位数，基于x循环移位后的所有样本表示为：xi＝pix,将所有的样本组成一个循环矩阵x，即为样本集：The superscript i of pi indicates the number of bits of the permutation matrix cyclic shift, and all samples after the cyclic shift based on x are expressed as: xi = pi x, Form all samples into a circular matrix x, which is the sample set:
对不同尺度的所有基样本都进行循环移位密采样后就可以得到一个多尺度样本集序列Xk，表示为：After cyclically shifting dense sampling for all basic samples of different scales, a multi-scale sample set sequence Xk can be obtained, expressed as:
其中xk(k＝0,1,…,2M+1)表示基样本序列。Where xk (k=0, 1, . . . , 2M+1) represents a basic sample sequence.
作为本发明一种多特征融合的尺度自适应目标跟踪方法的进一步优选方案，把跟踪问题看为一个线性回归模型的求解，构建线性分类模型f(x)＝wTx，对于RLS线性分类器,记y为回归类标签，找到对应的权重w，使得函数f(x)＝wTx在样本x上的响应与y的平方误差最小，即有其中λ是正则化项，则有解w＝(XHX+λI)-1XHy，其中XH＝(X*)T，*表示复共轭，T表示转置。As a further optimal scheme of the scale adaptive target tracking method of a kind of multi-feature fusion of the present invention, the tracking problem is regarded as the solution of a linear regression model, and the linear classification model f(x)=wT x is constructed, for the RLS linear classifier , record y as the regression class label, find the corresponding weight w, so that the square error between the response of the function f(x)=wT x on the sample x and y is the smallest, that is, Where λ is a regularization term, then there is a solution w=(XH X+λI)-1 XH y, where XH =(X* )T , * means complex conjugate, and T means transpose.
作为本发明一种多特征融合的尺度自适应目标跟踪方法的进一步优选方案，通过非线性映射函数将w表示为x的高维特征矩阵的线性加权利用核函数解决高维特征矩阵点积计算问，此时，对w的求解转换成对权重系数矩阵α的求解,α＝(K+λI)-1y，其中，α是系数矩阵；K是核矩阵；y是回归类标签矩阵，利用K为循环矩阵这一特性可实现快速计算：其中kxx是核矩阵K＝C(kxx)的第一行。As a further preferred solution of the multi-feature fusion scale-adaptive target tracking method of the present invention, w is expressed as a high-dimensional feature matrix of x through a nonlinear mapping function linear weighting of Using kernel function To solve the high-dimensional feature matrix dot product calculation problem, at this time, the solution to w is transformed into the solution to the weight coefficient matrix α, α=(K+λI)-1 y, where α is the coefficient matrix; K is the kernel matrix; y is the regression class label matrix, using the feature that K is a circulant matrix can realize fast calculation: where kxx is the first row of the kernel matrix K=C(kxx ).
作为本发明一种多特征融合的尺度自适应目标跟踪方法的进一步优选方案，根据核相关滤波的方法，对每个尺度的循环样本集都进行岭回归分析，得到一组不同尺度的参数：其中，2M+1为总的尺度数，yk和kk分别为对应尺度的标记和自相关核。As a further preferred solution of the scale-adaptive target tracking method of multi-feature fusion in the present invention, according to the method of kernel correlation filtering, the ridge regression analysis is performed on the circular sample set of each scale to obtain a set of parameters of different scales: Among them, 2M+1 is the total number of scales, and yk and kk are the markers and autocorrelation kernels of the corresponding scales, respectively.
作为本发明一种多特征融合的尺度自适应目标跟踪方法的进一步优选方案，多尺度核相关滤波分类器为：根据多尺度核相关滤波分类器的响应值进行目标的检测和尺度估计，本专利中多尺度核相关滤波分类器的响应值为一序列代表可信度的矩阵，矩阵最大元素值为目标最佳尺度和最佳位置的可信度，最大元素值在矩阵中的位置即新目标的最佳中心位置。As a further preferred solution of the multi-feature fusion scale-adaptive target tracking method of the present invention, the multi-scale kernel correlation filter classifier is: Target detection and scale estimation are performed according to the response value of the multi-scale kernel correlation filter classifier. In this patent, the response value of the multi-scale kernel correlation filter classifier is a sequence of matrices representing credibility, and the maximum element value of the matrix is the best for the target. The scale and the credibility of the best position, the position of the maximum element value in the matrix is the best center position of the new target.
作为本发明一种多特征融合的尺度自适应目标跟踪方法的进一步优选方案，在决策层对颜色特征和直方图特征进行融合：分别提取颜色特征和HOG特征后，单独训练相关分类器，得到的响应值最大的位置，作为相应的跟踪结果，再根据响应值的大小来计算两种特征各自的权重，计算如下式：As a further preferred solution of the scale-adaptive target tracking method of multi-feature fusion in the present invention, the color feature and histogram feature are fused at the decision-making layer: after extracting the color feature and HOG feature respectively, the relevant classifier is trained separately, and the obtained The position with the largest response value is used as the corresponding tracking result, and then the respective weights of the two features are calculated according to the size of the response value, and the calculation is as follows:
根据所得权重对最终的目标跟踪结果进行加权融合，在数学上表示为：According to the obtained weights, the final target tracking results are weighted and fused, which is expressed mathematically as:
p＝δpc+(1-δ)ph (5)p=δpc +(1-δ)ph (5)
其中，δ＝[0,1],若δ＝0，则代表该跟踪结果仅仅使用了HOG特征的跟踪结果，δ＝1表示跟踪结果仅仅使用了颜色特征的跟踪结果。Among them, δ=[0,1], if δ=0, it means that the tracking result only uses the tracking result of the HOG feature, and δ=1 means that the tracking result only uses the tracking result of the color feature.
本发明采用以上技术方案与现有技术相比，具有以下技术效果：Compared with the prior art, the present invention adopts the above technical scheme and has the following technical effects:
本发明可以很好地处理目标跟踪过程中的尺度变化问题，并能较好的处理跟踪中的遮挡问题，实现目标的精确跟踪；The present invention can well deal with the scale change problem in the target tracking process, and can better deal with the occlusion problem in the tracking process, so as to realize the precise tracking of the target;
CSK算法在MOSSE算法的基础上进行了改进，对样本进行密集采样并运用了核技巧来进行运算，大大提高了运算速度。KCF算法基于CSK算法，采用多通道特征连接和高斯核函数，进一步提升了跟踪性能；The CSK algorithm is improved on the basis of the MOSSE algorithm, and the samples are intensively sampled and the kernel technique is used for calculation, which greatly improves the calculation speed. Based on the CSK algorithm, the KCF algorithm adopts multi-channel feature connection and Gaussian kernel function to further improve the tracking performance;
针对目标跟踪中普遍存在的尺度变换问题和遮挡问题，在KCF算法的基础上，提出了一种多特征融合的自适应尺度变化的跟踪方法来弥补KCF算法的不足，能够较有效地应对目标尺度变化和遮挡的情况。Aiming at the ubiquitous scale transformation and occlusion problems in target tracking, on the basis of the KCF algorithm, a multi-feature fusion adaptive scale change tracking method is proposed to make up for the shortcomings of the KCF algorithm, which can effectively deal with the target scale. Variations and occlusions.
附图说明Description of drawings
图1是本发明的流程示意图；Fig. 1 is a schematic flow sheet of the present invention;
图2为本发明对存在尺度变化和遮挡情况的视频序列的跟踪结果图。FIG. 2 is a diagram of the tracking results of the present invention for video sequences with scale changes and occlusions.
具体实施方式Detailed ways
下面结合附图对本发明的技术方案做进一步的详细说明：Below in conjunction with accompanying drawing, technical scheme of the present invention is described in further detail:
下面结合附图对本发明做进一步描述。The present invention will be further described below in conjunction with the accompanying drawings.
如图1所示，一种多特征融合的自适应尺度跟踪的方法如下，包括如下步骤：As shown in Figure 1, a multi-feature fusion adaptive scale tracking method is as follows, including the following steps:
输入视频序列，获取目标位置和目标窗口尺度信息；Input a video sequence to obtain target position and target window scale information;
对于视频中的初始帧图像，先转为灰度图，For the initial frame image in the video, first convert it to a grayscale image,
根据第一帧图像给出的目标初始位置，以当前尺度为尺度初始值，对原图像进行不同比例的缩放，得到一系列不同尺度的基样本图像序列；According to the initial position of the target given by the first frame image, the current scale is used as the scale initial value, and the original image is scaled in different proportions to obtain a series of basic sample image sequences of different scales;
所有的缩放比例组成一组向量scalesi＝1±am,m＝0,1,.…M,i＝1,2,…,2M+1，其中a∈(0,1)是缩放比例的补偿，为正时，scalesi>1，表示放大的尺度，为负时，scalesi<1，表示缩小的尺度，2M+1是总的缩放比例数。All scaling ratios form a set of vector scalesi = 1±am,m=0,1,....M,i=1,2,...,2M+1, where a∈(0,1) is the compensation of scaling , when it is positive, scalesi >1 means the enlarged scale, when it is negative, scalesi <1 means the reduced scale, and 2M+1 is the total scaling number.
如图2所示，本发明对存在尺度变化和遮挡情况的视频序列的跟踪结果图，对当前帧按照设置好的缩放比例进行缩放操作。对缩放后的多尺度图像进行目标采样，得到多尺度基样本序列xi。循环移位构建多尺度样本集序列，多尺度样本集序列是对不同尺度的基样本进行循环移位密采样得到的一系列不同尺度的样本集。这可以通过一个置换矩阵来进行操作:As shown in FIG. 2 , the present invention performs a scaling operation on the current frame according to the set scaling ratio for the tracking result graph of the video sequence with scale change and occlusion. Perform target sampling on the scaled multi-scale image to obtain the multi-scale base sample sequence xi . The cyclic shift constructs a multi-scale sample set sequence, which is a series of sample sets of different scales obtained by performing cyclic shift dense sampling on the base samples of different scales. This can be done with a permutation matrix:
pi上标i表示置换矩阵循环移位的位数，基于x循环移位后的所有样本表示为：xi＝pix,The superscript i of pi indicates the number of bits of the permutation matrix cyclic shift, and all samples after the cyclic shift based on x are expressed as: xi = pi x,
将所有的样本组成一个循环矩阵X，即为样本集：Form all samples into a circular matrix X, which is the sample set:
对不同尺度的所有基样本都进行循环移位密采样后就可以得到一个多尺度样本集序列Xk，表示为After performing cyclic shift dense sampling on all basic samples of different scales, a multi-scale sample set sequence Xk can be obtained, expressed as
其中xk(k＝0,1,…,2M+1)表示基样本序列，where xk (k=0,1,...,2M+1) represents the basic sample sequence,
根据核相关滤波的方法，对每个尺度的循环样本集都进行岭回归，得到一组不同尺度的模型参数：According to the method of kernel correlation filtering, ridge regression is performed on the circular sample set of each scale to obtain a set of model parameters of different scales:
其中，2M+1为总的尺度数，yk和kk分别为对应尺度的标记和自相关核。Among them, 2M+1 is the total number of scales, and yk and kk are the markers and autocorrelation kernels of the corresponding scales, respectively.
把跟踪问题看为一个线性回归模型的求解，构建线性分类模型f(x)＝wTx，对于RLS线性分类器,记y为回归类标签，找到对应的权重w，使得函数f(x)＝wTx在样本x上的响应与y的平方误差最小，即有其中λ是正则化项，则有解w＝(XHX+λI)-1XHy，其中XH＝(X*)T，*表示复共轭，T表示转置。Think of the tracking problem as the solution of a linear regression model, construct a linear classification model f(x)=wT x, for the RLS linear classifier, record y as the regression class label, find the corresponding weight w, so that the function f(x) =wT The square error between the response of x on sample x and y is the smallest, that is, Where λ is a regularization term, then there is a solution w=(XH X+λI)-1 XH y, where XH =(X* )T , * means complex conjugate, and T means transpose.
通过非线性映射函数将w表示为x的高维特征矩阵的线性加权利用核函数解决高维特征矩阵点积计算问题。此时，对w的求解转换成对权重系数矩阵α的求解,α＝(K+λI)-1y，其中，α是系数矩阵；K是核矩阵；y是回归类标签矩阵。利用K为循环矩阵这一特性可实现快速计算：其中kxx是核矩阵K＝C(kxx)的第一行。Represent w as a high-dimensional feature matrix of x through a nonlinear mapping function linear weighting of Using kernel function Solve the high-dimensional feature matrix dot product calculation problem. At this point, the solution to w is converted to the solution to the weight coefficient matrix α, α=(K+λI)-1 y, where α is the coefficient matrix; K is the kernel matrix; y is the regression class label matrix. Using the feature that K is a circulant matrix can realize fast calculation: where kxx is the first row of the kernel matrix K=C(kxx ).
根据核相关滤波的方法，对每个尺度的循环样本集都进行岭回归分析，得到一组不同尺度的参数：其中，2M+1为总的尺度数，yk和kk分别为对应尺度的标记和自相关核。According to the method of kernel correlation filtering, the ridge regression analysis is performed on the circular sample set of each scale to obtain a set of parameters of different scales: Among them, 2M+1 is the total number of scales, and yk and kk are the markers and autocorrelation kernels of the corresponding scales, respectively.
多尺度核相关滤波分类器为：根据多尺度核相关滤波分类器的响应值进行目标的检测和尺度估计。本专利中多尺度核相关滤波分类器的响应值为一序列代表可信度的矩阵，矩阵最大元素值为目标最佳尺度和最佳位置的可信度，最大元素值在矩阵中的位置即新目标的最佳中心位置。The multi-scale kernel correlation filtering classifier is: Target detection and scale estimation are performed according to the response value of the multi-scale kernel correlation filtering classifier. In this patent, the response value of the multi-scale kernel correlation filter classifier is a sequence of matrices representing credibility, the maximum element value of the matrix is the credibility of the target’s optimal scale and position, and the position of the maximum element value in the matrix is Best central location for new targets.
在决策层对颜色特征和直方图特征进行融合；Fusion of color features and histogram features at the decision-making level;
分别提取颜色特征和HOG特征后，单独训练相关分类器，得到的响应值最大的位置，作为相应的跟踪结果，再根据响应值的大小来计算两种特征各自的权重，计算如下式：After extracting the color feature and HOG feature respectively, train the relevant classifier separately, and obtain the position with the largest response value as the corresponding tracking result, and then calculate the respective weights of the two features according to the size of the response value, and the calculation is as follows:
根据所得权重对最终的目标跟踪结果进行加权融合，在数学上表示为：According to the obtained weights, the final target tracking results are weighted and fused, which is expressed mathematically as:
p＝δpc+(1-δ)php=δpc +(1-δ)ph
其中，δ＝[0,1],若δ＝0，则代表该跟踪结果仅仅使用了HOG特征的跟踪结果，δ＝1表示跟踪结果仅仅使用了颜色特征的跟踪结果。Among them, δ=[0,1], if δ=0, it means that the tracking result only uses the tracking result of the HOG feature, and δ=1 means that the tracking result only uses the tracking result of the color feature.
目标跟踪结果的位置由特征决策融合的位置确定，尺度为多尺度样本训练分类的结果。上述实施例得到的结果示例如下表所示。表1为本发明和其他算法的精确度对比表；表2为本发明和其他算法的跟踪速度的对比表；The position of the target tracking result is determined by the position of feature decision fusion, and the scale is the result of multi-scale sample training classification. Examples of the results obtained in the above embodiments are shown in the table below. Table 1 is the accuracy comparison table of the present invention and other algorithms; Table 2 is the comparison table of the tracking speed of the present invention and other algorithms;
表1Table 1
表2Table 2
综上所述，本发明可以很好地处理目标跟踪过程中存在的尺度变化和遮挡问题，能基本实现基于尺度的自适应跟踪框大小的改变并能较好地应对遮挡问题。To sum up, the present invention can well deal with the scale change and occlusion problems existing in the target tracking process, can basically realize the scale-based adaptive tracking frame size change and can better deal with the occlusion problem.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810161561.8ACN108510521A (en) | 2018-02-27 | 2018-02-27 | A kind of dimension self-adaption method for tracking target of multiple features fusion |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810161561.8ACN108510521A (en) | 2018-02-27 | 2018-02-27 | A kind of dimension self-adaption method for tracking target of multiple features fusion |
| Publication Number | Publication Date |
|---|---|
| CN108510521Atrue CN108510521A (en) | 2018-09-07 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810161561.8APendingCN108510521A (en) | 2018-02-27 | 2018-02-27 | A kind of dimension self-adaption method for tracking target of multiple features fusion |
| Country | Link |
|---|---|
| CN (1) | CN108510521A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109445453A (en)* | 2018-09-12 | 2019-03-08 | 湖南农业大学 | A kind of unmanned plane Real Time Compression tracking based on OpenCV |
| CN109492537A (en)* | 2018-10-17 | 2019-03-19 | 桂林飞宇科技股份有限公司 | A kind of object identification method and device |
| CN109816698A (en)* | 2019-02-25 | 2019-05-28 | 南京航空航天大学 | A UAV Visual Target Tracking Method Based on Scale Adaptive Kernel Correlation Filtering |
| CN109886357A (en)* | 2019-03-13 | 2019-06-14 | 哈尔滨工程大学 | An adaptive weight deep learning target classification method based on feature fusion |
| CN109886130A (en)* | 2019-01-24 | 2019-06-14 | 上海媒智科技有限公司 | Target object determination method, device, storage medium and processor |
| CN109934147A (en)* | 2019-03-05 | 2019-06-25 | 北京联合大学 | Target detection method, system and device based on deep neural network |
| CN110147768A (en)* | 2019-05-22 | 2019-08-20 | 云南大学 | A kind of method for tracking target and device |
| CN110428450A (en)* | 2019-08-01 | 2019-11-08 | 中国矿业大学 | Dimension self-adaption method for tracking target applied to the mobile inspection image of mine laneway |
| CN110533689A (en)* | 2019-08-08 | 2019-12-03 | 河海大学 | Core correlation filtering Method for Underwater Target Tracking based on space constraint adaptive scale |
| CN110598715A (en)* | 2019-09-04 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Image recognition method and device, computer equipment and readable storage medium |
| CN111340842A (en)* | 2020-02-17 | 2020-06-26 | 江南大学 | A Correlation Filtering Target Tracking Algorithm Based on Joint Model |
| CN111368755A (en)* | 2020-03-09 | 2020-07-03 | 山东大学 | A vision-based approach to autonomous pedestrian following for a quadruped robot |
| CN111479062A (en)* | 2020-04-15 | 2020-07-31 | 上海摩象网络科技有限公司 | Method, device and handheld camera for displaying tracking frame of target object |
| CN111680713A (en)* | 2020-04-26 | 2020-09-18 | 中国科学院上海微系统与信息技术研究所 | A UAV-to-ground target tracking and approximation method based on visual detection |
| CN110097579B (en)* | 2019-06-14 | 2021-08-13 | 中国科学院合肥物质科学研究院 | Multi-scale vehicle tracking method and device based on pavement texture context information |
| CN113297982A (en)* | 2021-05-27 | 2021-08-24 | 南京航空航天大学 | Target detection method for improving combination of KCF and DSST in aerial photography |
| CN115240128A (en)* | 2022-06-06 | 2022-10-25 | 南京航空航天大学 | A Target Tracking Method Based on Scale Adaptive |
| CN116343915A (en)* | 2023-03-15 | 2023-06-27 | 电子科技大学长三角研究院(衢州) | Construction method of biological sequence integrated classifier and biological sequence prediction classification method |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106023248A (en)* | 2016-05-13 | 2016-10-12 | 上海宝宏软件有限公司 | Real-time video tracking method |
| US20160342837A1 (en)* | 2015-05-19 | 2016-11-24 | Toyota Motor Engineering & Manufacturing North America, Inc. | Apparatus and method for object tracking |
| CN107423702A (en)* | 2017-07-20 | 2017-12-01 | 西安电子科技大学 | Video target tracking method based on TLD tracking systems |
| CN107481264A (en)* | 2017-08-11 | 2017-12-15 | 江南大学 | A kind of video target tracking method of adaptive scale |
| CN107657630A (en)* | 2017-07-21 | 2018-02-02 | 南京邮电大学 | A kind of modified anti-shelter target tracking based on KCF |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160342837A1 (en)* | 2015-05-19 | 2016-11-24 | Toyota Motor Engineering & Manufacturing North America, Inc. | Apparatus and method for object tracking |
| CN106023248A (en)* | 2016-05-13 | 2016-10-12 | 上海宝宏软件有限公司 | Real-time video tracking method |
| CN107423702A (en)* | 2017-07-20 | 2017-12-01 | 西安电子科技大学 | Video target tracking method based on TLD tracking systems |
| CN107657630A (en)* | 2017-07-21 | 2018-02-02 | 南京邮电大学 | A kind of modified anti-shelter target tracking based on KCF |
| CN107481264A (en)* | 2017-08-11 | 2017-12-15 | 江南大学 | A kind of video target tracking method of adaptive scale |
| Title |
|---|
| JIANYU SHEN等: "A novel correlation filter based on kernel regression for visual tracking", 《ICIC EXPRESS LETTERS》* |
| MENGDAN ZHANG等: "Robust visual tracking using joint scale-spatial correlation filters", 《2015 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)》* |
| 赵璐璐: "基于相关滤波的目标跟踪算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》* |
| 钱堂慧等: "核相关滤波跟踪算法的尺度自适应改进", 《计算机应用》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109445453A (en)* | 2018-09-12 | 2019-03-08 | 湖南农业大学 | A kind of unmanned plane Real Time Compression tracking based on OpenCV |
| CN109492537A (en)* | 2018-10-17 | 2019-03-19 | 桂林飞宇科技股份有限公司 | A kind of object identification method and device |
| CN109492537B (en)* | 2018-10-17 | 2023-03-14 | 桂林飞宇科技股份有限公司 | Object identification method and device |
| CN109886130A (en)* | 2019-01-24 | 2019-06-14 | 上海媒智科技有限公司 | Target object determination method, device, storage medium and processor |
| CN109816698A (en)* | 2019-02-25 | 2019-05-28 | 南京航空航天大学 | A UAV Visual Target Tracking Method Based on Scale Adaptive Kernel Correlation Filtering |
| CN109934147A (en)* | 2019-03-05 | 2019-06-25 | 北京联合大学 | Target detection method, system and device based on deep neural network |
| CN109886357A (en)* | 2019-03-13 | 2019-06-14 | 哈尔滨工程大学 | An adaptive weight deep learning target classification method based on feature fusion |
| CN109886357B (en)* | 2019-03-13 | 2022-12-13 | 哈尔滨工程大学 | An Adaptive Weighted Deep Learning Target Classification Method Based on Feature Fusion |
| CN110147768B (en)* | 2019-05-22 | 2021-05-28 | 云南大学 | Target tracking method and device |
| CN110147768A (en)* | 2019-05-22 | 2019-08-20 | 云南大学 | A kind of method for tracking target and device |
| CN110097579B (en)* | 2019-06-14 | 2021-08-13 | 中国科学院合肥物质科学研究院 | Multi-scale vehicle tracking method and device based on pavement texture context information |
| CN110428450A (en)* | 2019-08-01 | 2019-11-08 | 中国矿业大学 | Dimension self-adaption method for tracking target applied to the mobile inspection image of mine laneway |
| CN110428450B (en)* | 2019-08-01 | 2021-11-16 | 中国矿业大学 | Scale-adaptive target tracking method applied to mine tunnel mobile inspection image |
| CN110533689A (en)* | 2019-08-08 | 2019-12-03 | 河海大学 | Core correlation filtering Method for Underwater Target Tracking based on space constraint adaptive scale |
| CN110598715B (en)* | 2019-09-04 | 2025-02-21 | 腾讯科技（深圳）有限公司 | Image recognition method, device, computer equipment and readable storage medium |
| CN110598715A (en)* | 2019-09-04 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Image recognition method and device, computer equipment and readable storage medium |
| CN111340842B (en)* | 2020-02-17 | 2023-04-18 | 江南大学 | Correlation filtering target tracking method based on joint model |
| CN111340842A (en)* | 2020-02-17 | 2020-06-26 | 江南大学 | A Correlation Filtering Target Tracking Algorithm Based on Joint Model |
| CN111368755A (en)* | 2020-03-09 | 2020-07-03 | 山东大学 | A vision-based approach to autonomous pedestrian following for a quadruped robot |
| CN111479062A (en)* | 2020-04-15 | 2020-07-31 | 上海摩象网络科技有限公司 | Method, device and handheld camera for displaying tracking frame of target object |
| CN111680713B (en)* | 2020-04-26 | 2023-11-03 | 中国科学院上海微系统与信息技术研究所 | A UAV ground target tracking and approximation method based on visual detection |
| CN111680713A (en)* | 2020-04-26 | 2020-09-18 | 中国科学院上海微系统与信息技术研究所 | A UAV-to-ground target tracking and approximation method based on visual detection |
| CN113297982A (en)* | 2021-05-27 | 2021-08-24 | 南京航空航天大学 | Target detection method for improving combination of KCF and DSST in aerial photography |
| CN115240128A (en)* | 2022-06-06 | 2022-10-25 | 南京航空航天大学 | A Target Tracking Method Based on Scale Adaptive |
| CN116343915A (en)* | 2023-03-15 | 2023-06-27 | 电子科技大学长三角研究院(衢州) | Construction method of biological sequence integrated classifier and biological sequence prediction classification method |
| CN116343915B (en)* | 2023-03-15 | 2023-11-24 | 电子科技大学长三角研究院(衢州) | Construction method of biological sequence integrated classifier and biological sequence prediction classification method |
| Publication | Publication Date | Title |
|---|---|---|
| CN108510521A (en) | A kind of dimension self-adaption method for tracking target of multiple features fusion | |
| Yang et al. | Class-agnostic few-shot object counting | |
| CN108665481B (en) | Adaptive anti-occlusion infrared target tracking method based on multi-layer deep feature fusion | |
| CN110097075B (en) | Deep learning-based marine mesoscale vortex classification identification method | |
| CN104281853B (en) | A kind of Activity recognition method based on 3D convolutional neural networks | |
| CN103632382B (en) | A kind of real-time multiscale target tracking based on compressed sensing | |
| Xiao et al. | Based on grid-search and PSO parameter optimization for Support Vector Machine | |
| CN109961024A (en) | Weed detection method in wheat field based on deep learning | |
| Zhao et al. | Cbph-net: A small object detector for behavior recognition in classroom scenarios | |
| CN110728694B (en) | Long-time visual target tracking method based on continuous learning | |
| CN112200121B (en) | A hyperspectral unknown target detection method based on EVM and deep learning | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN108182388A (en) | A kind of motion target tracking method based on image | |
| CN109859241B (en) | Adaptive Feature Selection and Temporal Consistency Robust Correlation Filtering for Visual Tracking | |
| CN112052772A (en) | Face shielding detection algorithm | |
| CN107067410B (en) | Manifold regularization related filtering target tracking method based on augmented samples | |
| Su et al. | FSRDD: An efficient few-shot detector for rare city road damage detection | |
| CN108038515A (en) | Unsupervised multi-target detection tracking and its storage device and camera device | |
| CN112785622A (en) | Long-time tracking method and device for unmanned ship on water surface and storage medium | |
| Kumar et al. | Mediapipe and cnns for real-time asl gesture recognition | |
| Suliman et al. | Arabic sign language recognition using deep machine learning | |
| CN110942472A (en) | A Kernel Correlation Filter Tracking Method Based on Feature Fusion and Adaptive Blocking | |
| CN106097381A (en) | A kind of method for tracking target differentiating Non-negative Matrix Factorization based on manifold | |
| Lai et al. | Improving classification with semi-supervised and fine-grained learning | |
| CN105334504A (en) | Radar target identification method based on large-boundary nonlinear discrimination projection model |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20180907 | |
| RJ01 | Rejection of invention patent application after publication |