CN109118517B - Multi-scale tracking method based on geometric estimation - Google Patents
Multi-scale tracking method based on geometric estimation Download PDFInfo
- Publication number
- CN109118517B CN109118517B CN201810775837.1A CN201810775837A CN109118517B CN 109118517 B CN109118517 B CN 109118517B CN 201810775837 A CN201810775837 A CN 201810775837A CN 109118517 B CN109118517 B CN 109118517B
- Authority
- CN
- China
- Prior art keywords
- target
- scale
- tracking
- template
- current
- 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
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention discloses a multi-scale tracking method based on geometric estimation, relates to the field of computer vision and image processing, can perform multi-target tracking, and has high calculation speed and good real-time performance. The invention comprises the following steps: inputting an image sequence, an initial position and an initial size of a tracking target, and setting a scale and a tracking frame; performing feature extraction and template training on the target in the tracking frame, and selecting a target position and a target scale variable; estimating the actual distance between the target and the camera according to the camera imaging principle, estimating the change of the target scale by using the actual distance, and reducing the value range of the target scale variable; feeding the scale variable back to the first step, circulating the steps, and finally training to obtain a filter capable of judging the target position and the scale; and tracking the scale change target through a filter. The invention reduces the calculation amount, ensures the frame rate of the algorithm, improves the real-time performance of the algorithm and realizes the accurate and quick target tracking effect.
Description
Technical Field
The invention relates to the field of computer vision and image processing, in particular to a multi-scale tracking method based on geometric estimation.
Background
With the development of artificial intelligence, computer vision is rapidly developed as an important part of the development, wherein target vision tracking is widely applied in the fields of intelligent transportation, human-computer interaction, intelligent monitoring and the like, and various tracking algorithms are also endlessly developed, so that the tracking algorithms become a hot problem for research of experts and scholars at home and abroad. The scale change problem is one of the great problems in the field of target tracking, and the tracking effect of many tracking algorithms is poor when the target scale changes. Therefore, the multi-scale tracking algorithm is developed, a plurality of target scales can be established by the multi-scale tracking algorithm to adapt to the change of the target scales, and the self-adaptation of the target scales is realized.
However, in the current multi-scale tracking algorithm, due to the fact that the establishment of a plurality of target scales causes the situations of large calculation amount and poor real-time performance, and cannot meet the requirements in practical engineering application, a multi-target scale tracking method with high calculation speed and good real-time performance is lacked in the prior art.
Disclosure of Invention
The invention provides a multi-scale tracking method based on geometric estimation, which sets a plurality of target scales, realizes the tracking of a target with scale change, and has high calculation speed and good real-time performance, thereby accurately and quickly realizing the target tracking.
In order to achieve the purpose, the invention adopts the following technical scheme:
a multi-scale tracking method based on geometric estimation, comprising:
s1, inputting an image sequence, tracking target initial position and tracking target initial size X, and setting a target scale variable Si＝{0.8，0.9，1.0，1.1，1.2}(i＝1～5)；
S2, according to the initial size X of the tracked target and the scale variable SiSetting 5 tracking frames, performing feature extraction and template training on a tracking target in the 5 tracking frames to obtain 5 candidate target templates, performing Gaussian kernel correlation operation on the candidate target templates and a target sample respectively to obtain maximum response, marking the candidate target template with the highest response value as a current target template, marking the position of the maximum response value as the current target position, and marking a scale variable corresponding to the candidate target template with the highest response value as a current target scale variable;
s3, estimating the actual distance d between the target and the camera according to the height h of the tracked target in the image sequence according to the camera imaging principle;
s4, estimating the change of the target scale by using the actual distance d, and reducing the target scale variable SiThe value range of (a) is set as the target scale variable siMultiplying the current tracking target size by the initial tracking target size X to obtain the current tracking target size, marking the current tracking target size as the initial tracking target size of the next frame in the input image sequence, marking the current target position as the initial tracking target position of the next frame in the input image sequence, and executing S1-S3 in a circulating manner, wherein a filter of the target position and scale can be obtained through training and iteration, and the filter can judge the position and scale of the target in each frame;
and S5, processing the input image sequence by using a filter to obtain the position and the scale of the tracking target, and realizing the tracking of the scale change target.
Further, S2 includes:
s21, according to the initial size X of the tracked target and the scale variable SiSetting 5 tracking frames at the target dimension si(i is 1-5) carrying out target feature extraction and template training in the corresponding tracking frame to obtain 5 candidate target templates with different scalesi(i＝1～5)；
S22, template of 5 candidate targetsi(i is 1-5), respectively performing correlation operation with the target sample of the next frame in the input image sequence,
yi＝f(templatei，sample)
f is the Gaussian kernel correlation function, yi(i is 1-5) comparing the 5 response results to obtain the maximum response value ymax；
S23, calculating the maximum response value SiAs the position of the target, and the candidate target template used in the responseiAs the current target template, a candidate target templateiThe scale of (2) is taken as the current target scale, so as to obtain the current target position and the current target scale variable.
Further, S3 includes:
s31, template according to the targetiObtaining the height h of the tracking target in the video sequence;
s32, substituting the height h of the tracking target into a camera imaging principle formula
Wherein a is the actual height of the target, f is the focal length of the camera, d is the actual distance, and a and f are fixed values, and the actual distance d is estimated.
Further, in S4, the target size variable S is reducediThe value ranges of (A) include:
judging the actual distance between the target and the camera according to the numerical change of the actual distance d, thereby judging the change of the target scale, and if the value of d is reduced, carrying out the next frame in the input image sequence on the scale variable siIs set as s1，s2，s3Obtaining a template as template1，template2，template3Finally in response to result y1，y2.y3Calculating the maximum value;
if the value of d becomes large, the scale variable s is changed in the next frame in the input image sequenceiIs set as s3，s4，s5Obtaining a template as template3，template4，template5Finally in response to result y3，y4，y5The maximum value is obtained.
The invention has the beneficial effects that:
the invention sets a tracking algorithm of a plurality of target scales, adopts a geometric estimation method, greatly reduces the calculated amount of the tracking algorithm of the plurality of target scales by estimating the sizes of the targets and combining the corresponding geometric relational expression, ensures the frame rate of the algorithm, improves the real-time property of the algorithm and realizes the accurate and quick target tracking effect.
Detailed Description
In order that those skilled in the art will better understand the technical solutions of the present invention, the present invention will be further described in detail with reference to the following detailed description.
The embodiment of the invention provides a multi-scale tracking method based on geometric estimation, which comprises the following steps:
a multi-scale tracking method based on geometric estimation, comprising:
s1, inputting the image sequence, the initial position of the tracking target and the initial size X (length and width) of the tracking target, and setting a target scale variable Si＝{0.8，0.9，1.0，1.1，1.2}，i＝1～5。
S2, according to the initial size X of the tracked target and the scale variable SiMultiplying to obtain 5 target tracking frames in a target scale si(i is 1-5) carrying out target feature extraction and template training in the corresponding tracking frame to obtain 5 candidate target templates with different scalesi(i＝1～5)；
Template of 5 candidate objectsi(i is 1-5), respectively performing correlation operation with the target sample of the next frame,
yi＝f(templatei，sample)
f is the Gaussian kernel correlation function, yi(i is 1-5) comparing the 5 response results to obtain the maximum response value ymax；
Will respond maximallyValue ymaxAs the position of the target, and the candidate target template used in the responseiAs the current target template, a candidate target templateiThe scale of (2) is taken as the current target scale, so as to obtain the current target position and the current target scale variable.
S3, template is tempted according to camera imaging principleiObtaining the height h of the tracking target in the video sequence;
substituting the height h of the tracked target into a camera imaging principle formula
Wherein a is the actual height of the target, f is the focal length of the camera, d is the actual distance, and a and f are fixed values, and the actual distance d is estimated.
S4, estimating the change of the target scale by using the actual distance d, and reducing the target scale variable SiThe value range of (a).
Wherein the target size variable s is reducediThe value ranges of (A) include:
judging the actual distance between the target and the camera according to the numerical change of the actual distance d, thereby judging the change of the target scale, and if the value of d is reduced, changing the scale variable s in the next frameiIs set as s1，s2.s3Obtaining a template as template1，template2，template3Finally in response to result y1，y2，y3Calculating the maximum value;
if the value of d becomes large, the scale variable s is set in the next frameiIs set as s3，s4，s5Obtaining a template as template3，template4，template5Finally in response to result y3，y4，y5The maximum value is obtained.
The target scale variable siMultiplying the current tracking target size by the tracking target initial size X to obtain the current tracking target size, marking the current tracking target size as the tracking target initial size in the next frame, and marking the current tracking target size as the tracking target initial size in the next frameThe current target position is marked as the initial position of the tracking target in the next frame, and the loop is executed from S1 to S3, and through training and iteration, a filter of the target position and scale can be obtained, and the filter can distinguish the position and scale of the target in each frame.
And S5, processing the input image sequence by using a filter to obtain the position and the scale of the tracking target, and realizing the tracking of the scale change target.
The embodiment is applied to a kernel correlation filtering tracking algorithm without scale change, and a computer running the improved algorithm is configured as follows: windows 8.1 operating system, processor Intel (R) core (TM) i5-4200M (2.50GHz), 8 GB. The operating environment is a Matlab platform.
The parameters are set as follows: learning factor β is 0.02, gaussian kernel σ is 0.5, regularization parameter λ is 1E-4, cell 4 × 4, HOG features in 9 direction. A video sequence with target scale change (video sequence name: Heman8) is selected from the data set, and the video sequence is tested by adopting the embodiment, and the test result is shown in the following table:
table: improved algorithm tracks changes in frame dimensions across different frames
According to the table, the tracking frame of the improved algorithm can change along with the change of the target dimension, and the fps of the algorithm is 100.5966, so that the calculation amount of the algorithm is reduced by adopting a geometric estimation method.
The invention has the beneficial effects that:
the invention sets a tracking algorithm of a plurality of target scales, adopts a geometric estimation method, greatly reduces the calculated amount of the tracking algorithm of the plurality of target scales by estimating the sizes of the targets and combining the corresponding geometric relational expression, ensures the frame rate of the algorithm, improves the real-time property of the algorithm and realizes the accurate and quick target tracking effect.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (4)
1. A multi-scale tracking method based on geometric estimation is characterized by comprising the following steps:
s1, inputting an image sequence, tracking target initial position and tracking target initial size X, and setting a target scale variable Si＝{0.8，0.9，1.0，1.1，1.2}(i＝1～5)，
S2, according to the initial size X of the tracked target and the scale variable SiSetting 5 tracking frames, performing feature extraction and template training on a tracking target in the 5 tracking frames to obtain 5 candidate target templates, performing Gaussian kernel correlation operation on the candidate target templates and a target sample respectively to obtain maximum response, marking the candidate target template with the highest response value as a current target template, marking the position of the maximum response value as the current target position, and marking a scale variable corresponding to the candidate target template with the highest response value as a current target scale variable;
s3, estimating the actual distance d between the target and the camera according to the height h of the tracked target in the image sequence according to the camera imaging principle;
s4, estimating the change of the target scale by using the actual distance d, and reducing the target scale variable SiThe value range of (a) is set as the target scale variable siMultiplying the current tracking target size by the initial tracking target size X to obtain the current tracking target size, marking the current tracking target size as the initial tracking target size of the next frame in the input image sequence, marking the current target position as the initial tracking target position of the next frame in the input image sequence, and executing S1-S3 in a circulating manner, wherein a filter of the target position and scale can be obtained through training and iteration, and the filter can judge the position and scale of the target in each frame;
and S5, processing the input image sequence by using the filter to obtain the position and the scale of the tracking target, and realizing the tracking of the scale change target.
2. The method according to claim 1, wherein the S2 includes:
s21, according to the initial size X of the tracked target and the scale variable SiSetting 5 tracking frames at a target scale siAnd performing target feature extraction and template training in the tracking frame corresponding to 1-5 to obtain 5 candidate target templates with different scalesi，i＝1～5；
S22, 5 candidate target templates are templatediI is 1-5, which is respectively related to the next frame target sample of the input image sequence,
yi＝f(templatei，sample)
f is the Gaussian kernel correlation function, yiAnd i is 1-5 response results, and the 5 response results are compared to obtain the maximum response value ymax；
S23, calculating the maximum response value ymaxAs the location of the target, the candidate target template used by the response is usediAs the current target template, the candidate target templateiAs the current target scale, thereby obtaining the current target position and the current target scale variable.
3. The geometric-estimation-based multi-scale tracking method according to claim 2, wherein the S3 includes:
s31, template is according to the goaliObtaining the height h of the tracking target in the image sequence;
s32, substituting the height h of the tracking target into a camera imaging principle formula
Where a is the actual height of the target, f is the camera focal length, and d isAnd the actual distance a and f are fixed values, and the actual distance d is estimated.
4. The multi-scale tracking method based on geometric estimation according to claim 1, wherein in the step S4, the reduced target size variable SiThe value ranges of (A) include:
judging the actual distance between the target and the camera according to the numerical change of the actual distance d, thereby judging the change of the target scale, and if the value of d becomes smaller, in the next frame in the input image sequence, changing the scale variable siIs set as s1，s2，s3Obtaining a template as template1，template2，template3Finally in response to result y1，y2，y3Calculating the maximum value;
if the value of d is larger, the scale variable s is used for the next frame in the input image sequenceiIs set as s3，s4，s5Obtaining a template as template3，template4，template5Finally in response to result y3，y4，y5The maximum value is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810775837.1A CN109118517B (en) | 2018-07-13 | 2018-07-13 | Multi-scale tracking method based on geometric estimation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810775837.1A CN109118517B (en) | 2018-07-13 | 2018-07-13 | Multi-scale tracking method based on geometric estimation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109118517A CN109118517A (en) | 2019-01-01 |
| CN109118517B true CN109118517B (en) | 2021-02-05 |
Family
ID=64862629
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810775837.1A Active CN109118517B (en) | 2018-07-13 | 2018-07-13 | Multi-scale tracking method based on geometric estimation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109118517B (en) |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9898829B2 (en) * | 2012-09-18 | 2018-02-20 | Hanwha Techwin Co., Ltd. | Monitoring apparatus and system using 3D information of images and monitoring method using the same |
| CN105184822B (en) * | 2015-09-29 | 2017-12-29 | 中国兵器工业计算机应用技术研究所 | A kind of target following template renewal method |
| CN106023257B (en) * | 2016-05-26 | 2018-10-12 | 南京航空航天大学 | A kind of method for tracking target based on rotor wing unmanned aerial vehicle platform |
| CN107016689A (en) * | 2017-02-04 | 2017-08-04 | 中国人民解放军理工大学 | A kind of correlation filtering of dimension self-adaption liquidates method for tracking target |
-
2018
- 2018-07-13 CN CN201810775837.1A patent/CN109118517B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109118517A (en) | 2019-01-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135503B (en) | Deep learning identification method for parts of assembly robot | |
| CN107369166B (en) | Target tracking method and system based on multi-resolution neural network | |
| CN107403175A (en) | Visual tracking method and Visual Tracking System under a kind of movement background | |
| CN110569723A (en) | Target tracking method combining feature fusion and model updating | |
| CN111582349B (en) | Improved target tracking algorithm based on YOLOv3 and kernel correlation filtering | |
| CN108846404B (en) | Image significance detection method and device based on related constraint graph sorting | |
| CN107067410B (en) | Manifold regularization related filtering target tracking method based on augmented samples | |
| CN106373145B (en) | Multi-object tracking method based on tracking segment confidence level and the study of distinction appearance | |
| CN106981071A (en) | A kind of method for tracking target applied based on unmanned boat | |
| CN105976397B (en) | A kind of method for tracking target | |
| CN107368802B (en) | Moving target tracking method based on KCF and human brain memory mechanism | |
| CN105046714A (en) | Unsupervised image segmentation method based on super pixels and target discovering mechanism | |
| US10115208B2 (en) | Image characteristic estimation method and device | |
| CN115375737A (en) | Target tracking method and system based on adaptive time and serialized space-time characteristics | |
| Zheng et al. | MD-YOLO: Surface Defect Detector for Industrial Complex Environments | |
| CN109118517B (en) | Multi-scale tracking method based on geometric estimation | |
| CN110570450B (en) | Target tracking method based on cascade context-aware framework | |
| Mbelwa et al. | Objectness-based smoothing stochastic sampling and coherence approximate nearest neighbor for visual tracking | |
| CN112633078B (en) | Target tracking self-correction method, system, medium, equipment, terminal and application | |
| Zou et al. | Bird detection on transmission lines based on DC-YOLO model | |
| CN108985216A (en) | A kind of pedestrian head detection method based on multiple logistic regression Fusion Features | |
| Dai et al. | OAMatcher: An overlapping areas-based network with label credibility for robust and accurate feature matching | |
| Xie et al. | Multi-scale patch-based sparse appearance model for robust object tracking | |
| Deng et al. | Few-Shot Steel Plate Surface Defect Detection with Multi-Relation Aggregation and Adaptive Support Learning | |
| Zhang et al. | Adaptively learning background-aware correlation filter for visual tracking |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230329 Address after: Room 201, Room 101, Building 1, No. 18, Daoyuan Road, High-tech Zone, Suzhou City, Jiangsu Province, 215000 Patentee after: SUZHOU ZHONGKETIANQI REMOTE SENSING TECHNOLOGY CO.,LTD. Address before: No. 29, Qinhuai District, Qinhuai District, Nanjing, Jiangsu Patentee before: Nanjing University of Aeronautics and Astronautics |
|
| TR01 | Transfer of patent right |