CN103679755A - Single-goal long-time tracking technique - Google Patents
Single-goal long-time tracking technique Download PDFInfo
- Publication number
- CN103679755A CN103679755A CN201310713690.0A CN201310713690A CN103679755A CN 103679755 A CN103679755 A CN 103679755A CN 201310713690 A CN201310713690 A CN 201310713690A CN 103679755 A CN103679755 A CN 103679755A
- Authority
- CN
- China
- Prior art keywords
- module
- target
- tracking
- time tracking
- motion
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to a single-goal long-time tracking technique. The single-goal long-time tracking technique comprises the following steps that: a tracking module estimates motion of a target by supposing that motion of an object between adjacent video frames is limited and a tracked target is visible; a detecting module supposes that each video frame is independent and performs whole-graph search for each frame of picture to position a region where the target may appear based on target models detected and learned in the past; like the other target detection methods, the detecting module in TLD (Track, Learn and Detection) may go wrong, and the error is nothing but a negative sample of the error and a positive sample of the error; a learning module assesses the two errors of the detecting module based on the result of the tracking module, generates a training sample based on the assessment result, updates the target model of the detecting module and updates key characteristics of the tracking module so as to prevent similar errors from appearing in future. The single-goal long-time tracking technique has more steady tracking effect, better robustness and better reliability.
Description
Technical field
The invention belongs to tracking field, particularly a kind of long-time tracking technique of single goal.
Background technology
For long-time tracking, a crucial problem is: when target reappears in the camera visual field, system should be able to detect it again, and starts again to follow the tracks of.But, the situation such as in long-time tracing process, change of shape, illumination condition variation, dimensional variation will inevitably occur tracked target, block.Traditional track algorithm, front end need to cooperatively interact with detection module, after tracked target being detected, just starts to enter tracking module, and after this, detection module just can not got involved in tracing process.But this method has a fatal defect:, when tracked target exists change of shape or blocks, follow the tracks of and be just easy to failure; Therefore, for long-time tracking, or there is the tracking in change of shape situation in tracked target, and a lot of people adopt the method for detection to replace following the tracks of.Although the method can be improved tracking effect in some cases, it needs the learning process of an off-line.That is: before detecting, need to select the sample of a large amount of tracked targets and learn and train.This also just means, training sample will contain the contingent various deformation of tracked target and various yardstick, attitude change and the situation of illumination variation.In other words, the method that utilization detects reaches the object of long-time tracking, most important for the selection of training sample, otherwise the robustness of tracking is just difficult to guarantee.
Summary of the invention
In order to overcome the shortcoming of prior art, the invention provides the long-time tracking technique of a kind of single goal.Its tracking effect is more stable, robust, reliable.
The present invention solves the technical scheme that its technical matters takes: the motion that comprises the following steps: object between tracking module hypothesis adjacent video frames is limited, and tracked target is visible, with this, carry out the motion of estimating target, if target disappears in the camera visual field, to cause and follow the tracks of unsuccessfully, detection module supposes that each is independent of each other depending on frame, and according to the object module that detected and learnt in the past, each frame picture is carried out to the region that full figure search may occur with localizing objects, the same with other object detection method, also likely there is mistake in the detection module in TLD, and mistake is nothing but wrong negative sample and wrong positive sample both of these case, study module is assessed these two kinds of mistakes of detection module according to the result of tracking module, and generate training sample according to assessment result, object module to detection module upgrades, " key feature points " of tracking module upgraded simultaneously, after avoiding with this, there is similar mistake.
Tracking effect of the present invention is more stable, robust, reliable.
Embodiment
Consider that simple tracking or simple detection algorithm all cannot reach desirable effect in long-time tracing process, so, (TLD) method of following the tracks of-study-detect just considers both to give combination, and add a kind of improved on-line study mechanism, thereby make whole target following more stable, effective.
The present invention includes the following step: between tracking module hypothesis adjacent video frames, the motion of object is limited, and tracked target is visible, with this, carry out the motion of estimating target, if target disappears in the camera visual field, to cause and follow the tracks of unsuccessfully, detection module supposes that each is independent of each other depending on frame, and according to the object module that detected and learnt in the past, each frame picture is carried out to the region that full figure search may occur with localizing objects, the same with other object detection method, also likely there is mistake in the detection module in TLD, and mistake is nothing but wrong negative sample and wrong positive sample both of these case, study module is assessed these two kinds of mistakes of detection module according to the result of tracking module, and generate training sample according to assessment result, object module to detection module upgrades, " key feature points " of tracking module upgraded simultaneously, after avoiding with this, there is similar mistake.
Claims (1)
1. the long-time tracking technique of single goal, it is characterized in that: the motion that comprises the following steps: object between tracking module hypothesis adjacent video frames is limited, and tracked target is visible, with this, carry out the motion of estimating target, if target disappears in the camera visual field, to cause and follow the tracks of unsuccessfully, detection module supposes that each is independent of each other depending on frame, and according to the object module that detected and learnt in the past, each frame picture is carried out to the region that full figure search may occur with localizing objects, the same with other object detection method, also likely there is mistake in the detection module in TLD, and mistake is nothing but wrong negative sample and wrong positive sample both of these case, study module is assessed these two kinds of mistakes of detection module according to the result of tracking module, and generate training sample according to assessment result, object module to detection module upgrades, " key feature points " of tracking module upgraded simultaneously, after avoiding with this, there is similar mistake.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310713690.0A CN103679755A (en) | 2013-12-20 | 2013-12-20 | Single-goal long-time tracking technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310713690.0A CN103679755A (en) | 2013-12-20 | 2013-12-20 | Single-goal long-time tracking technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103679755A true CN103679755A (en) | 2014-03-26 |
Family
ID=50317205
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310713690.0A Pending CN103679755A (en) | 2013-12-20 | 2013-12-20 | Single-goal long-time tracking technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103679755A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331901A (en) * | 2014-11-26 | 2015-02-04 | 北京邮电大学 | TLD-based multi-view target tracking device and method |
| CN104517125A (en) * | 2014-12-26 | 2015-04-15 | 湖南天冠电子信息技术有限公司 | Real-time image tracking method and system for high-speed article |
| CN107705319A (en) * | 2017-10-16 | 2018-02-16 | 中国电子科技集团公司第二十八研究所 | One kind is based on the real-time detecting and tracking method of empty day background Small object |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2008176504A (en) * | 2007-01-17 | 2008-07-31 | Toshiba Corp | Object detector and method therefor |
| US20120237081A1 (en) * | 2011-03-16 | 2012-09-20 | International Business Machines Corporation | Anomalous pattern discovery |
| CN102722725A (en) * | 2012-06-04 | 2012-10-10 | 西南交通大学 | Object tracing method based on active scene learning |
| CN102881024A (en) * | 2012-08-24 | 2013-01-16 | 南京航空航天大学 | Tracking-learning-detection (TLD)-based video object tracking method |
-
2013
- 2013-12-20 CN CN201310713690.0A patent/CN103679755A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2008176504A (en) * | 2007-01-17 | 2008-07-31 | Toshiba Corp | Object detector and method therefor |
| US20120237081A1 (en) * | 2011-03-16 | 2012-09-20 | International Business Machines Corporation | Anomalous pattern discovery |
| CN102722725A (en) * | 2012-06-04 | 2012-10-10 | 西南交通大学 | Object tracing method based on active scene learning |
| CN102881024A (en) * | 2012-08-24 | 2013-01-16 | 南京航空航天大学 | Tracking-learning-detection (TLD)-based video object tracking method |
Non-Patent Citations (2)
| Title |
|---|
| 周鑫 等: "改进后的TLD视频目标跟踪方法", 《中国图象图形学报》, vol. 18, no. 9, 30 September 2013 (2013-09-30) * |
| 高帆 等: "TLD目标跟踪算法研究", 《电视技术》, vol. 37, no. 11, 11 July 2013 (2013-07-11) * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331901A (en) * | 2014-11-26 | 2015-02-04 | 北京邮电大学 | TLD-based multi-view target tracking device and method |
| CN104517125A (en) * | 2014-12-26 | 2015-04-15 | 湖南天冠电子信息技术有限公司 | Real-time image tracking method and system for high-speed article |
| CN104517125B (en) * | 2014-12-26 | 2018-05-22 | 湖南天冠电子信息技术有限公司 | The image method for real time tracking and system of high-speed object |
| CN107705319A (en) * | 2017-10-16 | 2018-02-16 | 中国电子科技集团公司第二十八研究所 | One kind is based on the real-time detecting and tracking method of empty day background Small object |
| CN107705319B (en) * | 2017-10-16 | 2020-04-17 | 中国电子科技集团公司第二十八研究所 | Aerospace background-based small target real-time detection tracking method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10839547B2 (en) | Camera pose determination and tracking | |
| CN105405154A (en) | Target object tracking method based on color-structure characteristics | |
| CN109859266B (en) | Pre-transformation-based visual simultaneous positioning and drawing method under large visual angle change | |
| Zhou et al. | Robust plane-based structure from motion | |
| CN107862240B (en) | Multi-camera collaborative face tracking method | |
| Le Gentil et al. | Idol: A framework for imu-dvs odometry using lines | |
| CN101344965A (en) | Tracking system based on binocular camera shooting | |
| CN104408725A (en) | Target recapture system and method based on TLD optimization algorithm | |
| US20220156973A1 (en) | Information processing apparatus, information processing method, and program | |
| CN103679750B (en) | A kind of DE Camera Shake detection method based on video | |
| CN103679755A (en) | Single-goal long-time tracking technique | |
| CN109360225B (en) | Motion model optimization system and method | |
| CN103324932A (en) | Video-based vehicle detecting and tracking method and system | |
| CN103714553A (en) | Multi-target tracking method and apparatus | |
| CN111161318A (en) | Dynamic scene SLAM method based on YOLO algorithm and GMS feature matching | |
| Li et al. | Fast vision‐based autonomous detection of moving cooperative target for unmanned aerial vehicle landing | |
| CN103646242B (en) | Extended target tracking based on maximum stable extremal region feature | |
| Küçükyildiz et al. | Development and optimization of a DSP-based real-time lane detection algorithm on a mobile platform | |
| CN101950425A (en) | Motion behavior detection-based intelligent tracking arithmetic | |
| WO2014112407A1 (en) | Information processing system, information processing method, and program | |
| CN102456127A (en) | Head posture estimation equipment and head posture estimation method | |
| CN111260725A (en) | Dynamic environment-oriented wheel speed meter-assisted visual odometer method | |
| CN110826448B (en) | Indoor positioning method with automatic updating function | |
| CN116630764B (en) | Water surface target fusion identification method, system, storage medium, computer equipment and terminal | |
| Fu et al. | The Capture of Moving Object in Video Image. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20140326 |
|
| RJ01 | Rejection of invention patent application after publication |