摘要：目的:为了解决传统帧差法和动态特征分组法的局限性.方法:本文提出

摘要：目的:为了解决传统帧差法和动态特征分组法的局限性.方法:本文提出一种将帧差法和动态特征分组法相结合的实时目标检测跟踪算法. 首先提出了一种背景差值算法的扩充和有一种特征值跟踪分组算法, 随后引入了一种多级动态特征分组算法, 该算法能够适用于实时应用，处理各种大小的目标和提供稳定的轨迹.随后提出了一种能够适用于实时应用和从不完整的特征跟踪中产生高质量目标轨迹的动态特征分组. 通过用特征跟踪结果作为额外的线索给出稳定性更好的背景差值结果；同时，通过用背景差值作为线索给出更好的特征检测和分组结果.结果:利用VS2010进行编程,在CPU2.5G计算机上分析了了一份由大约80秒的包含车辆、自行车和行人的十字路口视频剪辑的结果,取得了非常好的效果.对比Ncut分组算法,本文算法执行时间45ms优于Ncut分组算法77ms.结论:本文方法对航空相机遥感、星载相机对地观测等领域具有很好的战略意义，可分析对地面坦克、机场等目标的实时检测跟踪。

結果 (英文) 1:

Abstract: objective: in order to address traditional limitations of dynamic characteristics of frame-difference method and grouping. methods: this paper presents a dynamic characteristics of frame-difference method and group real-time object detection and tracking algorithm combining. Background subtraction algorithm first proposed the expansion and eigenvalue of a trace grouping algorithm, then introduces a dynamic characteristics of multi-level grouping algorithm, this algorithm can be applied to real-time applications, dealing with the objectives of various sizes and provide a stable trajectory. subsequently offered a suitable for real-time applications and is never complete, produces high quality feature tracking of dynamic characteristics of target trajectory group. Through with features track results as extra of clues to out stability better of background poor value results; while, through with background poor value as clues to out better of features detection and group results. results: uses VS2010 for programming, in CPU2.5G computer Shang analysis Gets a copies by about 80 seconds of contains vehicles, and bike and pedestrian of crossroads video clip of results, made has very good of effect. contrast Ncut group algorithm, 45ms superior Ncut algorithm execution time grouping algorithm 77ms. conclusions: this method for aerial cameras, on-board camera for remote sensing Earth observation areas of great strategic importance, analysis of the tank on the ground,Real-time detection and tracking of targets such as the airport.

結果 (英文) 2:

Abstract: Objective: To address the limitations of conventional frame difference method and dynamic characteristics of grouping methods: This paper presents a real-time target detection and tracking algorithms will frame difference and dynamic characteristics of the Combination group first proposed a background difference expansion algorithm and there is a tracking eigenvalues ​​grouping algorithm, then the introduction of a multi-level dynamic feature grouping algorithm, which can be applied to real-time applications to handle various sizes of targets and provide a stable trajectory. then proposed a way to generate dynamic characteristics suitable for real-time applications and never complete characterization of the target trajectory tracking of high-quality packet given the difference between stability and better results by using a background feature tracking results as an additional clue; same time, by using background difference give better clues as feature detection and grouping Results: Using VS2010 programming on CPU2.5G a computer analysis of about 80 seconds from the included vehicles, bicycles and pedestrians crossroads video clip results achieved very good results contrast Ncut grouping algorithm, this algorithm is better than the execution time of 45ms 77ms grouping algorithm Ncut conclusion: this method of aerial camera remote sensing, earth observation spaceborne cameras and other areas with good strategic sense, can be analyzed ground tanks, airports and other real-time target detection and tracking.

結果 (英文) 3:

Abstract: Objective: in order to solve the limitation of the traditional frame difference method and dynamic grouping method. Methods: real time object detection and tracking algorithm is proposed in this paper to a frame difference method and dynamic feature grouping method combining. First proposed an extended background difference algorithm and a feature tracking algorithm, then the introduction of a multistage dynamic feature grouping algorithm, the algorithm can be applied to real-time application, processing of various sizes of goals and provide stable trajectory. Then put forward a kind of dynamic characteristics can be grouped in real-time application and never complete feature tracking to produce high quality target trajectoryThrough the use of feature tracking results as the background subtraction results better clues given additional stability; at the same time, by using the background difference as the feature cue gives better detection and grouping results. Results: the use of VS2010 programming, in CPU2.5G computer analysis of a vehicle, comprising about 80 seconds of bicycle and pedestrian crossroads video clip of the results, obtained very good effect. Compared with Ncut algorithm, the algorithm execution time 45ms is superior to Ncut algorithm 77ms. conclusion: this method of aerial camera, remote sensing satellite camera is of strategic significance for good field observation, analysis of ground tank,Real time detection and tracking the target airport etc..