IOE OpenIR  > 光电技术研究所博硕士论文
Alternative TitleFusion of Multiple Algorithms Based on Object Tracking
Thesis Advisor张启衡
Degree Grantor中国科学院光电技术研究所
Place of Conferral光电技术研究所
Degree Discipline光学工程
Keyword复杂场景 图像分割 目标跟踪 算法融合 并行处理
Abstract在目标跟踪过程中,背景复杂多变;目标发生变形﹑旋转﹑尺度变化及灰度变化;目标被遮挡等因素,都会不同程度的影响目标跟踪的稳定性。由于任何跟踪算法对复杂场景下目标跟踪的适应性都是有限,采用单一算法对复杂场景下的目标进行跟踪,容易导致跟踪失败。因此,充分发挥各种跟踪算法的优点,采用多个算法融合,才能有效的提高跟踪系统适应复杂场景的能力。 针对复杂场景下的目标稳定跟踪,深入研究了复杂场景下目标跟踪算法﹑多算法融合方法﹑多DSP并行处理平台﹑多算法融合实现等四个主要内容,取得了阶段性结果。主要研究内容和结果如下: (1)分析了复杂场景下影响目标跟踪稳定性的因素,了解了复杂场景下目标跟踪基础理论,阐述了提高系统跟踪稳定性的途径。 (2)针对变形目标的分割及跟踪,分析了基于传统水平集方法和M-S分割模型的局限性,提出了改进的分割模型,实现对边缘模糊的目标的分割;分别采用改进的分割模型和Snake模型实现对变形目标的跟踪。 (3)采用背景直方图加权的方法,改进了传统的均值平移算法,使得该算法对遮挡及目标尺度变化具有良好的适应性。 (4)针对多算法融合中建立决策判据的难题,采用贝叶斯网络建立多算法融合模型,实现多算法融合跟踪的智能决策。 (5)根据多算法融合实时跟踪需求,为满足目标跟踪系统中多通道数据处理、多传感器融合技术﹑复杂算法工程化的要求,研制了基于共享总线结构﹑CPCI架构的多DSP的并行处理系统。 (6)在分析各种目标跟踪算法的适应性及可用资源的基础上,优选多种跟踪算法进行融合。合理划分FPGA和DSP的功能,在并行处理平台上实施多算法融合,实现了对复杂场景下目标的稳定跟踪。 论文的最后,总结了本文的主要工作,并指出了存在的问题及进一步研究的方向。
Other AbstractThe algorithm for tracking in image sequence suffers from problems from sources such as: the complex background, object deformity, rotation of object, change in object scale, and illumination variation, occlusion. The difficulty with individual tracking algorithm is that it can not adapt to changes in conditions in the image thoroughly and the algorithm will fail. Given the wealth of tracking algorithms already available, a more promising solution seems to be to combine the results of multiple tracking systems-a type of algorithmic fusion. The main studying comprised of tracking algorithms in complex scene, fusion of multiple algorithms, parallel processing platform by using of multiple DSPs, and implement fusion of multi-algorithm. Having achieved some validation results, main result as following: (1) It is analyzed that influenced objects tracking in complex scene, knowing about basic theory of object tracking. It is expatiated that the robust tracking system can be achieved with fusion of multiple algorithms. (2) The limitation of traditional level set method and M-S model deformable is analyzed for deformable object segmentation, a new improved algorithm was presented for object without edge. Then adopting improved model and Snake model to segment and track deformable object. (3) Mean-shift algorithm is improved by use of background-weighted histogram, then using improved mean-shift to tracking object in gray image, results show that it has better adaptability to occlusion and object scale changing. (4) The major problem with fusion of multiple algorithm is how to decide which result is correct, we propose a decision framework for fusion of multiple algorithms based on Bayesian Networks. (5) For requirement of multiple tracking algorithm fusion, in order to meet the need of multi-channel image processing, multi-sensor fusion and complex algorithm implement, that parallel processing system with multiple DSPs based on shared bus and CPCI frame is developed. (6) Based on analyzing the adaptability of multiple tracking algorithm and usable algorithms, assigning reasonable function between DSP and FPGA, the real time tracking on moving object in complex scene is realized with fusion of several algorithm chosen . At last, the main work of this paper is summarized. Also, the inadequacies and further research directions are proposed.
Document Type学位论文
Recommended Citation
GB/T 7714
祁小平. 基于成像跟踪的多算法融合技术研究[D]. 光电技术研究所. 中国科学院光电技术研究所,2007.
Files in This Item:
File Name/Size DocType Version Access License
10001_20021801510335(18364KB) 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[祁小平]'s Articles
Baidu academic
Similar articles in Baidu academic
[祁小平]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[祁小平]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.