| 摘要: |
| 复杂背景条件下低对比度目标的跟踪和测量方法,是视觉领域的一个重要课题.低对比度,低信噪比,目标旋转、缩放、被遮挡等非理想状态给跟踪算法的研究带来很大困难,算法既要适应目标和背景的复杂变化,又要保证运算量小,满足工程实时性要求.提出一种基于似然相似度函数的低对比度目标跟踪方法.在建立模型阶段,利用棱锥面方程的单峰特性突出模型中的目标灰度信息,使目标与背景灰度信息的可区分性更高;在模型匹配阶段,从统计学中的极大似然估计方法得到启发,构造一种新的似然相似度函数,与传统的相似度量相比,度量值的可区分性更高,大大提高了匹配区域的无重复模式;最后,将目标跟踪过程转化为对目标跟踪位置的极大似然估计过程.目前,该算法已经成功嵌入TMS320C6416硬件平台.大量实验结果表明,该算法所能探测的目标对比度LSCR最低限度约为3.作为实例,给出复杂背景下低对比度LSCR=4.9时空中飞机的实验结果. |
| 关键词: 似然相似度函数 均值漂移 实时跟踪 Bhattacharyya系数 |
| DOI：10.13328/j.cnki.jos.004619 |
| 分类号: |
| 基金项目:国家自然科学基金(61172111) |
|
| Research on the Robust Illumination of the Likelihood Similarity Function in Tracking Target |
|
DI Nan1,2, ZHU Ming1,2, HAN Guang-Liang2
|
|
1.Key Laboratory of Airborne Optical Imaging and Measurement (Changchun Institute of Optics, Fine Mechanics and Physics, The Chinese Academy of Sciences), Changchun 130033, China;2.Department of Fast Capture and Real-Time Image Processing, Changchun Institute of Optics, Fine Mechanics and Physics, The Chinese Academy of Sciences, Changchun 130033, China
|
| Abstract: |
| Real-Time tracking low-contrast target in the complex environment is a key problem in the visual area. The algorithm needs to not only copy with the high similar between target and background, revolution, scale variations and target occlusions but also satisfy the real-time tracking. This paper provides a method based on the likelihood similarity function to resolve the low-contrast tracking. In the model construction phase, the new method uses the single peak of the pyramid surface equation to enhance the target information. In the model matching phase, it innovates a new likelihood similarity function which provides more distinguishable measurements than the traditional one. Finally, the tracking process transforms to the maximum likelihood estimate. The algorithm is applied in the TMS320C6416 hardware system and successfully copes with the low contrast (LSCR=4.9) airplane in the cluster background. A series of experiments results show that the lowest limitation of tracking the target by the proposed method is about 3 (LSCR value). |
| Key words: likelihood similarity function mean-shift real-time tracking Bhattacharyya coefficient |
