北京航空航天大学学报 ›› 2016, Vol. 42 ›› Issue (2): 310-317.doi: 10.13700/j.bh.1001-5965.2015.0113

• 论文 • 上一篇    下一篇

复杂动态场景下目标检测与分割算法

许冰, 牛燕雄, 吕建明   

  1. 北京航空航天大学仪器科学与光电工程学院, 北京 100083
  • 收稿日期:2015-03-05 出版日期:2016-02-20 发布日期:2016-03-01
  • 通讯作者: 牛燕雄,Tel.:010-82316906-868 E-mail:niuyx@buaa.edu.cn E-mail:niuyx@buaa.edu.cn
  • 作者简介:许冰 女,博士研究生。主要研究方向:目标的检测与识别。Tel.:010-82316906-868 E-mail:xubing@buaa.edu.cn;牛燕雄 男,博士,教授,博士生导师。主要研究方向:光电对抗。Tel.:010-82316906-868 E-mail:niuyx@buaa.edu.cn

Object detection and segmentation algorithm in complex dynamic scene

XU Bing, NIU Yanxiong, LYU Jianming   

  1. School of Instrumentation Science and Opto-electronics Engineering, Beijing University of Aeronautics and Astronautics, Beijing 100083, China
  • Received:2015-03-05 Online:2016-02-20 Published:2016-03-01

摘要: 在动态场景等复杂条件下,往往难以对序列图像目标进行准确的检测与分割。根据序列图像中目标在复杂条件下的成像特点,提出了一种基于融合尺度不变特征变换(SIFT)流特征显著模型的动态场景目标检测与分割算法。通过对SIFT流算法表示运动特征信息的优势进行分析,并结合图像国际照明协会(CIE)Lab颜色空间的颜色和亮度特征信息,建立四维特征向量空间。利用改进的多尺度中心-环绕对比方法生成各特征通道的显著图并进行线性融合,建立序列图像的动态场景目标显著模型。最后利用均值漂移聚类算法和形态学处理实现对检测目标的精确分割。实验结果表明,相比传统检测与分割算法,该算法在动态背景与航拍等复杂场景下能够分割出更为完整的目标区域,具有良好的鲁棒性和高分割精度。

关键词: 显著性检测, 运动目标, 尺度不变特征变换(SIFT)流, 图像分割, 动态场景

Abstract: In complex conditions of dynamic scenes, it is difficult to detect and segment objects accurately in image sequence. According to the image characteristics of the object in complex conditions, we propose an object detection and segmentation model which was fused with scale invariant feature transform (SIFT) Flow characteristics in dynamic scene. Through analyzing the advantages of the movement characteristic information by SIFT Flow, and combining the color and brightness information in Commission Internationale de L'Eclairage (CIE) Lab, we establish a four-dimensional vector space. We utilize the improved multi-scale center-surround comparison method to generate salient map in each channel and fuse by linear superposition, then establish the dynamic scene saliency object model in image sequence. Finally, mean-shift clustering algorithm and morphology are used to achieve object segmentation accurately. Experimental results indicate that the proposed method can segment more complete object region than the traditional method in complex dynamic scenes and aerial video. And it also has good robustness and high segmentation accuracy.

Key words: saliency detection, motion object, scale invariant feature transform (SIFT) Flow, image segmentation, dynamic scene

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发