计算机与现代化

• 人工智能 • 上一篇    下一篇

基于纹理特征改进的GFL运动目标提取方法

  

  1. (河海大学能源与电气学院,江苏南京211100)
  • 收稿日期:2018-05-25 出版日期:2019-01-30 发布日期:2019-01-30
  • 作者简介:窦修超(1995-),男,江苏仪征人,硕士研究生,研究方向:机器视觉与智能监护系统,E-mail: owen_dou@163.com; 李志华(1964-),男,江苏泰州人,教授,硕士生导师,博士,研究方向:人工智能与复杂系统故障诊断。
  • 基金资助:
    江苏省自然科学基金资助项目(BK20151500)

Improved GFL Moving Target Extraction Method Based on Texture Features

  1. (College of Energy and Electrical Engineering, Hohai University, Nanjing 211100, China)
  • Received:2018-05-25 Online:2019-01-30 Published:2019-01-30

摘要: 基于广义融合套索(GFL)前景模型,融合视频的纹理特征,提出一种基于纹理特征的运动目标提取方法。方法通过GFL前景模型提取前景运动目标和背景,再利用LBP算法提取前景与背景在多个方向上的纹理特征,比较两者纹理特征的相似度,去除前景中的投射阴影,解决由于运动目标遮挡产生的阴影问题,同时还引入误判率去描述模型的准确度。通过对广场、办公室以及体育馆等实际场景进行测试,实验表明提出的算法能够有效去除运动目标产生的阴影。

关键词: 前景检测, 纹理特征, 阴影去除

Abstract: Based on Generalized Fused Lasso (GFL) foreground model and the texture information of video, this paper proposes a moving target extraction method based on texture features. This method uses GFL foreground model to extract foreground moving object and background. Then it extracts the texture features of foreground and background in many directions by LBP algorithm, compares the similarity of two texture features, and removes the misjudged shadow regions in the foreground, which can reduce the cast shadows due to occlusion of moving targets. The paper also introduces the misjudgment rate to describe the accuracy of model. By testing real scenes that contain cast shadows, such as squares, offices, and gymnasiums, the proposed algorithm can effectively monitor the areas where shadows are cast.

Key words: foreground detection, texture features, shadow removal

中图分类号: