基于几何投影的运动阴影去除方法  被引量:1

A Moving Shadow Removal Method Based on Geometric Projection

在线阅读下载全文

作  者:王俊[1] 王文中[1,2] 汤进[1,2] 

机构地区:[1]安徽大学计算机科学与技术学院,安徽合肥230601 [2]安徽省工业图像处理与分析重点实验室,安徽合肥230039

出  处:《计算机技术与发展》2015年第12期37-40,46,共5页Computer Technology and Development

基  金:国家"863"高技术发展计划项目(2014AA015104);国家自然科学基金资助项目(61472002);国家科技支撑计划项目(2012BAH95F01)

摘  要:由于在使用图像差分法检测运动目标时,运动目标阴影会被当成运动目标而检测出,因此文中针对固定场景的运动目标阴影问题,提出了一种简单有效的基于几何投影的阴影去除方法。首先,根据图像灭点理论,通过场景先验知识计算场景图像的主方向。其次,结合帧差法和背景差法检测运动前景,把前景像素点在场景图像主方向的垂线方向上进行几何投影。根据目标轮廓和阴影轮廓的主方向与场景图像主方向的可区分特性,通过阈值化投影值则可分割出目标区域和阴影区域,进而去除阴影区域。最后,考虑到光照、目标自身大小、目标运动方向以及运动目标速度等因素对阴影的影响,文中在不同条件的视频上进行了实验,通过与原图对比和实验间相互对比验证了文中算法的有效性。As the image difference method were used to detect moving targets, moving targets shadow was detected as a motion target, so for the shadow problem of moving object detection under the fixed scene, proposed a simple and effective shadow removal approach based on geometric projection. Firstly, the principal image direction of the scene is computed by utilizing the vanishing point theory and scene prior knowledge. Secondly, combined the frames subtraction with the background subtraction, detect the foreground pixels and geometri- cally project them on the direction that is vertical to the computed principal direction. According to the fact that the principal directions of the target and shadow contours are distinguishable with the principal image direction of the scene, segment the shadow regions from the detected foreground pixels and remove it by thresholding the projection values. Finally, considering the influence of the light, their own size of targets,the direction of the target moving and moving target speed and other factors on the shadow ,make the experiment on some videos with different conditions, and demonstrate the effectiveness of the proposed approach.

关 键 词:运动目标检测 图像主方向 几何投影 阴影去除 

分 类 号:TP391.4[自动化与计算机技术—计算机应用技术]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象