We consider the extraction of accurate silhouettes of foreground objects in combined color image and depth map data.This is of relevance for applications such as altering the contents of a scene,or changing the depths...We consider the extraction of accurate silhouettes of foreground objects in combined color image and depth map data.This is of relevance for applications such as altering the contents of a scene,or changing the depths of contents for display purposes in 3DTV,object detection,or scene understanding.To展开更多
Motionless foreground objects are key targets in applications of home care monitoring and abandoned object detection,and pose a great challenge to foreground detection.Most algorithms incorporate the motionless foregr...Motionless foreground objects are key targets in applications of home care monitoring and abandoned object detection,and pose a great challenge to foreground detection.Most algorithms incorporate the motionless foreground objects into their background models because they have to adapt to environmental changes.To overcome this challenge,a foreground detection method based on nonlinear independent component analysis(ICA) was proposed.Considering that each video frame was actually a nonlinear mixture of the background image and the foreground image,the nonlinear ICA was employed to accurately separate the independent components from each frame.Then,the entropy of grayscale image was calculated to classify which resulting independent component was the foreground image.The proposed nonlinear ICA model was trained offline and this model was not updated online,so the method can cope with the motionless foreground objects.Experimental results demonstrate that,the method achieves remarkable results and outperforms several advanced methods in dealing with the motionless foreground objects.展开更多
移动主体获得准确的定位信息是构建稳定的混合现实(mixed reality,MR)系统的关键,然而MR中的前景对象对传统定位算法的精度影响较大.现阶段基于深度学习的定位算法可以通过识别前景对象来提升精度,但深度学习模型耗时过高,导致算法实时...移动主体获得准确的定位信息是构建稳定的混合现实(mixed reality,MR)系统的关键,然而MR中的前景对象对传统定位算法的精度影响较大.现阶段基于深度学习的定位算法可以通过识别前景对象来提升精度,但深度学习模型耗时过高,导致算法实时性下降.针对该问题,提出了一种MR中融合语义特征传播模型的前景对象感知定位算法.该算法依托语义分割网络与一种快速旋转的二进制独立稳定描述子特征(oriented fast and rotated binary robust independent elementary feature,ORB)提取算法构建了语义特征传播模型,实现高速语义特征提取;融合该模型和几何特征检测方法实现算法中的前景对象感知层,并依赖该感知层剔除MR中前景对象的特征点,构建了背景特征点集,实现高精度、高实时性的定位.实验结果表明:在慕尼黑工业大学(Technical University of Munich,TUM)公共数据集的高动态前景对象场景中,相比动态语义视觉同步定位与建图(dynamic semantic visual simultaneous localization and mapping,DS-SLAM)算法,该算法相对位姿误差降低了60.5%,定位实时性提升了39.5%,可见该算法在MR中具有较高的应用价值.展开更多
基金supported by Key Project No. 61332015 of the National Natural Science Foundation of ChinaProject Nos.ZR2013FM302 and ZR2017MF057 of the Natural Science Found of Shandong
文摘We consider the extraction of accurate silhouettes of foreground objects in combined color image and depth map data.This is of relevance for applications such as altering the contents of a scene,or changing the depths of contents for display purposes in 3DTV,object detection,or scene understanding.To
基金National Natural Science Foundations of China(Nos.61374097,61601108)the Fundamental Research Funds for the Central Universities,China(No.N130423006)the Foundation of Northeastern University at Qinhuangdao,China(No.XNK201403)
文摘Motionless foreground objects are key targets in applications of home care monitoring and abandoned object detection,and pose a great challenge to foreground detection.Most algorithms incorporate the motionless foreground objects into their background models because they have to adapt to environmental changes.To overcome this challenge,a foreground detection method based on nonlinear independent component analysis(ICA) was proposed.Considering that each video frame was actually a nonlinear mixture of the background image and the foreground image,the nonlinear ICA was employed to accurately separate the independent components from each frame.Then,the entropy of grayscale image was calculated to classify which resulting independent component was the foreground image.The proposed nonlinear ICA model was trained offline and this model was not updated online,so the method can cope with the motionless foreground objects.Experimental results demonstrate that,the method achieves remarkable results and outperforms several advanced methods in dealing with the motionless foreground objects.
文摘移动主体获得准确的定位信息是构建稳定的混合现实(mixed reality,MR)系统的关键,然而MR中的前景对象对传统定位算法的精度影响较大.现阶段基于深度学习的定位算法可以通过识别前景对象来提升精度,但深度学习模型耗时过高,导致算法实时性下降.针对该问题,提出了一种MR中融合语义特征传播模型的前景对象感知定位算法.该算法依托语义分割网络与一种快速旋转的二进制独立稳定描述子特征(oriented fast and rotated binary robust independent elementary feature,ORB)提取算法构建了语义特征传播模型,实现高速语义特征提取;融合该模型和几何特征检测方法实现算法中的前景对象感知层,并依赖该感知层剔除MR中前景对象的特征点,构建了背景特征点集,实现高精度、高实时性的定位.实验结果表明:在慕尼黑工业大学(Technical University of Munich,TUM)公共数据集的高动态前景对象场景中,相比动态语义视觉同步定位与建图(dynamic semantic visual simultaneous localization and mapping,DS-SLAM)算法,该算法相对位姿误差降低了60.5%,定位实时性提升了39.5%,可见该算法在MR中具有较高的应用价值.
