The accurate detection of cooperative targets plays a key and foundational role in unmanned aerial vehicle (UAV) landing autonomously. The standard method based on fixed threshold is too susceptible to both illuminati...The accurate detection of cooperative targets plays a key and foundational role in unmanned aerial vehicle (UAV) landing autonomously. The standard method based on fixed threshold is too susceptible to both illumination variations and interference. To overcome issues above, a robust detection algorithm with triple constraints for cooperative targets based on spectral residual (TCSR) is proposed. Firstly, by designing an asymmetric cooperative target, which comprises red background, green H and triangle target, the captured original image is converted into a Lab color space, whose saliency map is yielded by constructing the spectral residual. Then, the triple constraints are developed according to the prior knowledge of the cooperative target. Finally, the salient region in saliency map is considered as the cooperative target, and it meets the triple constraints. Experimental results in complex environments show that the proposed TCSR outperforms the standard methods in higher detection accuracy and lower false alarm rate.展开更多
针对现有深度学习光流计算模型在运动遮挡和大位移等场景下光流计算的准确性与鲁棒性问题,本文提出一种联合遮挡约束与残差补偿的特征金字塔光流计算方法.首先,构造基于遮挡掩模的光流约束模块,通过预测遮挡掩模特征图抑制变形特征的边...针对现有深度学习光流计算模型在运动遮挡和大位移等场景下光流计算的准确性与鲁棒性问题,本文提出一种联合遮挡约束与残差补偿的特征金字塔光流计算方法.首先,构造基于遮挡掩模的光流约束模块,通过预测遮挡掩模特征图抑制变形特征的边缘伪影,克服运动遮挡区域的图像边缘模糊问题.然后,采用特征图变形策略构建基于特征变形的光流残差补偿模块,利用该模块学习到的残差光流细化原始光流场,改善大位移运动区域的光流计算效果.最后,采用特征金字塔框架构建联合遮挡约束与残差补偿的光流计算网络模型,提升大位移和运动遮挡场景下的光流计算精度.分别采用MPI-Sintel(Max-Planck Institute and Sintel)和KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute)数据集对本文方法和代表性传统光流计算方法、深度学习光流计算方法进行综合对比分析,实验结果表明本文方法相对于其他方法能够有效提升大位移和运动遮挡场景下的光流计算精度与鲁棒性.展开更多
Message passing algorithms,whose iterative nature captures complicated interactions among interconnected variables in complex systems and extracts information from the fixed point of iterated messages,provide a powerf...Message passing algorithms,whose iterative nature captures complicated interactions among interconnected variables in complex systems and extracts information from the fixed point of iterated messages,provide a powerful toolkit in tackling hard computational tasks in optimization,inference,and learning problems.In the context of constraint satisfaction problems(CSPs),when a control parameter(such as constraint density)is tuned,multiple threshold phenomena emerge,signaling fundamental structural transitions in their solution space.Finding solutions around these transition points is exceedingly challenging for algorithm design,where message passing algorithms suffer from a large message fiuctuation far from convergence.Here we introduce a residual-based updating step into message passing algorithms,in which messages with large variation between consecutive steps are given high priority in the updating process.For the specific example of model RB(revised B),a typical prototype of random CSPs with growing domains,we show that our algorithm improves the convergence of message updating and increases the success probability in finding solutions around the satisfiability threshold with a low computational cost.Our approach to message passing algorithms should be of value for exploring their power in developing algorithms to find ground-state solutions and understand the detailed structure of solution space of hard optimization problems.展开更多
基金supported by the National Natural Science Foundation of China(61135001)the Scientific Research Program of Shaanxi Provincial Department of Education(16JK1499)+2 种基金the Doctoral Fund of Xi’an University of Science and Technology(2015QDJ007)the Cultivation of Xi’an University of Science and Technology(2014015)the Ministry of Education Key Laboratory of Information Fusion Technology(LIFT2015-G-1)
文摘The accurate detection of cooperative targets plays a key and foundational role in unmanned aerial vehicle (UAV) landing autonomously. The standard method based on fixed threshold is too susceptible to both illumination variations and interference. To overcome issues above, a robust detection algorithm with triple constraints for cooperative targets based on spectral residual (TCSR) is proposed. Firstly, by designing an asymmetric cooperative target, which comprises red background, green H and triangle target, the captured original image is converted into a Lab color space, whose saliency map is yielded by constructing the spectral residual. Then, the triple constraints are developed according to the prior knowledge of the cooperative target. Finally, the salient region in saliency map is considered as the cooperative target, and it meets the triple constraints. Experimental results in complex environments show that the proposed TCSR outperforms the standard methods in higher detection accuracy and lower false alarm rate.
文摘针对现有深度学习光流计算模型在运动遮挡和大位移等场景下光流计算的准确性与鲁棒性问题,本文提出一种联合遮挡约束与残差补偿的特征金字塔光流计算方法.首先,构造基于遮挡掩模的光流约束模块,通过预测遮挡掩模特征图抑制变形特征的边缘伪影,克服运动遮挡区域的图像边缘模糊问题.然后,采用特征图变形策略构建基于特征变形的光流残差补偿模块,利用该模块学习到的残差光流细化原始光流场,改善大位移运动区域的光流计算效果.最后,采用特征金字塔框架构建联合遮挡约束与残差补偿的光流计算网络模型,提升大位移和运动遮挡场景下的光流计算精度.分别采用MPI-Sintel(Max-Planck Institute and Sintel)和KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute)数据集对本文方法和代表性传统光流计算方法、深度学习光流计算方法进行综合对比分析,实验结果表明本文方法相对于其他方法能够有效提升大位移和运动遮挡场景下的光流计算精度与鲁棒性.
基金supported by Guangdong Major Project of Basic and Applied Basic Research No.2020B0301030008Science and Technology Program of Guangzhou No.2019050001+2 种基金the Chinese Academy of Sciences Grant QYZDJ-SSWSYS018the National Natural Science Foundation of China(Grant No.12171479)supported by the National Natural Science Foundation of China(Grant Nos.11301339 and 11491240108)。
文摘Message passing algorithms,whose iterative nature captures complicated interactions among interconnected variables in complex systems and extracts information from the fixed point of iterated messages,provide a powerful toolkit in tackling hard computational tasks in optimization,inference,and learning problems.In the context of constraint satisfaction problems(CSPs),when a control parameter(such as constraint density)is tuned,multiple threshold phenomena emerge,signaling fundamental structural transitions in their solution space.Finding solutions around these transition points is exceedingly challenging for algorithm design,where message passing algorithms suffer from a large message fiuctuation far from convergence.Here we introduce a residual-based updating step into message passing algorithms,in which messages with large variation between consecutive steps are given high priority in the updating process.For the specific example of model RB(revised B),a typical prototype of random CSPs with growing domains,we show that our algorithm improves the convergence of message updating and increases the success probability in finding solutions around the satisfiability threshold with a low computational cost.Our approach to message passing algorithms should be of value for exploring their power in developing algorithms to find ground-state solutions and understand the detailed structure of solution space of hard optimization problems.
