论文部分内容阅读
针对移动机器人面对的真实3维场景数据,提出一种基于频域和空域混合分析的视觉显著性检测方法.首先设计多通道特征融合算法融合RGB-D数据中包含的颜色和深度信息,然后通过超复数傅里叶变换在频域计算得到多尺度视觉显著图,接着利用非均匀超像素分割算法对得到的显著图进行平滑处理,从而消除离散背景噪声干扰,改善频域检测结果.最后,采用元胞自动机对多尺度视觉显著图进行有效融合,提取最终的显著性区域.在公开数据库上进行了多组实验,验证了所提出算法在移动机器人面对的真实复杂场景数据中的有效性.
Aiming at the real 3D scene data faced by mobile robot, this paper proposes a visual saliency detection method based on hybrid analysis of frequency domain and spatial domain.Firstly, multi-channel feature fusion algorithm is designed to fuse the color and depth information contained in RGB-D data, and then The multiscale visual saliency map is calculated by using the hypercomplex Fourier transform in the frequency domain, and then the saliency map is smoothed by using the non-uniform subpixel segmentation algorithm to eliminate the interference of discrete background noise and improve the detection results in the frequency domain.Finally, Cellular automaton is used to effectively fuse multi-scale visual saliency maps and extract the final salient regions.Many experiments are carried out on the open database to verify the effectiveness of the proposed algorithm in real complex scene data faced by mobile robots Sex.