基于时空纹理图象能量分析的运动和深度估计

本文提出用于运动理解和景深估计的三维时空纹理图象的运动能量方法，研究时空纹理和物体运动的关系，并从时域采样的角度研究了对应性问题。在此基础上，着重讨论了在已知运动的情况下，由时空运动纹理的频谱分布来定性估计深度的算法，并给出实验结果和分析。对于一般的运动，作者提出了采用图象注视变换和时域重采样的方法，使复杂的三维时空图象的分析简化为二维时空纹理的方向检测。本文研究可广泛应用于视觉导航、自动检测、安全监视和视频图象分类等领域。

Motion energy model for estimating range and motion using spatio-temporal textural images

he authors present the motion energy method for range and motion estimation using 3D spatio-temporal (ST) textural images. The relationship between motion and range with ST oriented textures is studied in spatio--temporal-frequency domain, and the correspondence problem is observed as a temporal sampling problem. A new approach of range estimation is proposed using the Fourier energy spectrum of 2D ST textural images under the translational motion. An arbitary motion is approximately transformed to an 1D translational motion by using techniques of image gaze transformation and temporal resampling. Experimental results with real scene images on campus have shown the efficiency and robustness of the approach. The work has practical applications in visual navigation, automatic detection, monitoring and surveillance, and video classification.