基于变分光流的三维运动检测与稠密结构重建

3D motion detection and dense structure reconstruction based on variational optical flow

提出了一种基于变分理论的由稠密光流检测图像中目标物体三维运动与结构的直接方法。首先在透视投影模型下将目标物体的三维运动参数与图像光流联系起来,构建了一个未知量是三维运动参数的能量函数,然后使用变分方法直接求出图像中目标运动物体的三维运动参数,最后由求出的三维运动参数恢复物体表面的三维稠密结构及图像光流。该方法在对目标物体进行三维运动检测时省略了"光流计算"这一中间过程,避免了图像光流计算给三维运动检测结果带来的误差影响和时间消耗。多组实验证明该算法具有较高的计算精度、较好的鲁棒性和较快的计算效率。

In this paper, a direct method of 3 D motion detection and dense structure reconstruction of the target object based on variational optical flow algorithm is proposed. Firstly, the relationship between 3D motion parameters and image optical flow under the perspective projection model is given. Then an energy function with the 3D motion parameters as its unknown variables is constructed and the 3 D motion parameters are calculated using the variational method. Finally, the 3D dense structure and image optical flow are restored utilizing the obtained 3D motion parameters. The new method omits the ＂optical flow calculation＂ in the 3 D motion detection process, which could avoid the large error and time cost in 3 D motion detection caused by the image optical flow calculation. Multiple groups of experiments prove that the proposed method features high precision, good robustness and fast calculation speed.