Fast Optical Flow Estimation And Spatio-Temporal Coherence Reconstruction

Optical flow is an important research field in computer vision.It is used in motion segmentation,object recognition,object tracking,video difference,3D reconstruction and so on.Optical flow estimation is a classical and fundamental problem in computer vision.Since the optical flow is proposed,a large number of researchers and scholars began to study the optical flow estimation.In recent years,experts and scholars continued to put forward many effective methods for optical flow estimation.At present,there are still some problems in optical flow estimation,for example,it is difficult to obtain satisfactory results in the practical engineering applications,facing large displacement problem.The computational complexity is very high.Even if the use of GPU is difficult to achieve real-time estimation.In this paper,a fast and effective optical flow estimation method for large displacement problems is proposed,which can effectively and quickly obtain the accurate optical flow estimation results.In this method,the matching method is combined with the traditional variational optimization method to deal with the problem of optical flow estimation under large displacement circumstance.With the help of feature matching,increasing the number of reliable matching points,the dense initial optical flow field can be obtained quickly.Based on the initialization of the optical flow field,the fast and high quality dense optical flow field estimation is realized by using a optimization method based on variational optimization.With GPU acceleration,our algorithm can obtain optical flow result very fast,which can satisfy some applications that requiring fast optical flow estimation.Our optical flow estimation algorithm is applied to the 3D reconstruction.Based on the correlation information between consecutive frames,the temporal and spatial coherence depth optimization is established to improve the depth recovery and the quality of 3D reconstruction.In this paper,based on the existing depth recovery methods,using the correlation information between consecutive frames,spatio-temporal consistency optimization is used to improve the accuracy of depth recovery and 3D reconstruction result.Experimental results show that the proposed method has good performance in KITTI data set.Our method can quickly obtain optical flow result,under the premise of ensuring a certain accuracy.Using our optical flow estimation result,we optimize the original depth with spatio-temporal constraints,and get more accurate depth information and 3D reconstruction results.