Research On RGBD Image Sequences Scene Flow Estimation Based On Depth Automatic Layer

3D scene flow is a three dimensional motion field of space scene or object movement,which contains the three-dimensional motion and structure information of scene or object.It has important research value in the direction of target motion estimation and tracking,attitude recognition,autonomous obstacle avoidance,path planning and so on.It is widely used in aerospace,military,industrial,meteorological,transportation and cultural relics protection and other fields.In recent years,with the popularity of consumer depth sensors,the use of RGBD sequence estimation 3D scene flow has become a hot topic in the field of computer vision research.Although the existing RGBD sequence scene flow calculation method can obtain more accurate estimation results,when the image sequence contains complex background and multiple moving objects,because the existing method usually uses the manual setting depth image initial stratification layer.The resulting initial scene segmentation map contains only the depth information,which leads to the inability to completely separate the independent moving target,resulting in poor performance of the scene flow estimation.In view of the above problems,this paper mainly studies the 3D scene flow estimation technology based on deep automatic hierarchical RGBD sequence.The main research work includes:1 Firstly,we introduce the classical algorithm of 3D scene flow calculation technology.The scene flow estimation based on RGBD image is analyzed,and we also introduce another classic scene flow estimation method based on binocular.2 Aiming at the problem that the existing 3D scene flow calculation method is poor in the complex background,an automatic hierarchical and segmentation optimization method based on optical flow is proposed.First,the initial stratified stratum is set,and then the K-means clustering is used to calculate the initial segmentation result of the depth image.Then,the adjacent layer is judged according to the optical flow estimation result of the RGB image sequence,and the stratified and segmented results of the depth image are finally obtained The Compared with the traditional artificial stratification method,this method can not only realize the automatic stratification,but also get the segmentation result related to the moving target,and separate the objects independently,which is more conducive to the calculation of the stratified scene flow.3 The automatic image segmentation and segmentation optimization method of depth image is applied to the calculation of RGBF image sequence hierarchical scene flow.In this paper,we introduce the energy functional of the scene of the RGBD sequence based on the deep automatic stratification,and then introduce the constraint items in detail.Finally,the algorithm of automatic stratification and segme ntation optimization is applied to the calculation of hierarchical scene flow.4 Using Middlebury 2003 test image set,Middlebury 2005 test image set,SRSF real scene image dataset,RGBD tracking scene image set test this paper depth image automatic layering and segmentation optimization in the stratified scene flow calculation application effect,and further testing The effectiveness of the method.The experimental results show that: 1)In this paper,the stratified scene flow calculation method realizes the automatic stratification,does not need to manually modify the stratified number to get the best stratification effect,and finally obtains the scene segmentation effect more accurately,Moving objects and background independent layering;2)The calculated method of the proposed method is smaller,and the result of the final scene is more in line with the real 3D motion of the scene.