Narrow-baseline Monocular And Binocular Depth Perception Based On Selfie UAV

With the development of portable photographing equipment,the quality of photography has become an important demand in the consumer market.The"Thirteen-Five" national strategic emerging industry development plan put forward to vigorously develop multi-purpose UAV which is of high market demand.Based on the current situation of Selfie UAV,this article innovatively put forward to complete the capture of shallow-depth-of-field images in Selfie UAV.Since most Selfie UAV do not have a depth-sensing camera,depth-of-field photo capture has to depend on the 3D reconstruction of the scene three-dimensional and the acquisition of depth information.The purpose of this paper is to realize the acquisition and optimization algorithm for depth information based on the input of Selfie UAV.This article put forward two depth acquisition algorithms,whose inputs can be monocular multi-frame sequences or binocular two frames.For different input application requirements,the technology route is divided into two types:monocular multi-frame depth acquisition + composite focus and binocular stereo matching depth acquisition + composite focus.The main contents and innovative achievements of this paper are as follows:(1)The acquisition and optimization of depth information based on monocular narrow-baseline image sequences are studied.Aiming at small jitter image sequence/video input from selfie UAV,a self-calibrating structure from small motion(SFSM)algorithm is proposed to estimate camera internal parameters,camera poses as well as the sparse 3D construction.The algorithm realizes a plane sweeping algorithm to estimate initial dense depth information,and proposes a fully connected conditional random field(CRF)model guided by color information and depth information to smooth and optimize initial depth results.The influence of small baseline input on the depth uncertainty can be effectively removed and the accuracy of the depth map are improved.(2)The acquisition algorithm of depth map based on binocular small-baseline image pair is studied.An edge-aware stereo matching algorithm is proposed for recovering depth result from small baseline image pair acquired by binocular camera.By constructing the smoothing term and data item of energy function in the bilateral space and solving the function,the depth map with accurate edge information and depth values can be obtained which can effectively reduce the storage resource consumption.(3)This paper studies the algorithm of image refocusing interaction based on scene depth information,which can be applied to Selfie UAV.Based on the depth capture algorithm,a hierarchical focus synthetic algorithm by dividing the depth space is proposed to produce shallow-depth-of-field images in Selfie UAV.The algorithm can effectively improve the accuracy of synthetic focus and visual effects.