Research On 3D Dense Reconstruction Algorithm Based On Monocular Image Sequences

Because of the size of the carrier,scene reconstruction is limited in binocular and multi vision system,but monocular camera is not limited.In this paper,a 3D dense reconstruction based on monocular camera sequence images is proposed to provide navigation and obstacle avoidance information for the aircraft.Firstly,the camera model and epipolar geometry theory are introduced.In order to solve the camera motion parameters,this paper proposes a uniform ORB(Oriented FAST and Rotated BRIEF)feature point detection algorithm,and a dynamic Hamming distance threshold method is used to eliminate the false matching,the camera motion parameters and image affine transformation model are solved by using the eliminating-error matching points.Then,in the dense matching part,the block matching technique is used to search the matching points on the epipolar line,and the whole image is densely matched,and the sub-pixel precision is obtained by using the parabola method.And an affine correction tracking matching algorithm is proposed to obtain the dense matching relationship between any two images.Dense depth field is reconstructed with the dense matching relationship between the reference image and any image.Median filtering technique is used to filter the image depth field sequence of reference frame.Considering the uncertainty of triangulation and the correlation between depth fields,the inverse depth filtering technique is used to fuse the depth field sequences in depth field.Considering the "hole" region in depth map,a spatial regularized median filtering algorithm based on gray constraint is proposed to filter the depth field.The experimental results show that the affine correction based tracking matching algorithm can effectively eliminate the mismatching caused by scale,rotation and illumination changes.Compared with sequential image depth field median filtering technique,the inverse depth filtering technique can obtain more stable and accurate depth field.Considering the "hole" areas,gray constraint space regularization algorithm based on median filtering is proposed in this paper,which effectively eliminates the noise in thedepth of field and the "hole",and the depth field is more smooth and fine.