A Parallel Research On Feature Point Detection Based On 3D Reconstruction With Multi-View

In the field of computer vision,3D reconstruction technique has become a major research focus. The three-dimensional reconstruction algorithm based on multi-view three-dimensional reconstruction is an important branch, the core idea of which is to apply more than one pictures of the same object or scene taken from different angles and the inside and outside parameters captured by the camera as input, generate sparse 3D point cloud data after a series of processes, then expand and filter the point cloud to obtain dense 3D point cloud data, and finally obtain three-dimensional model through the surface reconstruction. Feature point detection, an essential step based on multi-view 3D reconstruction, is to respectively extract feature point information of a multiple photos via feature point detection algorithm.This dissertation firstly introduces the development of three-dimensional reconstruction technology, and then focuses on the basic principles of multi-view 3D reconstruction algorithm (PMVS) based on patch. In feature point detection section, it focuses on the ideological principles of Harris’feature detection algorithm and DoG feature point detection algorithm. The dissertation applies parallel study on the Harris and DoG feature detection algorithm used in PMVS algorithm, and serial experiments and analysis on the feature point detection process. There is a certain degree of data independence in Harris feature detection algorithm and DoG feature point detection algorithm, and the quantity of operands is large, thus there is a feasibility of parallelism. This dissertation, on the basis of the OpenMP multi-core CPU programming model and the GPU built by parallel environment of CUDA and OpenCL, designs and implements parallel on the two feature point detection algorithm, and puts kermit and hallFeng image set of different data scales in experiments on different experimental platform. In experimental results, it analyzes the feature point detection efficiency, the detection accuracy and scalability to verify the parallelism and scalability of the feature point detection process.