Antenna Obstacle Detection And Semi-dense Point Cloud Reconstruction For MAV

With the continuous development of MAV(Miniature unmanned aerial vehicle)applications and the growing sophistication of visual technology,vision-based obstacle detection algorithms are widely used in MAV autonomous obstacle avoidance navigation.In the flying environment of a micro-drone,there are obstacles such as an antenna,hereinafter referred to as an antenna obstacle,which poses a great threat to the autonomous navigation of the micro-drone,and such obstacles use traditional distance sensors and binocular vision detection effects are not satisfactory,so antenna obstacle detection and depth estimation have their practical significance.In addition,the sparse point cloud can not meet the application needs,so semi-dense point cloud reconstruction technology received more and more scholars attention.In this paper,firstly,the research on antenna obstacle detection and depth estimation is studied in depth.Secondly,the semi-dense point cloud reconstruction and point cloud denoising smoothing techniques are deeply studied according to the application requirements.The main work of this paper is as follows:1)Research on antenna obstacle detection algorithm.This method uses texture extraction,texture denoising,image segmentation,grayscale histogram detection and depth constraint to extract the antenna obstacle.Firstly,the image is preprocessed by edge texture extraction and texture denoising.Secondly,the antenna ROI is extracted by image segmentation.Then,the non-antenna region of ROI is removed by gray histogram detection and the final antenna area is extracted by the depth constraint;Finally,the antenna detection work.is completed through the growth of antenna.2)Research on algorithm of antenna obstruction depth estimation.The depth of the antenna obstacle is equivalent to reconstruct the three-dimensional information of the antenna obstacle for the pinhole camera model.Firstly,the inter-frame matching technique is studied because the correct match is the prerequisite for solving the depth.Secondly,the uncertainty model of the matching uncertainty is studied and an uncertain model is proposed.Then,We study the fusion algorithm of depth based on Kalman Filter and Bayesian Estimation.Finally,a fast denoising and smoothing method is proposed.3)Research on reconstruction algorithm of semi dense point cloud.Firstly,the semi-dense point cloud is obtained by semi-global stereo matching.Then,the denoising of the semi-dense point cloud is carried out by density clustering of point cloud and the depth connected domain detection algorithm.A method of calculating adaptive denoising radius threshold is proposed according the deep analysis of the principle of point cloud acquisition.Finally,smoothing of point cloud is performed by the bilateral filter.