3D Digitalization Of Large-scale Unstructured Scene By Unmanned Helicopter

At present,3D digital models of large-scale and unstructured scenes have been widely used in various fields such as pattern recognition,reverse engineering,digital city,and reconstruction of ancient architectural sites.At the same time,a variety of depth acquisition devices are also widely appearing,such as laser scanning equipment,stereo-vision modules and various depth cameras adopting infrared sensors,structured light or time of flight principle.Among all kinds of schemes,it is popularand efficient modeling by the unmanned aerial platform equipped with laser scanning equipment.Because you can make full use of the flexible movement of UAV.This paper aims to solve some key problems relating to accuracy and efficiency during the modeling process with airborne laser equipment.For the scenes with complex structure and rich texture,in order to capture more details,a “fly-hover-scan” method is adopted.In the scheme,some fixed points around the scene are chosen for scanning.The aircraft will hover at the point and then laser equipment will scan the scene.But in this process,the aircraft disturbed by wind and some other external interference,will have a motion with small amplitude around the hover point.Unfortunately,the existing GPS/INS positioning system can not effectively and accurately perceive the slight movement,so the positioning error is introduced to the model error.This dissertation adopts monocular vision simultaneous localization and mapping technology combined with sensor fusion algorithm based on the EKF framework to perceive the slight movement around the hover point.The positioning result will be used for the correction of point cloud data,so as to effectively eliminate the distortion of single frame point cloud.When modeling scenes with simple texture,the “flying-scan” method is adopted.3D scanning is carried out by the motion of aircraft.But at the turning point of trajectory,air flow varies greatly and barometer can not provide height measurement accurately so that the positioning result has a significant jump in the height direction.The localization error leads to obvious stratification of the point cloud model.In order to solve this problem,this dissertation puts forward the idea of segmenting the point cloud into pieces and then putting them together after aligning them.Finally,several simulations and field experiments are provided to prove the validity and practicability of these methods.