Research On Calibration And Application Of Airborne Non-overlapping Cameras

Unmanned aerial vehicle(UAV)aerial photography plays an important role in military reconnaissance,security monitoring,and aerial surveying and mapping.Existing drone aerial photography equipment is mainly equipped with a single camera,and it is impossible to obtain large scene information at a single moment.In order to improve the perception ability of the aerial photography system at a single moment,a camera array system without public field of view is designed.Aiming at the problem that the overlap area of airborne non-overlapping cameras is small and the pixel relationship between frames is difficult to establish,this paper studies and designs the corresponding calibration method to complete the precise calibration of the system.On this basis,the aerial surveying and mapping and pedestrian detection related applications are completed based on the airborne non-overlapping cameras.The specific work is as follows:(1)Design the airborne non-overlapping cameras system.According to the system application requirements,carry out the investigation and selection of the airborne body,image acquisition board,and embedded camera.Comprehensive consideration of the field of view angle of the camera and the structure of the airborne body design the airborne non-overlapping cameras,and complete the overall system construction.(2)A calibration method of non-overlapping camera array based on cooperative target is studied.Aiming at the difficulty in establishing the pixel correspondence between nonoverlapping camera frames,an auxiliary camera is introduced to shoot the cooperative target,and the non-overlapping multi-camera problem is decomposed into the sub-problems of multiple adjacent binocular camera pose estimation,and adopted the bundle adjustment method optimizes the system calibration results.The experimental results show that the reprojection error of the ground-based non-overlapping camera array system and the air-borne non-overlapping camera array system is less than 0.2 pixels,and the calibration method based on the cooperative target has high calibration accuracy.(3)A measurement calibration method based on visual SfM(Structure from motion)and inertial measurement data for airborne non-overlapping camera arrays is proposed.First,use the consistency information of the stationary objects in the shooting scene in different cameras to establish the pixel correspondence between non-overlapping visible cameras,and use the SfM method to calculate the relative poses between non-overlapping visible cameras.Secondly,by constructing a nonlinear equation between the inertial measurement acceleration and the camera measurement acceleration,the measurement scale estimation of the calibration result is completed.Finally,through experimental analysis,compared with the calibration results of the airborne non-overlapping camera array based on the cooperative target,the estimated relative attitude angle error of this method is less than 0.2°,the position error is less than 0.015 m,and the measurement scale error is less than 2%.The calibration of the airborne non-co-visible camera array can be completed more accurately in the scene without a cooperative target.(4)Research on the application of airborne non-overlapping camera array system.On the basis of system calibration,this article applies it to aerial 3D measurement and pedestrian detection.On the basis of metric calibration,the three-dimensional metric reconstruction of the object is completed.Experiments show that the average accuracy of the system's scale measurement is 97.28%.The system is used for pedestrian detection,through the multidirectional image acquisition of the non-overlapping cameras system,and the use of YOLOv4-Tiny to perform pedestrian detection on the acquired image,which can realize the monitoring of the area with a larger field of view at a single moment.