Research On Panoramic Stitching Of Drone Aerial Images And Dynamic Target Trajectory Reconstruction Technology

With the development of UAV(unmanned aerial vehicle)technology,the research of aerial image related technology based on UAV has become a research hotspot.Different from the image content collected by ground camera equipment,drone aerial photography scenes are mostly regional,engineering,disastrous,and military regional monitoring.Its aerial images have the characteristics of high resolution,high timedependent correlation,large parallax,and flexible shooting angles.However,due to the influence of drone flight stability,wind,light,drone aerial images also have image characteristics.Stability,large image inclination,large number,irregular overlap,put forward higher requirements on UAV aerial image processing technology.This paper takes UAV aerial images as the research goal,and mainly studies key technologies such as panoramic image stitching and target trajectory reconstruction under the large parallax view of UAV.The main work is as follows:1.Propose a method for generating large parallax panoramic images for UAV aerial images.Due to the performance limitations of imaging equipment,current UAV imaging systems mostly use sequence image online dynamic splicing technology to obtain largearea,high-resolution panoramic images.Aiming at the problem of panorama stitching with large parallax from the perspective of drones,this paper proposes a panorama stitching algorithm based on semantic segmentation.This algorithm introduces image registration methods and uses a semantic segmentation network to segment the foreground and background of the stitched image.Only the background score is matched,which effectively solves the problem of the false matching of feature points caused by the movement of the foreground target,which leads to the blurring of the stitching result,and effectively solves the constraint that the traditional panoramic stitching algorithm can only be applied to the stitching of static large parallax images problem.Experiments show that the method proposed in this paper has achieved good results in panoramic image stitching under UAV scenes with large parallax.2.A lightweight moving target tracking algorithm for UAV aerial images is proposed.Compared with satellite remote sensing images,a significant advantage of UAV aerial images is its ability to track moving targets.Therefore,the research of moving target trajectory modeling technology represented by vehicles,pedestrians,etc.is another hot issue in UAV image research.Aiming at the problem of moving target trajectory reconstruction from the perspective of drones,this paper proposes a target tracking algorithm based on deep learning.The algorithm uses a mixture of Gaussian background modeling to optimize the uninteresting area,and the color extraction method is used to extract the panoramic image.The color of the marked moving target is extracted,and the final trajectory of the moving target is reached.In addition,through lightweight tailoring of the target network,the algorithm can be transplanted under the limited computing power of the airborne computer.Experiments show that this method has good target trajectory reconstruction accuracy.