Research On Preprocessing Method For 3D Model Reconstruction Based On Redundant Image Filtering

The popularity of civilian small drones provides strong data support for the construction of urban 3D models.The visualization advantages of 3D models over 2D models have gradually emerged with the development of hardware devices.This also brings new challenges: due to the susceptibility of civilian small unmanned aerial vehicles to equipment jitter and harsh natural environments during the shooting process,their captured images are prone to a series of distortion issues such as blurring and out of focus.However,3D reconstruction of distorted unmanned aerial vehicle images can have a significant impact on the quality of the reconstructed model.In addition,due to the limitations of aerial photography height and image size,the number of images captured by drones is generally large and redundant,greatly increasing the reconstruction time.Therefore,at present,using civilian small unmanned aerial vehicles to collect data and perform 3D reconstruction is often slow,and the reconstruction results are prone to quality issues,which is not conducive to efficient modeling of large-scale complex scenes at low cost.In response to the above issues,research has shown that filtering the drone image set can be used to preprocess the image set participating in 3D reconstruction,thereby achieving the goal of improving the overall efficiency of 3D reconstruction.This includes image redundancy reduction methods for fixed thinning of drone image overlap.However,due to the susceptibility of drones to self shaking and crosswind during the shooting process,the overlap of their shots may differ from the preset values.Therefore,using a fixed thinning method to reduce the redundancy of the image set may result in insufficient overlap of some images.Images with insufficient overlap may affect the3 D modeling results during the 3D reconstruction process due to insufficient matching of feature points.Due to the emergence of this issue,this article proposes a new drone aerial image filtering method.This method can delete as many drone images with poor quality as possible while maintaining the minimum overlap requirement for 3D reconstruction in the image set,thereby achieving the goal of improving the efficiency of 3D reconstruction modeling.Which distortion will affect the reconstruction results of the3 D model;How to determine the degree of image distortion;How to obtain actual overlap data of drone images;How to develop a screening strategy is a key issue in the drone aerial image screening method.This article conducts the following research on the above issues:(1)Explored the impact of some unsatisfactory factors in drone imaging on 3D reconstruction,found a quality evaluation algorithm suitable for drone images,and improved it.This article first analyzes the common distortion factors in the imaging process of unmanned aerial vehicles,and conducts experiments on the motion blur and defocus factors among them.Based on the experimental results,the impact of these two types of distorted images on 3D model reconstruction is verified.At the same time,in order to find suitable evaluation algorithms for drone images,this article trained four algorithms based on natural scene statistical NSS,and found that the BRISQUE algorithm had the best performance through comparison.In order to make the algorithm more accurate in quality evaluation of UAV images,this paper adjusts the loss function epsilon of BRISQUE algorithm.(2)Explored the impact of overlap on 3D reconstruction and proposed a method for calculating overlap based on feature point matching.This article analyzes the 3D modeling results under different overlapping images through experiments,and obtains the impact of overlapping on the 3D reconstruction modeling results.At the same time,the proposed method for calculating the overlap of drone images based on feature point matching can quickly and accurately calculate the overlap of a large number of drone images compared to traditional manual statistical methods.(3)A drone image filtering method based on image quality and overlap is proposed.This method evaluates the quality of drone images to obtain the quality of each image in the drone image set;Simultaneously simulate the drone route to accurately obtain the corresponding heading and side facing images of each image;After sorting the quality of drone images,recursive methods are used to sequentially determine whether the remaining images after deletion meet the minimum overlap requirements.After experimental comparison,the drone image redundancy reduction method proposed in this article overcomes the irregular and missing blocks in the reconstruction results caused by traditional redundancy reduction methods,and the efficiency optimization is not inferior.While improving the efficiency of 3D reconstruction,it can also ensure the quality of reconstruction results and achieve more complete and high-quality reconstruction results.