基于增量式SFM的特定目标加速定位方法

Specific Target Acceleration Positioning Method Based on Incremental SFM

针对当前增量式运动恢复结构法中每加入一幅图片需要循环多次光束平差迭代造成的大计算量和三维重建过程点云目标针对性弱等问题,提出两步改进方法。首先,引入一种分段调节函数。在三角化步骤之后,计算系统的重投影误差,通过比较此误差与事先设定的阈值,以此判断后续局部迭代优化步骤的运行方式。随后,待所有图像注册完毕,使用SURF算法进行二次图像匹配。确定特定目标的图像坐标和对应空间相对位置,并完成对应点云重建过程,同时滤除无用冗余信息。实验证明,改进方法在基本维持原有定位精度的基础上,可以较大的提高系统运行速度,并且能够从众多目标中快速准确地找到所需目标位置,最终生成空间点云,具有较强的实用价值。

Aiming at the problem that the large amount of computation caused by multiple local BA iterations and the weak target of the point cloud target in the three-dimensional reconstruction process are added for each incremental SFM in the current incremental SFM,a two-step improved method is proposed.First,a segmentation adjustment function was introduced.After the triangulation step,the reprojection error of the system was calculated,and the operation mode of the subsequent local iteration BA step was judged by comparing the error with a preset threshold.Subsequently,after all images have been registered,the SURF algorithm was used for secondary image matching.The image coordinates of the specific target and the relative position of the corresponding space were determined,and the corresponding point cloud reconstruction process was completed,while filtering out the useless redundant information.The experiment proves that the improved method can greatly improve the system running speed based on the basic maintenance accuracy,and can find the required target position from many targets quickly and accurately and generate a spatial point cloud finally with strong practical value.