如何突破卫星影像建模难点?重建大师这样做!

日前,由重建大师生成的首个“珞珈三号01星”卫星影像三维模型一经发出,引发业内关注。这是重建大师(卫星版)基于卫星所获取的多角度成像数据,生成高质量实景三维模型的最新实践成果。

卫星影像动图

相较于传统倾斜摄影数据建模,卫星影像数据采集区域广、不受航空管制、采集周期的影响,更符合当前大场景实景三维建模与城市三维模型快速构建的需要。但从获取的影像数据来说,卫星影像数据与传统航摄影像数据存在较多差异:

01.卫星影像通常是采用线阵传感器单元以推扫方式采集而来,其投影方式与传统框幅式成像模型不同;

02.卫星传感器到地面的距离过大,基线短,容易导致计算结果不稳定;

03.卫星影像的像幅过大,单幅影像可包含数十亿像素,处理困难。

因此,想要实现卫星影像快速三维重建并不能采用现有的针对传统航摄影像的处理方法,需要在空三解算和匹配算法上进行优化。

大势智慧研发团队经过长期探索,攻克多个技术难点,推出重建大师卫星版本,在已具备超大规模地理场景高质量生成与地理实体智能化生产的同时,成为业内为数不多的、能有效满足多个卫星影像数据高效率三维模型构建的软件

►►►

优化特征提取与匹配算法

解决卫星影像空三难

卫星影像采用RPC有理函数模型来表达三维坐标到影像坐标之间的投影关系,每个影像的RPC模型参数由卫星厂商给出,但是该参数存在微小的误差,必须经过校正才能用于后续的重建。

大势智慧研发出业界领先的针对卫星影像优化的图像匹配算法,具有匹配点数量多、精度高、匹配效率高等特性,平差方法采用一种适合于RPC模型的交替求解法,在无控制点的情况下仍可以获得高精度的解。

►►►

攻克多个技术难点

解决卫星影像重建难

For 3D reconstruction based on satellite images, it is generally necessary to perform epipolar correction on satellite images to obtain stereoscopic images, and then dense point clouds are obtained through dense matching. Finally, products such as DSM and grid model are output.

However, in the actual reconstruction process, there are many technical difficulties in the satellite image reconstruction process, such as epipolar line correction, dense matching, and DSM merging . Dashi Wisdom has successfully solved these problems by optimizing and rebuilding the master algorithm.

Due to the complexity of the RPC projection model, epipolar correction cannot be performed directly. Reconstruction Master performs epipolar correction based on the affine projection model. The correction algorithm has good stability and high precision, and is suitable for various satellite viewing angles, resolutions and terrain scenes. When densely matched, Reconstruction Master effectively solves the problem of unknown matching parallax range by matching multi-level resolution images, greatly enhancing the success rate of matching.

Point cloud produced by dense matching of some satellite images

In order to obtain the optimal DSM, Reconstruction Master combines local elevation reliability and global consistency when selecting the optimal elevation, while ensuring the elevation accuracy and continuity of the merged DSM.

grid model

Based on a series of algorithm upgrades, at present, the reconstruction master satellite version has completed the inspection of the same-orbit stereo and different-orbit stereo satellite image data sources with visible light resolution better than 1 meter, and the modeling efficiency can reach up to 600km²/node/day, which can Provide high-efficiency real-scene 3D model construction for the world, and provide data basis for urban 3D model LOD1.3.

The satellite image 3D reconstruction technology of Reconstruction Master (Satellite Edition) can completely replace foreign satellite processing software. After years of research and development, Dashizhizhi can not only realize localized replacement in 3D reconstruction of aerial or UAV oblique images, but also completely replace foreign software in satellite processing.

The following are satellite model results generated by Rebuild Master:

Based on the 0.3-meter image data of Beijing No. 3

(Thanks to Beijing 21st Century Company)

Guess you like

Origin blog.csdn.net/Daspatial_/article/details/129586919