オルソモザイクが正しく処理されていません

McKenzieBredemeyer

私はArcGIS Flightのオフラインで森林内のフライトプランを作成し、それをSiteScanに送信して処理しました。ミッションエリア全体で画像は正しく撮影されたにもかかわらず、オルソモザイクには大きな穴があり、端だけでなく内部にも多くの歪みがあります。画像の再処理を何度も試み、新しいプロジェクトにアップロードしても状況は変わりませんでした。これはArcGIS Flight/SiteScanでの4回目のフライトですが、他の3回はより大きく複雑なミッションであったにもかかわらず、問題なく処理されました。どなたか洞察や将来的に再フライトを避けるための対策をご存知でしょうか？SiteScanバージョン4.210.2、ArcGIS Flightバージョン5.1、DJI Mavic 2 Proドローン、フライトプランは地形認識オンで完全にオフラインで実行され、飛行中に異常はありませんでした。この4回目のミッションだけですが、このドローンは過去1年間使用しており、DroneDeployなど他のソフトウェアで画像からオルソモザイクへの処理で問題が起きることはほとんどありませんでした。<\/P>

<\/span><\/P><\/span><\/P> <\/P> <\/P>

NicoBonnafoux

Hi @McKenzieBredemeyer, 

Thanks for raising this, great question and two different yet common scenarios that both pose challenges: 

1 - Mapping forested areas resulting in holes and distortion in the orthomosaic

Dense vegetation (like forest canopy) is one of the most challenging environments for photogrammetry. Even when image coverage looks good, the core challenge is feature matching:

• Forests contain highly repetitive textures (leaves, branches, shadows), which makes it difficult for the software to identify stable, unique tie points between images.
• When confidence in image matching drops, this can result in:
  • Gaps where surfaces can’t be reconstructed reliably
  • Local distortion, not just around the edges

One important nuance, especially given the comparison to other software, is how gaps are handled:

Some photogrammetry engines will aggressively interpolate or “fill in” areas where matching confidence is low. This can make an orthomosaic appear more complete, but those filled areas are often estimated rather than directly supported by strong image evidence, which can reduce accuracy.

With the Reality Engine used in Site Scan, our approach has historically been more conservative. When the software cannot establish sufficient confidence (common in dense or repetitive vegetation) we prefer to leave gaps rather than fabricate detail. From an engineering standpoint, we generally favor exposing uncertainty over hiding it.

That said, we also recognize that large canopy gaps aren’t always desirable for end users. Based on customer feedback, we’re beginning to apply limited, targeted interpolation in heavily vegetated areas, while still prioritizing accuracy and transparency. This is an area of active improvement that has started to be introduced and you will see further enhancements in the two months or so. 

A few practical considerations for future flights in forested environments:

• Increase overlap, which is the single most effective way to improve reconstruction reliability
• Maintain consistent lighting where possible
• If you’re considering a future drone purchase, a platform with a mechanical shutter can also help. Mechanical shutters reduce motion‑related distortion compared to rolling‑shutter sensors (like the one on the Mavic 2 Pro), which can be beneficial in complex, high‑texture scenes such as vegetation
  
  
  

2- Mapping around tall structure/buildings

Flight planning applications, including ArcGIS Flight, calculate overlap assuming the mapped surface is at ground level, based on sensor field of view and planned flight height. When you fly near tall buildings, the effective camera‑to‑surface distance is reduced, which means:

• Each image covers less area on the building surfaces
• Your effective overlap is lower than the planned 70/70, even if photo positions look dense. See the attached illustration: the image capture rate is consistent and desired overlap is achieved at ground level, but the top of building only appears in a single image. 
• This is why holes often appear around rooftops and facades

While crosshatch flights improve viewing geometry, they do not replace overlap when mapping vertical structures.

To compensate, especially when you can’t increase altitude due to airspace restrictions, you should:

• Increase frontlap and sidelap, even for crosshatch missions. You could perhaps fly a separate mission over the building itself with higher overlap, keeping the other area that is near ground level unchanged, then merge the images for processing. 
• Plan overlap based on the tallest features, not just ground level
• When possible, capture both nadir and oblique images. 

Hope this helps, and thanks for sharing the detailed context, it’s a great discussion.

Nico