I am in pursuit of the perfect 3dModel , i am using a headland of about 1Km with a few structures atop it from human sized to small house sized. i am limited by the camera on a mavic 2 pro, i am building a manual capture flight using litchi unless a pix4Dcapture flight can be shown to give me the granular results of any particular site, and i am am not limited to processing as i am using an AWS cloud GPU with 192G of RAM etc I am trying to understand some best practices and will post results and hope to get the help of community and PIX4d in what will result in a fairly exciting high profile commercial project given my results. Thus i query input on these and any other recommendations
Whenever i zoom into my model , it loses clarity and definition as i approach a 1:1 ratio compared to the image shot. What factors wil improve this in pix4D mapper given i am at the highest standard settings for texture and modeling. is it about creating more coordinating points and how should i understand scalling these up and how it improves quality
does it make a difference if i use adjusted DNG ( raw) files or their corresponding converted JPGs. aside from the image size itself , will i approximate quality with proximity to the objects i am seeking to model. i understand the option of handmodelling, i am trying to understand the limits of drone photogrammetry reconstruction.
i have had success with my own traversing low level flights with gimbal angles of anything from -30% to -50% for best results, Pix4D applies either a cross hatch nadir flight or an orbital . This i assume to be optimal after much research but what tweaks might improve further such as
- combination of both
- terrain following from the first initial point
- distance from the object for best quality not practicality
- and gimbal angles
At what point does overlap not optimise the results further or the more overlap the better. Am i correct in my layman’s assumption that too much overlap creates a burden of processing without improving results after a certain point or does the process ignore redundancy and a 99% overlap will be better than an 80% overlap of shooing images.
Does the app benefit from varied camera angles and gimbal angles for calculation of textures and shape or does it benefit form as standardised rigid a shot format at possible.