How to processing drone images fast?

[AkhilChhibber1]

I am having 3000 drone images which I want to process to get my final output as:

1.) 2d mosaic(tiff)

2.) 3d mesh

This issue is that while processing 3000 images with drone2map its been 3 days that the processing have started but still it is 50%. I tried using it in pix 4d but Pix4d hangs whenever I try more than 2000 images. I tried using Agiosoft but even that is too slow.My pc is also good enough with i7 processor, 64gb ram, 4tb harddisk, 3.5 ghz processor and 8 gb ram still I can't do.

Is there a way by which I could process 3000 drone images(fastly) to get my final output i.e. 2d mosaic(tiff) and 3d mesh.

[XanderBakker]

You can spin up a server in the cloud with for instance 48 cores and have a go at it. I heard about cases that processing on a desktop could take several days and running Drone2Map on a server in the cloud the images were processed in less than a hour. Having a server in the cloud for a day should not be very expensive. 

[AkhilChhibber1]

Are there some cloud server available for trial such that we could just try processing?????

[DanPatterson_Retired]

If you are processing in arcmap your memory is still limited to 4 Gb on a good day.  If you have background geoprocessing enabled, that will improve things .. but .. only if the tool you are using supports it.  

Processing in batches/chuncks may improve things, or do your work in ArcGIS Pro to take full advantage of memory in any circumstance.  Check the tools that you use and see if background geoprocessing is supported would be a first step, switch to pro if the tool(s) is/are available there, then process in chunks as the final option.

[XanderBakker]

Hi Dan Patterson , this refers to processing images captured by drone with Drone2Map (or other specific software). Drone2Map allows you to define the amount of cores you want to dedicate to the process. So now ArcMap or Pro involved in these steps. 

[AkhilChhibber1]

Can you please provide me some info regarding where could I buy cloud server for a day.

I mean that information would be of good help.

[DanPatterson_Retired]

ahhh .. I will esri would drop the ...for ArcMap from stuff, , at least we agree on 'chunks'

[XanderBakker]

Hi Akhil Chhibber ,

There might be trials available have a look at AWS or Azure or any other cloud server. Drone2Map is an application that can be installed at the server without any dependencies to other software of the ArcGIS platform.

Also see if perhaps you could process the data (perhaps local) with the rapid template and less accurate output just to validate the coverage of the images and quality in the report. Or process in chunks (with overlap).

[PhilipRoyal]

Hi Akhil -

One thing we have learned in Drone2Map processing is that using GPU accelerated processing helps speed the processing up considerably.  This will allow thousands of cores to process your workload.  We saw processing of around 1,000 Drone2Map images drop from 23 hours to 8 hours by using a couple of NVIDIA processors.  I am not sure how effective it will be processing 2,000-3,000 images (as mentioned by Xander above you may need to break up your areas into smaller parts) but it will make a difference.

[PrashanthMohan]

Hi,

We are also facing exactly similar situation, we are trying to process 1000+ images with Drone2Map.

As mentioned in the above thread, we have partitioned the whole data into 3 datasets and processed individually. The results were individually good but the issue was that the 3 output 3D meshes had distinct spatial shifts. So, we are unable to aggregate the three 3D meshes to visualize the whole area.

Having done this, we came to know that partitioning the large input data into small datasets won't serve the purpose. Thus, we are left with the only option of processing all the input images at single time.

So, kindly suggest on how to accomplish the task without segregation of input data.

Most importantly, even in the processed output 3D meshes, most of the trees are not extruded. While the small shrubs, houses are clearly extruded, we are wondering why trees are not extruded. Note that, we have sufficiently 60% forward overlap and 40% lateral overlap.

As we have been working on this for the last 3 weeks, we have tried all the possible combinations of processing parameters but could not get the trees extruded. Can it be a shortcoming with the back ground algorithms?