search

OpenDroneMap / WebODM

User-friendly, commercial-grade software for processing aerial imagery. 🛩
https://www.opendronemap.org/webodm/
GNU Affero General Public License v3.0
2.89k stars 961 forks source link

Idea: Implement raster vision module/plugin for AI object detection #792

Open pierotofy opened 4 years ago

pierotofy commented 4 years ago

Could be useful for things like tree counting, building detection, cars counting, etc.

Would like to get some feedback on this from users as to whether this would be a useful feature, possible use cases aside from those listed, etc.

smathermather commented 4 years ago

It's quite possible in certain deciduous forests flown at high resolution , we can count trees in the forest and potentially extract crown size as well: image

We just laid out a project to play with this idea, but hadn't yet chosen a framework (though rastervision seems ideal). Integration into WebODM would be icing.

ChipwizBen commented 4 years ago

This would be very useful. I've seen livestock counted in the same way.

coreysnipes commented 4 years ago

This would be a very powerful addition. I was looking into solutions for this recently, as well. These may not all have suitable licensing or tech stack, but other candidates for consideration:

timotay07 commented 4 years ago

This would be amazing. I have thought of implementing computer vision to find GCP targets that are QR codes to simplify surveying workflow processes.

smathermather commented 4 years ago

@timotay07 -- maybe if you have the time, you could look into integrating this as an extension into WebODM: https://github.com/dronemapper-io/aruco-geobits

JamesLeonDufour commented 4 years ago

Definately a good ideas , counting shelter for estimating displacement population etc... https://www.arcgis.com/apps/Cascade/index.html?appid=5fdca0f47f1a46498002f39894fcd26f

pierotofy commented 2 years ago

I think this could be implemented, but requires developing our own smaller library for AI detection; rastervision is just very difficult to install (in terms of dependencies).

SabraHashemi commented 1 year ago

I was thinking about this right now, why this amazing software dose not have object detection, have you done anything already? I look forward to this feature

AffiTheCreator commented 1 year ago

So I agree with @SabraHashemi and so I started developing my own plugin for ODM to use my own or any AI model to perform basicly any type of detection/recognition ... I plan on developing this further and create a whole ecosystem around it but that can be discuss later :)

I'm currently at the stage where I have all the code for a first test and have created a simple interface using pieces of other plugins. I have also reached a stage where I can build the plugin successfully with ./webodm.sh start --dev but that is all that happens unfortunately, hence my little trip to this thread.

The only modification i did has to coreplugins, is there something else missing? Because the plugin does not show in dev tools in the browser under plugins like the others and nothing is being rendered.

I have also looked for documentation for this specific subject and couldn't find anything. You explain very well the framework and the infrastructure but there is a substantial hole when it comes to documentation for devs,in my humble opinion.

Anyways I appreciate the help and the nice work you guys have been doing... if you could clarify how to move forward. I didn't want to create a new issue since this one already exists.

SabraHashemi commented 1 year ago

@AffiTheCreator can you join the community of this project? there they will answer your quests, and I support your work and efforts

AffiTheCreator commented 1 year ago

@SabraHashemi I have created an account in the community forum. Are you aware of any current thread where we can discuss this or should i create a new one?

Also @smathermather would it be possible to share what you have on this subject to compare to what I have currently?

SabraHashemi commented 1 year ago

@SabraHashemi I have created an account in the community forum. Are you aware of any current thread where we can discuss this or should i create a new one?

Also @smathermather would it be possible to share what you have on this subject to compare to what I have currently?

We should create a thread on object detection plugin and also change detection(object based) is a good idea

Currently I’m working on live videos but its not for this app, after that i will join to contribute to object detection plugin

Create a thread and i will support and ask the developers to join and help

AffiTheCreator commented 1 year ago

@SabraHashemi I'll start the thread and share some of my ideas, if you could provide feedback on what you think is best I would appreciate.