Object Detection/Recognition in OpenDroneMap
Related Github Issue: Idea: Implement raster vision module/plugin for AI object detection · Issue #792 · OpenDroneMap/WebODM · GitHub
The objective of this thread is to start a discussion about how to build this new plugin as mention in the issue.
There are couple solutions we can use:
- Add a feature to ODM engine to extract results from the images uploaded
- Add a new plugin to webODM that would either use an orthophoto tile or the original picture as input to the model. It then should be able to generate a geoJSON file with all the locations of the results. The geoJSON can be used with the layer function similar to the location of each photo as a way of showing the final results.
Such plugin or new feature would need a some parameters: defining the framework (tensorflow , keras ,onx , etc) and all their parameters, defining the model to use and his parameters, among some others to be defined.
I have started some work, and I’m using approach number 2 for now. It seemed the easiest way to implement a proof of work and to understand how the software works.
I´m open to suggestions and more ideas.
Happy Coding