Hello guys,
I’m doing a masters thesis on segmenting LiDAR point clouds. The main focus is on how to synthesize training data with which we can train models, that are usable on real world data.
I’ve seen this post AI in Unreal Engine: Learning Through Virtual Simulations - Unreal Engine, which indicates that work is already being done in this direction. The reason I’m posting is just that I want to connect with people that have already done research in this direction or a somewhat connected direction. Perhaps if we share our knowledge we could see mutual benefits.
I’m very interested on how I could create synthetic scenes that resemble real world conditions procedurally. I would like to scan these scenes with a simulated sensor and use this data to train a machine learning model.
I’ve seen threads where community members were asking about a tool that simulates LiDAR or other sensors in unreal engine, but I haven’t really seen anyone actually build the tool.
I’ve made a tool that simulates an airborne LiDAR sensor, and render the resulting point cloud every couple of frames using the UE4 PointCloud plugin. I also made a tool that allows you to select a subset of points within a point cloud and export it to txt format. Please let me know if there is any interest in these tools and I will share them.