Here’s an interview that I recently did with IBM DeveloperWorks TV at the recent World of Watson conference. In it I discuss a project I’ve been working on that analyzes drone imagery to perform automatic damage detection using the Watson Visual Recognition, and generates 3D models from the drone images using photogrammetry processes. The best part – the entire thing runs in the cloud on IBM Bluemix.
It leverages the IBM Watson Visual Recognition service with custom classifiers to detect the presence of hail damage on shingled roofs, Cloudant for metadata/record storage, the IBM Cloud Object Storage cross-region S3 API for massively scalable & distributed image/model/asset storage, and Bare Metal servers for high performance computing.
Bare Metal servers are dedicated machines in the cloud: not shared, and not virtualized. I’ve got mine setup as a linux server with 24 cores (48 threads), 64 Gigs of RAM, a SSD RAID array, multiple GPUs, etc… and it improved my photogrammetry rendering from hours on my laptop down to merely 10 minutes (in my opinion the best part).
I’ve done all of my testing with DJI Phantom and DJI Inspire aircraft, but really, it could work with any images, from any camera that has embedded GPS information.
Check out the video to see it in action.
