Our deep-learning damage recognition solution can use all data types in the form of an image (out of cameras, laser scans, thermal cameras, aerial imaging, etc.). Either with pre-trained models, or models trained from your data, we can recognize various types of damages and automatize this time-consuming process that is still done manually. Applications include buildings, power poles, bridges, tunnels, roads or pipelines.
After digitizing damages and objects, we can combine them with information like geo-location, or timestamps so that we keep track of changes over time. We can also implement custom risk assessment and management procedures according to your case. Heavy deep-learning training runs on the cloud, but inference (using a trained model on your data) can also be done on premises, so we adapt to your use-case and keep costs low.
An example application of our solution is to automatically detect damages in tunnels for our client Amberg. Images taken by laser scanners are given to a deep network, specifically trained to recognize different types of damages (cracks, leakages etc.) or objects (pipes etc.).