As part of the energy and heat transition, we are developing methods for digitalizing the inventory of heating and photovoltaic systems so that installers can quickly record the inventory digitally, allowing the automated creation of individual offers for system optimization. We use commercially available devices such as smartphones to ensure that the algorithms are widely applicable. Our process is based on current deep learning models from the computer vision section. They were initially developed to record heating systems, but can also be transferred to new issues such as the recording of PV roof-top systems or PV power plants.
The desired image objects in a video sequence are identified and classified with pixel accuracy using AI-based segmentation models. To improve the evaluation, type labels are also identified and the labeling is interpreted using language models. In the 3D scene analysis, the identified components are recorded in their spatial reference, for which we use the integrated LiDAR (Light Detection and Ranging) technology of the mobile devices. The use of commercially available devices allows a wide range of applications for the algorithms.
Our video annotation tools, which enable efficient and user-friendly labeling of the entire recording sequence with just a few clicks, are ideal for training the algorithms.