Computer Vision and Machine Learning
Semantic understanding of indoor environments
A 3D point cloud is a most complete 3D representation containing points with associated coordinates, color and other properties. Point cloud processing and semantic understanding are very active research topics
Room segmentation allows to automatically partition huge point cloud data containing millions of points into semantically meaningful parts, like buildings and rooms. To solve this problem, one typically uses machine learning.
Object segmentation and matching allows to automatically recognize the objects within indoor environments. To solve this challenging problem, deep learning is commonly used. The recognized objects can be further used for applications, such as architecture, robot navigation and improved building management.
A current project deals with the autonomous inspection of airplanes with a drone. A variety of sensors is exploited to achieve this task, such as LiDAR, camera, RGB-D and IMU. The inspection data is sent in real-time to a ground station for further analysis and automatic damage classification using machine learning techniques. Find out more here.