Large-Scale 3D Scene Reconstruction


The ability to produce accurate maps is known to be of key importance in most autonomous vehicles tasks, but it can also be useful to increase a person’s awareness of its surrounding. In this project we focus on the real-time generation of 3D surface models based on raw sensor information, and the extraction of meaningful features that can be used in a variety of tasks, such as user localization, object detection and obstacle avoidance. The use of state-of-the-art machine learning techniques ensures a continuous improving model that is able to reason over partial occlusions, sensor failure and dynamic motion to produce consistent maps even after long periods.