Description

This research is focused on the recognition and detection of dynamic objects using lidar data and image sequences with applications to mobile robotics. Analyzing scene dynamics is a challenging task because objects not only change appearance, but also become partially or completely occluded during motion. Moreover moving objects might remain still for large periods of time, making it difficult to classify them as dynamic or static without scene context. In this work we will investigate to what extent the use of dense range data, typically coming from lidar sensing devices, together with image sequences, can be used to improve segmentation, classification, recognition, and reconstruction tasks of dynamic objects and people in the scene. The fusion of lidar data and image sequences for recognition and segmentation of dynamic objects will be demonstrated through improved results for SLAM in highly dynamic scenes, with the added benefit of accurate 3D reconstruction of the dynamic objects.

The work is under the scope of the following projects:

• PAU: Percepción y acción ante incertidumbre (web)