Nachricht

Am Mittwoch, den 26. August 2015, hält Thomas Westfechtel um 16:30 Uhr im Raum P 1.4.17 einen Vortrag über seine Masterarbeit mit dem Titel:

3D Object Recognition by Combining 3D Feature Descriptors and Exploiting their Inter-Relations

Zusammenfassung:

The task of object recognition is quite a challenging topic of recent research. It is the most essential task for reacting to and interacting with the environment. Currently, the Tadokoro lab is developing an autonomous mobile robot, Quince, for condition-monitoring of an industrial environment. In the daily route of the robot, there are several stairs the robot has to climb. For the perception of the surrounding, the Qunice robot is equipped with one UTM-30LX LRF (Laser Range Finder). The LRF is used for the acquisition of a 3D point cloud. Within the point cloud data, stairs are to be detected. If stairs are detected, the robot should then change into a different mode for climbing stairs. For this process, the position of the stairs has to be localized accurately. A pre-known 3D model of the stairs can be assumed to aid with this task. However, the model may be incomplete or changes can happen to the stairs over time. In addition, some part of the stairs may be occluded. The algorithm should account for these hindrances. Besides the recognition of stairs, the object recognition algorithm could be expanded to other simple objects, like pillars, pipes or walls. These objects can be used to improve the localization of the robot, thereby allowing for a better navigation, which is particular important in narrow spaces.