Towards Modular Spatio-temporal Perception for Task-adapting Robots
Schlagworte: robotics, multi-cue vision, machine learning, DDC: 004 (Data processing, computer science, computer systems)
AbstractIn perception systems for object recognition, the advantage of multiple modalities, of combining approaches, and several views is emphasized, as they improve accuracy. However, there are great variances in the implementation, suggesting that there is no consensus yet on how to approach this problem. Nonetheless, we can identify some common features of the methods and propose a flexible system where existing and future approaches can be tested, compared and combined. We present a modular system in which perception routines can be easily added, and define the logic of making them work together based on the lessons learned from different experiments.