Towards Hybrid Camera Sensor Simulation for Autonomous Vehicles
| Autor: | Dieter Balemans, Jens de Hoog, Siegfried Mercelis, Yves De Boeck, Peter Hellinckx, Ali Anwar |
|---|---|
| Rok vydání: | 2020 |
| Předmět: | |
| Zdroj: | Advances on P2P, Parallel, Grid, Cloud and Internet Computing ISBN: 9783030611040 3PGCIC Advances on P2P, Parallel, Grid, Cloud and Internet Computing: proceedings of the 15th International Conference on P2P, Parallel, Grid, Cloud and Internet Computing (3PGCIC-2020) : 7th International Workshop on Distributed Embedded Systems in conjunction with 15th 3PGCIC-2020 Conference, October 28-30, 2020, Yonago, Tottori, Japan |
| DOI: | 10.1007/978-3-030-61105-7_29 |
| Popis: | To accurately test and validate algorithms used in autonomous vehicles, numerous test vehicles and very large data sets are required, resulting in safety constraints and increased financial cost. For this reason, it is desired to train the algorithms at least partly in simulation. In this work we lay the focus on the camera sensor and propose a novel methodology for injecting the instance of simulated vehicles into the camera data of real vehicles. To get qualitative results and improve generalization capabilities, the simulated data must sufficiently correspond to real-world sensor data in order to prevent loss of performance when moving the model to a real environment after training. The realism of the output is evaluated by object detection systems and a realism score produced by a CNN. Results show the potential of this approach for improving hybrid simulators for the validation of autonomous vehicles. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|