Access Restriction

Author Billy, Antoine ♦ Pouteau, Sébastien ♦ Chaumette, Serge ♦ Desbarats, P. ♦ Domenger, Jean-Philippe
Source Hyper Articles en Ligne (HAL)
Content type Text
File Format PDF
Language English
Subject Keyword info ♦ Computer Science [cs]/Image Processing [eess.IV]
Abstract In robotic mapping and navigation, of prime importance today with the trend for autonomous cars, simultaneous localization and mapping (SLAM) algorithms often use stereo vision to extract D information of the surrounding world. Whereas the number of creative methods for stereo-based SLAM is continuously increasing, the variety of datasets is relatively poor and the size of their contents quite small. Because today most of these technologies are embedded on on-board systems, the power consumption and real-time constraints turn to be key requirements. Our contribution is twofold: we propose an adaptive SLAM method that reduces the number of processed frames with minimum impact error, and we make available a synthetic exible stereo dataset with absolute ground truth, which allows to run new benchmarks for visual odometry challenges. This dataset is available online at Adaptive SLAM Algorithm Our algorithm smartly select the most usefull frames that are going to be processed by a SLAM algorithm. It focuses on strongly reducing the number of frames when the trajectory is mostly straight, and keeping a high frame rate during rotations. The gure below illustrates two outputs of the same SLAM algorithm with the same number of input frames with and without our adaptive frames selection. With our method (in black and red) the estimated trajectory ts perfectly the ground truth (in green) while a naive frame selection (in blue) strongly moves away. Real-time Dense D Reconstruction Pipeline Our adaptive SLAM method lets us build a real-time dense D reconstruction process on a small device (such as a Raspberry Pi) without the need of a huge computationnal cost. The gure below shows the reconstructed point cloud correctly tting the unused given GPS coordinates (in red). The Alastor Dataset Alastor is, to our knowledge, the rst synthetic stereo dataset for SLAM algorithms. The real advantages of synthetic dataset: Alastor allows you to switch to any vector car or UAV. Alastor comes with absolute ground truth. Alastor lets you adjust the frame rate you need. Alastor can simulate any lightning conditions. Results Evaluation of adaptive SLAM algorithm using to libvisoo: on the KITTI dataset (left) and on the Alastor dataset (right). Those diagrams show that we effciently reduce the number of processed frames without increasing the overall error. Examples of generated dense point clouds for each dataset: Conclusions We have presented an optimization of SLAM algorithms that reduces the number of processed frames without increasing the resulting error. Increasing performances in real-time scenarios, and applicable to well known datasets. We have shown that the results are perfectly suitable for real-time dense reconstruction. Additionally, we made available the rst synthetic stereo dataset for SLAM applications: Alastor. The experiments that we have conducted have shown that adaptive SLAM performs great on simulated datasets. We hope that our dataset will help design and evaluate innovative methods that will take advantage of the tunable parameters in order to improve the effectiveness of SLAM solving algorithms in the future. Main references A. Geiger, P. Lenz, and R. Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In Computer Vision and Pattern Recognition (CVPR), IEEE Conference on, pages-66. IEEE,. A. Howard. Real-time stereo visual odometry for autonomous ground vehicles.
Educational Use Research
Learning Resource Type Poster