17–19 Mar 2020
European Space Research and Technology Centre (ESTEC)
Europe/Amsterdam timezone
NEW!!! SEFUW 2023!!! The 2020 edition was postponed. Presentations and abstracts are left here for reference. Please check the 2023 edition of SEFUW at https://indico.esa.int/event/439/

Computer Vision Acceleration on NG-MEDIUM and NG-LARGE FPGAs for Rovers

18 Mar 2020, 15:00
25m
Newton 1 and 2 (European Space Research and Technology Centre (ESTEC))

Newton 1 and 2

European Space Research and Technology Centre (ESTEC)

Keplerlaan 1 2201AZ Noordwijk ZH The Netherlands
FPGAs: High Performance FPGAs: High Performance

Speaker

Dr George Lentaris (National Technical University of Athens, Greece)

Description

The high-performance and reconfiguration capabilities offered by the new European FPGA technology can be leveraged in planetary exploration scenarios, where rover autonomy relies on multiple diverse and computationally intensive algorithms for Computer Vision. A number of HW/SW pipelines for rover localization and mapping were developed on Xilinx technology during past ESA activities (SPARTAN/SEXTANT/COMPASS). These pipelines include classical feature extraction such as Harris corner detection, Canny edge detection, SIFT description, SURF detection & description, FAST, BRIEF, they include feature matching via L1/L2/chi2/Hamming distances, as well as depth extraction from stereo images via Disparity-based and SpaceSweep algorithms. With a twofold purpose, these HW/SW pipelines are now being ported and optimized on the NG-MEDIUM and NG-LARGE FPGAs under new ESA activities (QUEENS1/QUEENS2 and NXARTAN). First, we perform a methodical assessment of the new European FPGA by using these VHDL kernels as high-performance benchmarks and testing all possible options/capabilities of the new SW & HW tools. Second, we optimize and deliver accelerators based on European technology, which offer considerable gains over the conventional LEON-FT approach. The current talk summarizes the results of the aforementioned activities with respect to the assessment of NanoXmap and its progress through successive SW versions, as well as the performance of the CV pipelines implemented on NG-LARGE.

Primary author

Dr George Lentaris (National Technical University of Athens, Greece)

Co-authors

Mr Vasileios Leon (National Technical University of Athens, Greece) Mr Ioannis Stamoulias (National Technical University of Athens, Greece) Mr Konstantinos Maragos (NTUA, Greece) Dimitrios Soudris (National Technical University of Athens)

Presentation materials

There are no materials yet.