Home

Awesome

Open Visual Inertial Odometry (OVIO) Core

Current status

Work in progress. PCB have arrived for v1.0 of the board and bring-up has started.

Issues with v1.0 hardware (fixed in v1.1)

Images from the assembly and results here: https://twitter.com/korken89/status/1341126909485342721

Motivation

Aim: A "low cost" Visual Inertial Odometry accelerator.

A project to try out different vision processing pipelines, for Visual Inertial Odometry (VIO), on the ECP5 FPGA. As far as I know, the open hardware landscape of VIO cameras is slim to non-existent. Especially for systems which perform a majority of the pre-processing in the camera.

Things such as image convolutions, FAST corners, multi-scale images, non-maximum suppression, image gradients, etc, are extremely well suited for FPGA implementation.

Why do it on a PC? Let it run the actual Kalman Filter / MHE / nonlinear optimizer, which the FPGA is not well suited for.

alt text

alt text

Specs

Do you want one?

Previous projects has been made on Xilinx FPGAs, however the high cost of their FPGAs never allowed for trying to distribute a camera. Now it is expected that the hardware cost for this system should be sub 100 Euros, even with the largest ECP5 FPGA.

If there is interest for acquiring a camera, contact me (via any medium) and we'll figure something out.

Lenses and lens holder

Contributors

License

This hardware is open source.

TODO: Find a proper license