A new month means a new release! Today we would like to show off our surface tracking tech.
See how the system reliably captures features on the surface and does not let them go, despite camera movement, zoom, frame updates, as well as rotation of the drone around its axis.
This is an eye candy, pure Computer Vision using Optical Flow, i.e. without ML (neural networks), GPU and all that kind of computation heavy stuff. Source video is public from yesterday’s Andriy Tsaplienko channel post.
Reliable surface tracking can be used for:
- Motion detector where object detection fails (not enough features, far distance, unusual objects etc)
- To separate the real movement of objects in the conditions of a hyper-active drone operator
- For optical navigation or flight stabilization in GPS denied / EW affected conditions
- To estimate distance to the surface and other camera movement if telemetry is not available
- Helpful in a more complex pipeline for motion compensation and to reduce false positives (e.g. discard all false movements, perform object detection on real movements) – something we’ve been doing for over 1 year now in our full system implementations
Another advantage of this method is that it’s computationally light and can work on a CPU.
More examples:
Land rover FPV