SRV-1 Blackfin vision-based horizon finder function - version 2

This is an update to the simple horizon finding algorithm posted earlier for the Surveyor SRV-1 Blackfin camera firmware (software only on the Blackfin processor - no other sensors besides the camera). As before, it uses a built-in edge detection function, but we're now computing an actual slope and intercept for the detected horizon via a least-squares line fit, and then some filtering is added to smooth the resulting data and introduce a small time lag. This is an open source project ... for more information, visit http://www.surveyor.com or http://code.google.com/p/surveyor-srv1-firmware/

Category:

Science & Technology

Tags:

• autopilot
• robotics
• srv-1
• blackfin
• robot
• navigation
• vision

[motters2001]

Looks like there's some amount of lag. Is this due to filtering?

[surveyorcorp]

Exactly. The horizon function maintains a buffer of the past 'n' slope & intercept samples (n = [1, 10]), and the display is therefore based on a moving average. For the video, n = 5.

By the way, you suggested on the earlier video using RANSAC instead of least-squares fit for computing slope, and I am inclined to give this a try, as it will probably improve the handling of outlier data such as the tree that shows up in the middle of the video.

[frank26080115]

this is purely visual and not assisted by accelerometers or gyros? cool

[surveyorcorp]

yes - the horizon is derived just from onboard processing of frames captured from the Blackfin processor's attached camera - there are no other sensors.

[motters2001]

Yes, RANSAC usually gives better results when trying to fit a line to noisy data.