PYTHON · NUMPY

Estimate Motion

Estimate how far the scene shifted between two frames — optical flow — the GPS-free motion cue behind position hold and VIO.

01Challenge

Two frames, one small unknown shift between them. Recover the shift in pixels without any external position sensor, accurately enough to feed a velocity estimate.

02Model

A drone indoors has no GPS, but its camera still moves over the world, and that motion is written in how the image shifts frame to frame. Estimate that shift and you have a velocity — the core of optical flow and visual-inertial odometry. The cleanest estimator: try candidate shifts, and pick the one where the two frames line up best (lowest squared difference over the overlap). That best-match shift, divided by the time step, is your GPS-denied velocity.

03Guided practice

Worked: synthesize a frame and a shifted copy. Faded: search integer shifts, scoring by overlap SSD. Independent: return the best shift and convert to velocity.

04Feedback

PASS WHEN Block matching recovers the exact inter-frame shift and converts it to a velocity.

If your run fails, check:

FAIL: wrong shift — the search range may not cover the true motion; widen it.
FAIL: off by a sign — the shifted-frame indexing has dx/dy swapped or negated.

05Retrieve & space

Why does this work without GPS? (Scene motion in the image is a direct cue to camera motion.)
What is the cost of a bigger search window? (Quadratically more comparisons per frame — the real-time limit.)
What sensor pairs with flow to fix scale and drift? (An IMU — that is visual-inertial odometry.)

06Mastery & project

Recovers inter-frame motion by block matching and converts it to a GPS-denied velocity (L3: estimate egomotion from images alone).

This velocity is a measurement stream for the Module-3 filter and the fallback when GPS drops in the field block.

← Find the Target Fuse the Sensors →

PYTHON · NUMPY

Estimate Motion

Find the inter-frame shift by block matching, then read off velocity.

You read

the arrays and values already in scope

You change

the code you write in each cell

Fixed

the dataset and the checks that grade you