Video + Tracking
CSC3104 · Lab

Kalman prediction & matching

Two people walk towards each other, stop to talk behind a pillar, then carry on. Track them with motion alone (SORT) and with appearance features (DeepSORT), and catch the ID switch as it happens.

  1. 1 · Predict0/1
  2. 2 · Experiment0/2
  3. 3 · Explain0/1
  4. 4 · Code it0/1
  5. 5 · Challenge
  6. 0/4 tasks done

Predict first

Commit to a guess before you touch anything. Being wrong here is how it sticks.

An object at x = 120 px moves at 12 px/frame. Where does a constant-velocity Kalman filter predict it after 5 frames?

Experiment

Move the controls and watch the outputs. Goals tick off automatically when you reach them.

Two people crossing

Scrub or play to frame ~20, where they cross.

Explain it

Put what you saw into words. This is what the exam will ask you to do.

What does the Hungarian algorithm do in SORT, and why did DeepSORT add a re-identification network?

Code it

Write real code against real data. Python and SQL run right here in your browser and hidden tests check your work; the AI reviews it like an examiner.

Pythonexam

SORT's predict-and-match step

Write:

  • predict(box, velocity, steps=1) → the box moved by velocity × steps (constant-velocity Kalman prediction).
  • match(tracks, detections, min_iou=0.3) → (matches, unmatched_tracks, unmatched_dets) where matches is a list of (track_id, det_index). Predict each track one step, build the cost matrix 1 − IoU, solve it optimally with scipy.optimize.linear_sum_assignment (the Hungarian algorithm), and reject pairs with IoU below min_iou.
Ctrl/⌘ + Enter to run

Challenge

Fixed variations, a new problem every day, and unlimited AI challenges built from your current settings.

Constant-Velocity Kalman Step

A tracked pedestrian is at x = 100 px moving at a constant velocity of 12 px/frame. Where does the Kalman filter predict the box after 3 frames?