Motion Tracking & Optical Flow

Iris includes algorithms to isolate moving objects, track dynamic structures, and evaluate pixel motion vectors.

Background Subtractor

A pipeline that models stationary scenes and isolates transient moving foreground elements by computing running averages and absolute differences.

// Create background subtractor (learning_rate = 0.1, threshold = 0.05)
let mut subtractor = BackgroundSubtractor::new(0.1, 0.05);

// Process frame sequence
let mask = subtractor.apply(&frame)?; // grayscale mask highlighting motion

Object Tracking (MOSSE)

Tracks a specified bounding box across subsequent video frames using the MOSSE correlation filter.

let mut tracker = Tracker::new(TrackerType::MOSSE);

// Initialize tracker on first frame
let init_box = Rect::new(50, 50, 100, 100);
tracker.init(&frame1, init_box)?;

// Update on subsequent frames
let updated_box = tracker.update(&frame2)?;

Supported tracker types: TrackerType::KCF, TrackerType::CSRT, TrackerType::MOSSE.

Optical Flow

Optical flow tracks movements of pixels between two images.

Dense Flow (Farneback)

Calculates flow vectors (dx, dy) for all pixels, returning a flow tensor of shape [2, H, W].

let flow_tensor = OpticalFlow::calc_dense_farneback(&frame1, &frame2)?;

Sparse Flow (Lucas-Kanade)

Tracks specific keypoints across frames using spatial image pyramids.

let prev_pts = vec![Point::new(10.0, 10.0)];
let (next_pts, status) = OpticalFlow::calc_sparse_pyr_lk(&frame1, &frame2, &prev_pts)?;