Crate optical_flow_lk 

High-performance computer vision algorithms for real-time applications

Provides implementations of:

• Lucas-Kanade optical flow
• Shi-Tomasi feature detection
• Optimized image processing pipelines

Designed to be compatible with WebAssembly (Wasm).

Structs

Corner

A detected corner: its pixel position and its Shi-Tomasi score, the minimum eigenvalue of the mean 3x3 structure tensor.

DetectorCapacity

Upper bounds for the work a detector context will see, used to allocate every buffer up front with DetectorContext::with_capacity or DetectorContext::reserve.

DetectorContext

Reusable owner of every buffer the Shi-Tomasi detector touches: gradient planes, structure-tensor row buffers, the score plane, the candidate list, the spacing grid and the result vector.

FloatDetectorContext

Shi-Tomasi corner detector for f32 images in physical units, e.g. thermal frames in °C. The counterpart of DetectorContext, owning every buffer it touches.

FloatImageView

A borrowed, validated, row-major f32 image, e.g. a thermal frame in °C.

FloatTrackerContext

Pyramidal Lucas-Kanade tracker for f32 images in physical units, e.g. thermal frames in °C. The counterpart of TrackerContext, owning both frame pyramids and every buffer the tracking hot path touches.

Grid

Detection grid of good_features_to_track_grid and detect_grid: the image is split into cols x rows equal cells, and each cell keeps up to max_per_cell corners, counting existing points.

ScoreThreshold

Corner score threshold of the f32 detector.

TrackOptions

Options of a tracking call beyond TrackParams, for TrackerContext::track_with and TrackerContext::track_fb_with and their FloatTrackerContext counterparts. The default reproduces track and track_fb.

TrackParams

Lucas-Kanade parameters of a tracking call.

TrackResult

Per-point result of calc_optical_flow_ex.

TrackerCapacity

Upper bounds for the work a tracker context will see, used to allocate every buffer up front with with_capacity or reserve of TrackerContext and FloatTrackerContext.

TrackerContext

Reusable owner of every buffer the tracking hot path touches: both frame pyramids, the Lucas-Kanade scratch, the result vector and the forward-backward intermediates.

Enums

FloatImageError

Why FloatImageView::new rejected an image.

TrackKernel

Arithmetic of the 8-bit tracker’s window kernels, see TrackOptions::kernel.

TrackStatus

Why a feature point ended up where it did after tracking.

Constants

CONVERGENCE_EPSILON

Per-level convergence threshold, in pixels of the current pyramid level.

DEFAULT_FB_THRESHOLD

Default forward-backward round-trip threshold (pixels) for calc_optical_flow_fb.

DEFAULT_MIN_EIGEN_THRESHOLD

Default TrackParams::min_eigen_threshold of the 8-bit tracker.

Functions

build_pyramid

Builds a pyramid of images where each successive layer is half as large in width and height

build_pyramid_into

Builds the pyramid into an existing buffer, reusing each level’s storage when its dimensions are unchanged.

calc_optical_flow_ex

Compute optical flow using the pyramidal Lucas-Kanade method, returning a per-point TrackResult with status and photometric error.

calc_optical_flow_fb

Track points prev->next, then re-track the results next->prev, and flag any point whose round-trip lands further than fb_threshold pixels from where it started as TrackStatus::FbInconsistent.

good_features_to_track

Finds good features points using the Shi-Tomasi algorithm

good_features_to_track_grid

Finds good feature points with uniform frame coverage by detecting per grid cell, while respecting features that are already being tracked.

Type Aliases

FloatPyramidLevel

One level of an f32 image pyramid, as built by FloatTrackerContext.