Crate spatialrust_vision

Expand description

AI-ready CPU image processing and vision algorithms.

Algorithms are feature-gated by area. GPU implementations belong in an explicit backend and must not introduce hidden device transfers.

Structs§

AbsolutePose: Object-to-camera pose X_cam = R X_obj + t used by absolute orientation (PnP).
AdaptiveBackgroundModel: Stateful single-Gaussian grayscale background model.
AffineTransform: A source-to-destination 2D affine transform.
AreaResizeU8Plan: Reusable weighted-span plan for packed u8 area resize.
BackgroundModelOptions: Adaptive Gaussian background-model controls.
BackgroundUpdate: Output of one background-model update.
BilinearResizeU8Plan: Reusable source coordinates and fixed-point coefficients for u8 bilinear resize.
BinaryMask: Validated binary mask (0 background, 1 foreground).
BoundingBox2: Axis-aligned bounding box using half-open (min, max) coordinates.
CameraMatrix3: Calibrated pinhole intrinsic matrix and its analytic inverse.
CannyOptions: Validated Canny thresholds and gradient settings.
CannyResult: Edge map and numerical stages produced by Canny.
CannyWorkspace: Reusable caller-owned storage for the allocation-light Canny path.
ComponentStats: Per-component statistics.
ConfidenceMap: Dense confidence values constrained to [0, 1].
ConnectedComponents: Connected-component labeling output.
Contour: One closed polygonal contour on pixel-grid corner coordinates.
CornerSelectionOptions: Shared response thresholding and spatial suppression options.
DenseFlowOptions: Integer block-matching controls for dense grayscale optical flow.
DepthMap: Metric or relative dense depth map. Non-positive and non-finite values are invalid.
DescriptorBuffer: Row-major descriptor matrix with a fixed width and explicit scalar kind.
Detection: One scored detection used by class-aware post-processing.
DistanceTransformWorkspace: Reusable host scratch storage for the unit-spacing exact distance transform.
Essential3: A calibrated two-view epipolar constraint matrix.
ExposureFusionOptions: Well-exposedness fusion controls.
FastOptions: FAST detector configuration for the standard 9-of-16 radius-three circle.
FeatureMatch: One descriptor correspondence from a query feature to a train feature.
FeatureSet2: Keypoints paired one-to-one with descriptor rows.
FlowField: Dense (dx, dy) optical-flow field in pixel units.
Fundamental3: An uncalibrated two-view epipolar constraint matrix.
GaussianBlurU8Workspace: Reusable scratch storage and fixed-point kernel cache for gaussian_blur_u8_into.
GeometricEstimate: A geometric model paired with one inlier decision and residual per input row.
GridSelectionOptions: Controls deterministic spatial distribution of candidate keypoints.
HarrisOptions: Harris detector configuration.
Homography3: A pixel-to-pixel planar projective transform.
IntegralImage: Summed-area table with a zero top row and left column.
Kernel1D: Validated one-dimensional correlation kernel.
Kernel2D: Validated two-dimensional correlation kernel.
Keypoint2: One scale-space image keypoint in pixel coordinates.
LabelImage: Connected-component label image (0 is background).
LetterboxTransform: Mapping between a source image and its letterboxed output.
LucasKanadeOptions: Configuration for sparse Lucas–Kanade tracking.
MaskRle: Alternating zero/one run lengths, always beginning with a zero run.
MatchOptions: Filtering applied to brute-force descriptor correspondences.
MemoryVideoSource: Deterministic in-memory source used for adapter conformance and replay.
MonocularOdometryEstimate: Scale-ambiguous monocular odometry result.
MultiObjectTracker: Deterministic same-class greedy IoU multi-object tracker.
MultiObjectTrackerOptions: IoU tracker controls.
NearestResizeU8Plan: Reusable source-coordinate plan for packed u8 nearest-neighbor resize.
ObjectImageCorrespondence: One ordered object-to-image correspondence for absolute pose estimation.
ObjectTrack: One deterministic detection track.
OrbOptions: Multi-scale ORB detector and descriptor configuration.
Padding: Padding applied around an image.
Panorama: Pairwise panorama and its integer world-coordinate origin.
PanoramaOptions: Bounded pairwise panorama settings.
PerspectiveTransform: A source-to-destination projective 2D transform.
PointCorrespondence2: One ordered pixel correspondence from a source image to a target image.
PointMap: Dense per-pixel XYZ point map. Non-finite triples represent invalid points.
RectMorphologyWorkspace: Reusable scratch storage for packed grayscale rectangular morphology.
RelativePose: Target-camera pose expressed in the source-camera coordinate frame.
RelativePoseEstimate: Essential-matrix pose disambiguation and per-correspondence triangulation.
RgbdOdometryEstimate: Metric previous-camera to current-camera odometry result.
RgbdOdometryOptions: Depth filtering and PnP RANSAC controls for metric RGB-D odometry.
RobustEstimationOptions: Deterministic robust-model sampling and inlier classification settings.
RobustTrackOptions: Forward/backward consistency settings layered over pyramidal LK.
RobustTracks: One target coordinate, validity decision, and round-trip error per source point.
ScoredIndex: Index and possibly updated score returned by Soft-NMS.
ShiTomasiOptions: Shi–Tomasi minimum-eigenvalue detector configuration.
StereoBmOptions: Block-matching stereo options.
StereoRectifyMaps: Remap grids produced by stereo rectification for explicit warp::remap.
StereoRig: Calibrated two-camera stereo geometry.
StructuringElement: A validated binary neighborhood mask and anchor.
TrackedPoints: Result of sparse point tracking between two frames.
TriangulatedPoint: One two-view triangulation with cheirality and reprojection diagnostics.
VideoFrame: Timestamped owned video frame.

Enums§

AdaptiveThresholdMethod: Adaptive neighborhood statistic.
BorderMode: Out-of-bounds sampling behavior for CPU image operations.
Connectivity: Pixel connectivity used by binary-mask algorithms.
DescriptorKind: Scalar representation used by a descriptor matrix.
Interpolation: Sampling filter used by image resampling and geometric warps.
MorphologyOperation: Composite morphology operation.
MorphologyShape: Built-in structuring-element geometry.
OrbScoreType: Response used to rank ORB keypoints.
RleOrder: Linearization order used by run-length encoded masks.
SoftNmsMethod: Soft-NMS score-decay strategy.
ThresholdType: Point-wise threshold transformation.
TrackState: Track lifecycle state.
VisionError: Errors raised by image and vision algorithms.

Constants§

INVALID_DISPARITY: Invalid disparity sentinel written by stereo_block_match.

Traits§

PixelComponent: Scalar component supported by generic CPU image kernels.
VideoFrameSource: Pull-based video source implemented by optional codec/camera adapters.

Functions§

adaptive_threshold: Applies adaptive binary thresholding to an 8-bit grayscale image.
approximate_polygon: Simplifies a closed contour with Ramer–Douglas–Peucker approximation.
batched_nms: Class-aware greedy NMS over detection records.
bilateral_filter: Applies an edge-preserving bilateral filter.
box_blur: Applies a normalized rectangular box filter.
build_gaussian_pyramid: Builds a packed Gaussian pyramid including the input as level zero.
canny: Finds edges in a single-channel u8 image.
canny_into: Finds edges into caller-owned output using reusable CPU workspace.
canny_with_intermediates: Runs Canny and retains gradient, magnitude, and suppression stages.
clahe: Applies contrast-limited adaptive histogram equalization to grayscale u8.
connected_components: Labels foreground components and computes bounding boxes and centroids.
connected_components_u8: Labels connected non-zero pixels from packed u8 mask storage.
convolve2d: Applies true convolution by reversing the supplied kernel around its anchor.
crop: Copies a checked rectangular region into a packed image.
decode_rle: Decodes a run-length mask and validates its total length.
dense_flow_block_match: Computes a dense integer flow field with deterministic local block matching.
depth_map_to_point_cloud: Unprojects a depth map with a calibrated camera into an XYZ point cloud.
depth_tensor_to_depth_map: Converts a host depth tensor into a DepthMap.
detect_and_describe_orb: Detects oriented multi-scale keypoints and computes 256-bit rotated BRIEF descriptors.
detect_fast: Detects FAST-9/16 corners in scan order, matching OpenCV’s radius-three ROI.
detect_harris: Detects Harris corners and returns strongest-first keypoints.
detect_shi_tomasi: Detects Shi–Tomasi corners and returns strongest-first keypoints.
detection_tensor_to_detections: Decodes axis-aligned detections from an [N,6] tensor of x0,y0,x1,y1,score,class.
dilate: Dilates each channel independently.
dilate_rect_u8: Dilates a grayscale u8 image with a rectangular element in linear time.
dilate_rect_u8_into: Dilates into caller-owned packed output using reusable scratch storage.
disparity_to_depth: Converts horizontal disparity to metric depth with Z = f * B / d.
disparity_to_xyz: Reprojects disparity into left-camera XYZ using rectified intrinsics.
distance_transform_edt: Computes the exact Euclidean distance from every foreground pixel to the nearest background pixel using unit pixel spacing.
distance_transform_edt_into: Computes the exact unit-spacing Euclidean distance transform into a reusable slice.
distance_transform_edt_u8_into: Computes a unit-spacing exact distance transform from a packed u8 mask.
distance_transform_edt_with_spacing: Computes the exact Euclidean distance transform with physical pixel spacing.
encode_rle: Encodes a binary mask into alternating runs.
equalize_histogram: Equalizes an 8-bit grayscale histogram.
erode: Erodes each channel independently.
erode_rect_u8: Erodes a grayscale u8 image with a rectangular element in linear time.
erode_rect_u8_into: Erodes into caller-owned packed output using reusable scratch storage.
estimate_and_stitch_panorama: Estimates a source-to-target homography and stitches the pair.
estimate_essential: Estimates an essential matrix after normalizing pixels with both cameras.
estimate_essential_ransac: Robustly estimates an essential matrix in normalized-camera coordinates.
estimate_fundamental: Estimates a fundamental matrix from at least eight correspondences.
estimate_fundamental_ransac: Robustly estimates a fundamental matrix using deterministic eight-point RANSAC.
estimate_homography: Estimates a homography from at least four correspondences using normalized DLT.
estimate_homography_ransac: Robustly estimates a homography using deterministic four-point RANSAC.
estimate_monocular_odometry: Estimates calibrated monocular motion with essential RANSAC and cheirality.
estimate_rgbd_odometry: Estimates metric motion from source depth and source-to-target pixel tracks.
filter2d: Applies OpenCV-style correlation and converts the result back to the input dtype.
filter2d_f32: Applies OpenCV-style correlation and preserves signed/fractional results as f32.
find_contours: Extracts oriented boundary loops, including hole contours.
fuse_exposures: Fuses aligned RGB exposures with normalized per-pixel well-exposedness.
gaussian_blur: Applies a separable Gaussian blur with explicit odd sizes and standard deviations.
gaussian_blur_u8: Applies a specialized 3×3, 5×5, or 7×7 Gaussian blur to interleaved u8 input.
gaussian_blur_u8_into: Applies the specialized u8 Gaussian blur into caller-owned packed output and workspace.
gray_to_rgb: Replicates a gray channel into RGB.
gray_world_white_balance: Applies deterministic gray-world white balance to an RGB image.
harris_response: Computes an unnormalized Harris structure-tensor response image.
histogram: Computes a configurable one-channel histogram, optionally through a mask.
histogram_u8: Computes the exact 256-bin histogram of an 8-bit grayscale image.
integral_image: Computes a summed-area table for one selected channel.
laplacian: Computes a Laplacian response as signed f32 values.
letterbox: Aspect-preserving resize followed by centered constant padding.
match_descriptors: Matches each query descriptor to its nearest train descriptor.
median_blur: Applies a per-channel median filter with an odd square aperture.
morphology_ex: Applies a composite morphology operation.
morphology_rect_u8: Applies composite grayscale u8 morphology using the rectangular fast path.
morphology_rect_u8_into: Applies composite rectangular morphology into caller-owned packed output.
nms: Greedy non-maximum suppression. Returned indices are score-descending.
normalize: Converts packed scalar values to f32, then applies value * scale, mean subtraction, and per-channel standard-deviation normalization.
normalize_into: Normalizes interleaved pixels into caller-owned f32 storage.
otsu_threshold_u8: Selects an Otsu threshold for an 8-bit grayscale image and applies it.
otsu_threshold_u16: Selects an Otsu threshold for a 16-bit grayscale image and applies it.
pack_chw: Packs and normalizes an interleaved image into planar CHW storage.
pack_chw_into: Packs and normalizes interleaved pixels into a reusable planar CHW slice.
pad: Pads an image with a constant pixel value.
planar_f32_to_nchw: Packs a planar CHW f32 image into contiguous NCHW [1,C,H,W].
point_map_to_point_cloud: Flattens valid point-map pixels into an XYZ cloud, optionally filtering by confidence.
project_object_point: Projects an object point with an absolute pose into pixel coordinates.
pyr_down: Reduces each dimension by approximately two using the canonical 5×5 kernel.
pyr_up: Doubles each dimension using zero insertion and the canonical 5×5 kernel.
recover_relative_pose: Recovers the essential-matrix pose with the most positive-depth points.
remap: Samples an image at absolute coordinates supplied by two single-channel maps.
resize: Resizes an interleaved image while preserving semantic metadata.
resize_into: Resizes into caller-owned storage without allocating.
resize_pack_chw: Fuses bilinear RGB resize, normalization, and CHW packing without an intermediate image.
resize_pack_chw_into: Fuses bilinear RGB resize, normalization, and CHW packing into caller-owned storage.
resize_rgb_to_gray: Fuses bilinear RGB resize and BT.601 grayscale conversion without an intermediate RGB image.
resize_rgb_to_gray_into: Fuses bilinear RGB resize and BT.601 grayscale conversion into caller-owned storage.
rgb_to_gray: Converts RGB u8 pixels to BT.601 luma using fixed-point coefficients.
rgb_to_gray_into: Converts RGB u8 pixels into caller-owned gray storage without allocating.
rgb_to_hsv: Converts RGB u8 to OpenCV-style HSV (H in 0..=179, S/V in 0..=255).
rgb_u8_to_nchw_f32: Letterboxes an RGB u8 image and packs it as contiguous NCHW f32 [1,3,H,W].
scharr: Computes a 3×3 Scharr first derivative as signed f32 values.
score_tensor_to_binary_mask: Converts a host score tensor into a binary mask via threshold.
select_keypoints_grid: Retains the strongest keypoints in each image cell.
separable_filter: Applies horizontal and vertical kernels using an f64 intermediate buffer.
separable_filter_f32: Applies separable kernels and preserves signed/fractional output as f32.
shi_tomasi_response: Computes an unnormalized Shi–Tomasi minimum-eigenvalue response image.
sobel: Computes a Sobel derivative as signed f32 values.
sobel_3x3_u8: Computes a first-order 3x3 Sobel derivative directly into signed f32 output.
sobel_3x3_u8_into: Computes a first-order 3x3 Sobel derivative into caller-owned packed output.
sobel_abs_3x3_u8: Computes an exact absolute first-order 3x3 Sobel response as saturated u8.
sobel_abs_3x3_u8_into: Computes an exact absolute first-order 3x3 Sobel response into packed output.
sobel_l1_magnitude_u8: Computes exact 3×3 Sobel L1 magnitude (|Gx| + |Gy|) for grayscale u8.
sobel_l1_magnitude_u8_into: Computes exact fused 3×3 Sobel L1 magnitude into caller-owned packed output.
sobel_threshold_3x3_u8: Computes a binary edge mask from absolute 3x3 Sobel response.
sobel_threshold_3x3_u8_into: Computes a binary absolute-Sobel edge mask into caller-owned packed output.
soft_nms: Soft non-maximum suppression with deterministic score ordering.
solve_pnp: Estimates an object-to-camera pose from at least four correspondences.
solve_pnp_ransac: Robust PnP using deterministic six-point RANSAC and inlier refinement.
spatial_gradient_u8: Computes exact 3×3 Sobel X/Y gradients together for grayscale u8 input.
spatial_gradient_u8_into: Computes exact paired 3×3 Sobel gradients into caller-owned packed output.
stereo_block_match: Dense SAD block matching on already-rectified grayscale stereo images.
stereo_rectify: Builds left/right absolute remap grids that make epipolar lines horizontal.
stitch_panorama_pair: Warps source into target coordinates and feather-blends their overlap.
swap_red_blue: Swaps the red and blue channels of a three-channel image.
threshold: Applies a fixed threshold independently to every component.
track_points_forward_backward: Tracks points in both directions and rejects inconsistent round trips.
track_points_lucas_kanade: Tracks sparse points from previous to next with pyramidal Lucas–Kanade.
triangulate_correspondence: Triangulates one calibrated correspondence for a known relative pose.
warp_affine: Warps an image with a source-to-destination affine transform.
warp_perspective: Warps an image with a source-to-destination projective transform.

Type Aliases§

VisionResult: Result type for image and vision operations.

Crate spatialrust_vision

Crate spatialrust_vision Copy item path

Structs§

Enums§

Constants§

Traits§

Functions§

Type Aliases§

Crate spatialrust_vision