Skip to content
WebARKitWebARKitWebARKit Docs

jsfeatNext

jsfeatNext is a TypeScript port of jsfeat, a JavaScript computer-vision library, made for WebARKit. It covers the natural-feature pipeline from end to end, in plain JavaScript with no WebAssembly:

FAST or YAPE detection → ORB descriptors → brute-force Hamming matching → GMS filtering → RANSAC homography → camera pose (R, t)

Its algorithms are checked against the original jsfeat with an automated test suite, every public class and method has TSDoc comments, and it comes as both a UMD and an ES module build.

Install

Section titled “Install”
Terminal window
npm install @webarkit/jsfeat-next

The package ships dist/ and the type declarations already built. Use the npm package for production installs: installing from a git reference builds it on install, and needs the dev dependencies.

Use

Section titled “Use”
import jsfeatNext from '@webarkit/jsfeat-next';


// The algorithm modules are singletons: call them directly, no `new`.
const src = new jsfeatNext.matrix_t(width, height, jsfeatNext.U8_t | jsfeatNext.C1_t);
jsfeatNext.imgproc.grayscale(rgbaPixelData, width, height, src);

In a plain page, load the UMD build and use the jsfeatNext global:

<script src="dist/jsfeatNext.js"></script>
<script>
  const m = new jsfeatNext.matrix_t(320, 240, jsfeatNext.U8_t | jsfeatNext.C1_t);
</script>

The modules are attached to the jsfeatNext namespace: imgproc, fast_corners, yape, yape06, orb, bfmatcher, gms, homography2d, affine2d, pose_estimator, optical_flow_lk, linalg, matmath and the data structures such as matrix_t and keypoint_t. pose_estimator is a constructor (new jsfeatNext.pose_estimator(K)), because it keeps the inverted camera intrinsics.

The repository has runnable camera demos for most modules (blur, Canny, FAST, YAPE, optical flow, ORB matching with a reference image, warps). The ES module demos must be served over HTTP.

Limits

Section titled “Limits”
  • Not every class of the original jsfeat is ported yet: the Haar-cascade and BBF object detectors are missing.
  • The transform module takes matrix_t arguments where the original used raw arrays.
  • Coming from 0.8 or older? The jsfeatNext.jsfeatNext double namespace and new jsfeatNext.imgproc() were removed in 0.9: see the migration guide.
  • A hosted API reference is not available yet; it is tracked in webarkit.github.io#49. Meanwhile npm run docs builds it locally.

See also

Section titled “See also”