executable file 306 lines (220 sloc) 7.49 KB


Build Status Coverage

OpenCV bindings for Node.js. OpenCV is the defacto computer vision library - by interfacing with it natively in node, we get powerful real time vision in js.

People are using node-opencv to fly control quadrocoptors, detect faces from webcam images and annotate video streams. If you're using it for something cool, I'd love to hear about it!


You'll need OpenCV 2.3.1 or newer installed before installing node-opencv. Note that OpenCV 3.x is not yet fully supported.

Specific for Windows

  1. Download and install OpenCV (Be sure to use a 2.4 version) @ For these instructions we will assume OpenCV is put at C:\OpenCV, but you can adjust accordingly.

  2. If you haven't already, create a system variable called OPENCV_DIR and set it to C:\OpenCV\build\x64\vc12

    Make sure the "x64" part matches the version of NodeJS you are using.

    Also add the following to your system PATH ;%OPENCV_DIR%\bin

  3. Install Visual Studio 2013. Make sure to get the C++ components. You can use a different edition, just make sure OpenCV supports it, and you set the "vcxx" part of the variables above to match.

  4. Download peterbraden/node-opencv fork git clone

  5. run npm install

$ npm install opencv


Run the examples from the parent directory.

Face Detection

cv.readImage("./examples/files/mona.png", function(err, im){
  im.detectObject(cv.FACE_CASCADE, {}, function(err, faces){
    for (var i=0;i<faces.length; i++){
      var x = faces[i]
      im.ellipse(x.x + x.width/2, x.y + x.height/2, x.width/2, x.height/2);

API Documentation


The matrix is the most useful base data structure in OpenCV. Things like images are just matrices of pixels.


new Matrix(rows, cols)

Or if you're thinking of a Matrix as an image:

new Matrix(height, width)

Or you can use opencv to read in image files. Supported formats are in the OpenCV docs, but jpgs etc are supported.

cv.readImage(filename, function(err, mat){

cv.readImage(buffer, function(err, mat){

If you need to pipe data into an image, you can use an ImageDataStream:

var s = new cv.ImageDataStream()

s.on('load', function(matrix){


If however, you have a series of images, and you wish to stream them into a stream of Matrices, you can use an ImageStream. Thus:

var s = new cv.ImageStream()

s.on('data', function(matrix){


Note: Each 'data' event into the ImageStream should be a complete image buffer.

Accessing Data

var mat = new cv.Matrix.Eye(4,4); // Create identity matrix

mat.get(0,0) // 1

mat.row(0)  // [1,0,0,0]
mat.col(4)  // [0,0,0,1]


var buff = mat.toBuffer()

Image Processing

im.canny(5, 300)

Simple Drawing

im.ellipse(x, y)
im.line([x1,y1], [x2, y2])

Object Detection

There is a shortcut method for Viola-Jones Haar Cascade object detection. This can be used for face detection etc.

mat.detectObject(haar_cascade_xml, opts, function(err, matches){})

For convenience in face detection, cv.FACE_CASCADE is a cascade that can be used for frontal face detection.





Using Contours

findContours returns a Contours collection object, not a native array. This object provides functions for accessing, computing with, and altering the contours contained in it. See relevant source code and examples

var contours = im.findContours();

// Count of contours in the Contours object

// Count of corners(verticies) of contour `index`

// Access vertex data of contours
for(var c = 0; c < contours.size(); ++c) {
  console.log("Contour " + c);
  for(var i = 0; i < contours.cornerCount(c); ++i) {
    var point = contours.point(c, i);
    console.log("(" + point.x + "," + point.y + ")");

// Computations of contour `index`
contours.arcLength(index, isClosed);

// Destructively alter contour `index`
contours.approxPolyDP(index, epsilon, isClosed);
contours.convexHull(index, clockwise);

Face Recognization

It requires to train then predict. For acceptable result, the face should be cropped, grayscaled and aligned, I ignore this part so that we may focus on the api usage.

** Please ensure your OpenCV 3.2+ is configured with contrib. MacPorts user may port install opencv +contrib **

const fs = require('fs');
const path = require('path');
const cv = require('opencv');

function forEachFileInDir(dir, cb) {
  let f = fs.readdirSync(dir);
  f.forEach(function (fpath, index, array) {
    if (fpath != '.DS_Store')
     cb(path.join(dir, fpath));

let dataDir = "./_training";
function trainIt (fr) {
  // if model existe, load it
  if ( fs.existsSync('./trained.xml') ) {

  // else train a model
  let samples = [];
  forEachFileInDir(dataDir, (f)=>{
      cv.readImage(f, function (err, im) {
          // Assume all training photo are named as id_xxx.jpg
          let labelNumber = parseInt(path.basename(f).substring(3));
          samples.push([labelNumber, im]);

  if ( samples.length > 3 ) {
    // There are async and sync version of training method:
    // .train(info, cb)
    //     cb : standard Nan::Callback
    //     info : [[intLabel,matrixImage],...])
    // .trainSync(info)
  }else {
    console.log('Not enough images uploaded yet', cvImages)

function predictIt(fr, f){
  cv.readImage(f, function (err, im) {
    let result = fr.predictSync(im);
    console.log(`recognize result:(${f}) id=${} conf=${100.0-result.confidence}`);

//using defaults: .createLBPHFaceRecognizer(radius=1, neighbors=8, grid_x=8, grid_y=8, threshold=80)
const fr = new cv.FaceRecognizer();
forEachFileInDir('./_bench', (f) => predictIt(fr, f));


Using tape. Run with command:

npm test.

Code coverage

Using istanbul and lcov. Run with command:

make cover

Build version of opencv.node will be generated, and coverage files will be put in coverage/ directory. These files can be remvoved automatically by running make clean.

MIT License

The library is distributed under the MIT License - if for some reason that doesn't work for you please get in touch.