The getUserMedia webcam access API in modern browsers offers exciting possibilities for "runs anywhere" Computer Vision, but with the limitation that performance is typically limited by the use of JavaScript.

Meanwhile, Vision algorithms are increasingly being implemented on the GPU with CUDA and OpenCL. This project takes advantage of the GPU within a browser environment by using WebGL shaders to implement a face detection algorithm in a fast, parallel way.

Since WebGL is primarily a graphics API, this approach has some rough edges, and could be considered more of a stop-gap measure until projects like WebCL reach maturity. But the viability of Vision on the web is definitely on the rise, and with it comes the need for Vision libraries, a web-based analogue to OpenCV.

This repository has somewhat ambitiously been titled WebCV, despite only doing face detection so far. It was my MEng project at Imperial, supervised by Andrew Davison.

As it grows, it would be nice to have backend-agnostic approach, giving an API that would not be tied to WebGL, and could take advantage of WebCL or other technologies if available. (Similarly to how THREE.js can fallback to a Canvas2D renderer)

Works best in Chrome (Firefox can be unstable)

On Windows, the ANGLE DirectX conversion makes things slow. To use native OpenGL, launch chrome with

chrome.exe --use-gl=desktop

Or in Firefox set webgl.prefer-native-gl=true in about:config

Basic detection demo

Head tracking