Haar cascades in a browser tab: object detection without the server round-trip

What it does

js-objectdetect runs Viola-Jones object detection entirely in the browser using HTML5 getUserMedia. It detects faces, eyes, mouths, and upper bodies from webcam streams or static images without sending frames to a server. The API wraps a canvas element and returns bounding boxes you can draw over the video feed.

The interesting bit

This is a hand-port of OpenCV’s C++ Haar cascade classifier to JavaScript — not a wrapper around WebAssembly or a cloud API. The README notes it “reimplements” the algorithm, which in 2013 meant wrestling with typed arrays and avoiding the still-nascent WebAssembly path entirely. That choice aged strangely: fast for its era, now likely outpaced by TensorFlow.js or native MediaPipe, but with zero dependencies and no build step required.

Key highlights

• Ships with pre-trained OpenCV cascades (face, eye, mouth, upper body)
• Works on static images, video elements, or live webcam streams
• No server required; frames never leave the browser
• ~1200 stars suggests it solved a real need in the pre-WebAssembly era
• Single-file include: <script src="js/objectdetect.js"></script> and go

Caveats

• Repository appears unmaintained; last meaningful activity looks to be 2013–2014 based on the README’s browser compatibility notes (“works in all modern browsers”)
• No mention of performance benchmarks or frame rates; “real-time” is claimed but not quantified
• Haar cascades are generally less accurate than modern deep-learning approaches; the README doesn’t compare accuracy or discuss false positive rates

Verdict

Grab this if you need a zero-dependency, no-build object detection snippet for a quick prototype or an air-gapped environment. Skip it if you want state-of-the-art accuracy or a maintained codebase — MediaPipe and TensorFlow.js have eaten this lunch, and they brought GPU acceleration.