Visage Technologies

FaceTrack locates and follows up to 151 facial feature points on one or more faces, in images and video, from any standard camera or video file. You get 2D and 3D head pose, the fitted 3D face model, and coordinates for landmarks like the chin tip, lip corners, and irises. It starts tracking as soon as a visible face shows up, and it recovers after occlusions or when someone turns away.

The tracker reports eye closure, eye movement, a 3D gaze vector, and pupil coordinates. Pair it with age and gender estimation from the analysis module and you cover most of the attention metrics a study needs. With a short calibration it also outputs screen space gaze, which is the point on the monitor a person is looking at. That covers attention studies and control schemes that follow the eyes, and it pairs with the iris radius data for distance estimates.

FaceAnalysis estimates age to within about five years under decent lighting, classifies gender, and returns a probability distribution across happiness, sadness, anger, fear, surprise, disgust, and neutral. The vendor notes that accuracy depends on lighting and head pose, so slightly turned faces give shakier numbers. More complex emotion sets are available on request.

FaceRecognition turns a face into a numeric descriptor, then compares it against enrolled descriptors to score similarity. Because only mathematical templates get stored, images and names stay out of the system, and the templates can't be reversed into a photo. Optional liveness detection blocks a printed photo or a phone screen held up to the camera.

The same SDK runs on Windows, macOS, Linux distributions like Ubuntu and RedHat, Android, iOS, and HTML5, plus embedded targets including Raspberry Pi, Xilinx, and Ambarella. Most modules work fully offline, which matters for cars, factory floors, and anywhere a stable connection is a luxury. A Unity plugin covers game and AR projects.

FaceTrack and FaceRecognition don't store images, names, or other personal data unless a client builds that layer. The toolkit ships as separate modules, so you can license tracking alone, or add recognition and analysis when the product needs them. Configuration files expose speed, face size range, and search limits, which lets a team trade accuracy for frame rate on weaker hardware.