Section: New Software and Platforms

PrimaCV

• Participants: Rémi Barraquand, Claudine Combe, Lukas Rummelhard, Amaury Negre, Sergi Pujades-Rocamora and James Crowley

• Contact: James Crowley

Functional Description

PrimaCV is a software library for detecting, observing and tracking faces and emotions using the cameras on mobile devices. The PrimaCV library uses a scale invariant pyramid to construct receptive field descriptors for images. These are used by a coarse to fine multiscale "scanning window" face detector constructed as a cascade classifier constructed using an highly optimised version of Ada Boost. Because the system uses coarse to fine search within a scale invariant pyramid it automatically adapts to the number of pixels and scale of the imager. The coarse-to-fine search algorithm has been shown to provide a dramatic gain in performance over classic scanning window detectors. The algorithm produces a probability of a face for each possible scale and position in the image. Local maximum in probability are fed to a Bayesian face tracker.

Normalized imagettes of tracked faces can be fed to procedures for estimating face orientation, recognising identity, estimating parameters of emotions.