This project is to construct an end-to-end pipeline with Renderscript, Allocations, and Java (as necessary), in order to provide real-time (30fps+) hardware-independent face detection and tracking on Android devices. This will involve writing RenderScript code to extract relevant visual features for Haar Cascades or other methodologies at multiple resolutions, then clustering/classifying image segments as representing facial regions or not.
Further adaptations of this pipeline may be discussed, based on the same underlying features.
I am able to provide greyscale facial training data in the order of 10k's images, as well as guidance on parameterization and engineering considerations.
Candidates should have demonstrated experience with computer vision, RenderScript projects (whether together or separately), clear communication, and clean coding/testing practices.
This will be an excellent portfolio project, as well as leading to continued engagement on similar and more sophisticated pipelines in future, depending on performance.