The mainstream technique to recognize faces from an arbitrary image is based on template learning and matching. This is a tedious, error-prone and inaccurate way of “learning” software how to look for a specific face in a picture. It requires a lot of input and manual labor, and can be sensitive to the training set used.
Current state of the art enables neural networks to be used, enabling the software to learn by itself what a certain face must look like. The software is fed with matching and non-matching pictures of a certain face, the software then trains itself to recognize a similar face in unknown pictures.
In this assignment, the student will design, implement and test a deep neural network to recognize faces from a video stream. This recognition algorithm is then used in a real application to welcome regular inhabitants and guests in our building. Specific attention will be paid to learning with a sparse data set, since the ultimate goal would be to recognize a guest based on one (1) image only.
An experienced tech lead will supervise, coach and support this Master level assignment. The result of this assignment will be included in an IoT / Big Data application that is currently being developed at Sioux.
In all of our assignments, your own contributions are highly valued so this assignment description may be altered to suit your own expertise or ideas.
Sioux is an innovative technology partner that supports leading high tech companies in the development and manufacturing of their products. With 600 dedicated engineers we support or act as the Research and Development department of our customers. With our excellent productivity, we help to shorten the development time and to create a sustainable and competitive advantage at a better price/performance ratio.
Sioux strives to build the best quality software in its customer domains. More and more, complex systems require sophisticated data analysis for maintenance prediction or data mining. Neural networks play an important role in these analytics.