EFFICIENT IDENTIFICATION OF FACES IN VIDEO STREAMS USING LOW-POWER MULTI-CORE DEVICES [chapter]

Donavan Prieur, Eric Granger, Yvon Savaria, Claude Thibeault
2015 Handbook of Pattern Recognition and Computer Vision  
The recognition of individuals based on their facial traits provides a powerful alternative to traditional methods for access control in many mobile and distributed applications. In these cases, the fuzzy ARTMAP neural network classifier is well adapted for fast and efficient matching of faces captured in video streams against the model of individuals enrolled to the access control system. In this paper, fuzzy ARTMAP networks co-jointly optimized for accuracy and efficiency are implemented and
more » ... ompared for one-against-many face matching using the Intel Core i3-530, the Intel Atom N270 and the Octasic Vocallo MGW processors. The performance of these implementations is studied from several standpoints including processing time, memory requirements, energy consumption and classification accuracy. Experimental results obtained using real-world video data show that implementing fuzzy ARTMAP networks optimized using multi-objective PSO on low-power parallel processors allows to significantly reduce energy consumption over traditional processor solutions while maintaining a high level of classification accuracy.
doi:10.1142/9789814656535_0023 fatcat:ncorik52mnblli4cko5cllhwya