How to Speed-Up an Embedded Face Detector?

A Face detector based on multiple cascaded classifiers runs fast, however, the algorithm must be further optimized for real time applications because an embedded system has usually a fixed point operation core and low speed memory access. Some strategies of speeding-up a face detector are proposed in this paper, including usages of scaling image and feature respectively, narrow of searching range and optimization of software design, and so on. Experimental results show that these strategies can greatly speed-up the detector in the embedded ARM system.