Abstract
We present a reconfigurable architecture model for rotation invariant multi-view face detection based on a novel two-stage boosting method. A tree-structured detector hierarchy is designed to organize multiple detector nodes identifying pose ranges of faces. We propose a boosting algorithm for training the detector nodes. The strong classifier in each detector node is composed of multiple novelly designed two-stage weak classifiers. With a shared output space of multicomponents vector, each detector node deals with the multidimensional binary classification problems. The design of the hardware architecture which fully exploits the spatial and temporal parallelism is introduced in detail. We also study the reconfiguration of the architecture for finding an appropriate tradeoff among the hardware implementation cost, the detection accuracy, and speed. Experiments on FPGA show that high accuracy and marvelous speed are achieved compared with previous related works. The execution time speedups range from 14.68 to 20.86 for images with size of 160×120 up to 800×600 when our FPGA design (98 MHz) is compared with software solution on PC (Pentium 4 2.8 GHz).
Highlights
Great advances have been achieved on face detection research [1], which has already been widely applied in many real-world applications, such as biometrics, visual surveillance, human-computer interaction, to name a few
The break through happened in 2001 when Viola and Jones [4] developed their Boosted Cascade Framework whose remarkable performance owes to the fast speed of Haar-like feature calculation based on the integral image, the high accuracy of boosted strong classifiers, and the asymmetric decision making of the cascade structure
Frontal face images at 384 × 288 resolutions were reported to be fairly reliably detected at 15 frames per second by using PC with PIII 700 MHz CPU
Summary
Great advances have been achieved on face detection research [1], which has already been widely applied in many real-world applications, such as biometrics, visual surveillance, human-computer interaction, to name a few. In the past few years, many derivatives of Viola’s work have been proposed for rotation invariant frontal face detection and MVFD These derivatives can be categorized into four aspects: the detector structure, designing of strong classifiers, training of weak classifiers, and selecting of features. Since different applications have different demands on the accuracy, speed, and resources cost, it is necessary to research on the reconfigurable hardware architecture for RIMVFD. The main contributions of this paper are (i) RIMVFD with all ± 90◦ ROP and 360◦ RIP pose changes is achieved by using tree-structured detector hierarchy and fine-classified boosting method;. (iii) by dynamically reconfiguring the hardware architecture, the tradeoff among the hardware implementation cost, the detection accuracy, and speed is well tuned, so that the proposed design can meet the demands of different applications.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
