Research And Implementation Of Sight Detection System Architecture On ASIC

Sight detection plays an important technology in human-computer interaction.According to the characteristics of each task in the system, the Adaboost facedetection algorithm and eye characterized extraction algorithm used in the sightdetection system are optimized and implemented on hardware or software in order tocomplete the demand for real-time implementation of the sight detection.The goal of this study is to implement real-time sight detection system hardwarearchitecture based on ASIC. Its working process is below. First, get the human facesize and position in the picture shot by camera using Adaboost algorithm. Second, getthe coarse position eye region in face area and complete the precise position of the eyeregion in the crude eye region. Third, find the position of the pupil and the innercorner of the eye. Forth, use gaze estimation model and position of the pupil and theinner corner of the eye to estimate the user’s sight direction and position.Fully considered the parallel of Adaboost algorithm and the hardwarecharacteristics, the scaling, detection mode and the classifiers calculation of thealgorithm are optimized and implemented on dedicated processing unit in order toachieve improved face detection speed, reduced resource consumption and highsystem performance. Meanwhile, because of the low computation and complex logicoperations, the eye feature point detection algorithms are implemented on software.And adjusted and optimized the memory use and computing, the algorithms can run inembedded core.Under the high detection rate and a low operating frequency (100MHz)conditions, the resource consumption of the system is relatively small. The system isverified on Altera’s Stratix IV and able to achieve12frames/sec for a resolution of640×480image.