基于Mega2560的人脸检测与定位跟踪  

Face detection and location tracking based on Mega2560

作　　者：李东洁[1] 何岸花 

机构地区：[1]哈尔滨理工大学自动化学院,黑龙江哈尔滨150080

出　　处：《计算机工程与科学》2016年第10期2101-2108,共8页Computer Engineering & Science

基　　金：黑龙江省高校青年学术骨干支持计划(1254G023);哈尔滨理工大学青年拔尖创新人才培养计划(2013)

摘　　要：为了解决存在外界干扰的人脸的检测和准确定位与跟踪问题,避免过于依赖软技术造成的算法复杂性,采用软硬件结合的方式,基于Mega2560单片机、VS2008开发环境与OpenCV函数库,实现了人脸检测与准确定位与跟踪。首先,利用Haar-like特征法寻找出图像中的人脸特征矩形,通过AdaBoost算法生成级联分类器,实现人脸检测;然后,将图像显示窗口中心选为坐标原点,计算人脸位置,用线性变换的方法获取人脸位置偏差,进而转换成角度偏差发送给下位机Mega2560;最后,单片机根据接收到的角度偏差信息,控制摄像头使人脸始终在坐标原点位置,实现人脸的准确定位与跟踪。通过实验表明系统具有较好的稳定性和准确性。In order to solve the problem of external interference on human face detection and accurate tracking and avoid relying too heavily on soft technology which can increase algorithm complexity, we propose a method for face detection and accurate location tracking based on Mega2560 microcomputer （MCU） ,VS2008 development environment and OpenCV function libraries. The proposal combines software with hardware. To realize face detection,we first use Haar-like features method to find out the facial feature rectangule in the image and generate a cascade classifier via the AdaBoost algorithm. Then we select the image display window as the origin of the coordinate and calculate the location of face. Using the linear transformation method we can acquire face location deviation, which is converted into angular deviation and sent to the lower machine Mega2560. Finally we employ the MCU to control the camera according to the received angular deviation information, which guarantees the face staying at the origin position all the time,thus acomplishing face location and accurate tracking. Experimental results show that the proposed system has good stability and accuracy.

关 键 词：人脸检测 Haar-like特征法 ADABOOST算法 定位与跟踪 

分 类 号：TP393.4[自动化与计算机技术—计算机应用技术]