Research And Implementation Of Human Face Detection Based On Embedded System

Face detection technology can effectively applied in many fields such as video security monitoring. The traditional face detection system often uses PC as a development platform,such a system is very flawed on the portability and development costs. Embedded technology is becoming mature now, and the embedded systems is small, portable,low-cost.So embedded face detection system will bring greater competitive advantage.Based on the study of the foundation of embedded technology and Adaboost face detection algorithm, the Functionality and performance requirements analysis of face detection system is finished. And a system development platform is built, the ARM11 processor and USB camera are used as the hardware platform,and the Linux operating system,OpenCV and Qt library are used as software development platform.Then, writing the application,using V4L2 interface to control camera to capture live video images,using Adaboost algorithm to do a face detection for each frame of the video.The detection result is output to the LCD screen. Finally, This paper implements a embedded face detection system,and improves the function and performance of the system as follows.To overcome the shortcomings that the traditional face detection system has a single function and occupies a large space for storing data. This paper implements the monitoring alarm and video frame selective memory function. While the system detects the presence of a face in the video, it trigger an alarm buzzer immediately. At the same time, the system stores the detected face of the video frame image in the SD card. And use the current system time to name the stored image.If there is no face in the video,the video frame will not be stored. Thus saving the system memory space,and it is very easy to search results.To solve the problem of poor real-time characteristic when systems running on embedded ARM platform, and improve the face detection speed of the system, four aspects, i.e. Set suitable face detection parameters, cut out the face classifier, image scaling, and color regionsettings are considered to solve the problem. The experimental result shows that the average detection time reduced from 2260 ms to 80 ms, and the detection rate is 92.3%, which meet the system requirements.