| In recent years, face detection has been a basic and important research area in the field of pattern recognition and computer vision. With the development of embedded technology and intelligent device, the face detection system should meet the requirement of mobile and outdoor application field. We implement the multi-view face detection system on three platforms: DSP, PowerPC and Microblaze. And we compare the performance of three systems. It gives the reference of the embedded face detection system design. The content and contribution of the thesis are listed as follow:1. Algorithm Research: After the algorithm analysis, we choose the Real Adaboost algorithm introduced by Schapire to train a cascade based on view. And we optimize the process of classifier training and the output of the weak classifier. It is high-performance in multi-view face detection.2. Algorithm implementation: We construct the multi-view face detection system on three different platforms: DSP, PowerPC and Microblaze. After algorithm and platform optimization, the system based on DSP platform can detect faces at a frame rate of 18 frames per second. On the Microblaze platform, we design the coprocessor to finish the multiply-add and extraction operation which is time-consuming to improve the performance of the system. But it does not achieve the real time detection. On the PowerPC platform we design the coprocessor to calculate the integral image and the system can detect an image at a frame rate of 16 frames per second.The experiments show that the face detection based on Real Adaboost algorithm is high-performance in multi-view face detection. Depends on the high processing speed of TMS320DM642, the system based on DSP can achieve the real-time detection. The system based on PowerPC can detect an image within 0.062s just like DSP. But due to the performance of Microblaze, and the algorithm is not adapted to the coprocessor strategy of Microblaze, it does not achieve the real-time detection. |
PDF Full Text Request