Description

Real-time Multiple Face detection using Raspberry Pi with Intel Movidius Stick 2

Introduction

Real-time face detection is usual, but detecting multiple faces at the same time, is a little difficult since you need more computational power to detect every face in the crowd. This project works on Raspberry pi, but performance is accelerated using Intel Movidius Neural Compute Stick 2 with the help of OpenVino running on the backend.

Abstract

This project uses Raspberry Pi as the core, it doesn’t have enough computational power to process the frame in real-time and to recognize the face. SO to accelerate the frame rate with high computational power, Intel Movidius neural compute stick is used, which makes the application more efficient by processing every process from NCS instead of processing in Raspberry Pi. It leads to detecting the face from the real-time video at 40 to 60 Frames per second.

Existing system
In the existing system, face detection is done by using the “Haar cascade Frontal face “algorithm in the format of the XML file. 

Disadvantage 

• It cannot detect the multiple faces with a high frame rate 
• The system process will have a slow frame rate.

Proposed System

In this proposed system, this project uses Intel Movidius Neural compute stick to take the whole computational process making the application to be real-time with more efficient with the high frame rate.

Disadvantage 

• Recognition is real-time
• Convolutional Neural Network is a very advanced deep learning algorithm 
• The number of data set is high hence accuracy is also high
• Uses a simple USB camera for video capture
• No 3D camera is used 
• cost-effective 

Block diagram

Hardware Required

• Raspberry Pi
• Intel Movidius Neural Compute Stick 2
• USB Camera
• SD Card

Software tools

• Raspberry pi OS: Raspbian stretch
• Programming Platform: python 3 IDLE
• Programing language: python 3
• Library: OpenCV

Reference:

• [1] Y. Adini, Y. Moses, and S. Ullman, “Face recognition: The problems of compensating for changes in illumination direction,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 19, no. 7, pp. 721–732, Jul. 1997. 
• [2] M. S. Bartlett, H. M. Lades, and T. Sejnowski, “Independent component representation for face recognition,” in Proc. SPIE Conf. Human Vis. Electron. Imag. III, San Jose, CA, 1998, vol. 3299, pp. 528–539. 
• [3] P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman, “Eigenfaces vs. fisher face Recognition using class specific linear projection,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 19, no. 7, pp. 711–720, Jul. 1997. 
• [4] “Face recognition vendor test (FRVT),” 2006 [Online]. Available: http://www.frvt.org/FRVT2006/ 
• [5] X. Geng, D.-C. Zhan, and Z.-H. Zhou, “Supervised nonlinear dimensionality reduction for visualization and classification,” IEEE Trans. Syst. Man Cybern. B, Cybern., vol. 35, no. 6, pp. 1098–1107, Dec. 2005.