| According to statistics from a German car market research institution, the number of vehicles breaks through billion in 2010 and it will increase twenty percent next seven years. With the increasing of vehicles, people are more and more attention to traffic safety. According to figures from The World Health Organization, because of traffic accidents, 1.2 million people lost their lives in the world every year. Fatigue driving is one of the three most main factors which cause traffic accidents. Traffic accidents which be caused by Fatigue driving account for 25%~30% in total traffic accidents. In order to reduce casualties and property losses, many countries begin to research fatigue driving.The main research contents of the thesis are to develop a set of driver fatigue driving early warning devices which possesses practical value. According to comparing various test algorithms each other, combining with drivers'characteristic and rule, this thesis finally chooses using pattern recognition and image processing technology algorithm to analyze drivers'eyes states. If the algorithm is working, the next job is transplanting the algorithm into hardware platform which bases on DSP chip. The development of the device in this thesis needs to satisfy vehicle-mounted fatigue driving early warning system for real-time, non-contact basic requirements.This dissertation selects popular MB-LBP feature to describe face and eyes, advantage of the feature is insensitive with noise, chooses Gentle adaboost algorithm, which is the most excellent algorithm in adaboost algorithm family to train face detection and eye location classifiers. High detection rate weak classifiers are selected through training. According to weighting, weak classifiers make up strong classifiers, and then strong classifiers make up cascade classifiers. Finally repeating experiments and adjusting parameters make the device possess higher eyes detection rate in PC.The algorithm can be transplanted into DM642 development board by using CCS3.1 which was developed by TI Company. In the course of transplant, various optimization methods are used to make the program simply to satisfy the requirement of real-time testing.After completing algorithm development and program transplant, the third step is to establish respectively experiment devices based on PC and DSP, use experiment devices to analyze real-time and correctness of the algorithm. Through the experiment, the algorithm is improved to achieve basically desired objectives.If the algorithm works well in PC and DSP hardware platform, the next job is to design the circuit and PCB of the device on the basis of DM642 development board to make the whole device work in the car. |
