Study On Detection Algorithm Of Fatigue Driving Based On Combination Of Adaboost And Active Shape Model

Traffic accidents are increasing as a result of the development China’s communications and transportation. Among the factors of traffic accidents, fatigue driving takes up a big ratio. In order to decrease traffic accidents and ensure the safety of drivers, researches on automatic detection are of great significance. Based on the theory of machine vision, it is focused on detection software of fatigue driving realizing the combination of Adaboost and ASM. Taking the stability of the detection system, improvements on two aspects are put forward: one is the adaptation to the driver’s posture and the condition of illumination; another one is to utilize various standards including the algorithm of Perclos and the frequency of nictitating and yawn to judge the fatigue condition. The detailed steps are as following:(1) Research on drivers’ fatigue detection system based on the combination of Adaboost and Active Shape ModelOne of the key steps of fatigue detection is eye allocation. In order to improve the accuracy of eye allocation, the advantages of both Adaboost and ASM are combined. Then detection software of driver’s fatigue condition is developed on the basis of both a combined algorithm and OpenCV. First, collected pictures of driver’s face are tested using the algorithm of Adaboost. Then the tested data of face area will be sent to ASM, which is defined as the initial area of the matching domain. After the process of matching the accurate location of eye will be obtained. Numerous simulative experiments show that this software is able to detect drivers’ fatigue conditions successfully, including collecting pictures of drivers, pre-processing, detecting faces, allocating eye, judging drivers’ fatigue condition and so on. What’s more high accuracy of eye allocation can be satisfied.(2) Research on the stable fatigue detection algorithm for the adaptation to drivers’ gestures and illuminationFactors like drivers’ gestures and illumination are ignored in traditional fatigue detection systems. In order to improve this phenomenon, it is focused on a more stable fatigue driving detection algorithm based on the combination of Adaboost and ASM. This new algorithm first unitizes the face picture and spins itself when faces can not be detected, which makes the system adapt to the condition of head-leaning. Second, this system can also adapt itself to various illumination conditions by adding samples pictures of training faces and column diagrams. Tests on simulative videos show that this stable algorithm can allocate eyes more precisely. Above all this algorithm can not only adapt to the conditions of head-leaning but it can also decrease the influence of illumination, which improve the stability of the detection system.(3) Judgment of fatigue driving states using Preclos, frequency of nictitating, yawn and the time of continuous eye closing.In connections with the time limitation of Perclos, frequency of nictitating, yawn and the time of continuous eye closing are also put forward as standard of judging fatigue driving conditions. First Perclos numerical value can be calculated according to the level of nictitating. Meanwhile frequency of nictitating, yawn and the time of continuous eye closing can also be obtained. Whatever the value of Perclos is, the state of eye will be defined as abnormal or absent if the frequency of nictitating and yawn are too high or too low. Additionally it will be defined as dangerous driving if the time of continuous eye closing is more than three seconds. By using those four standards together, the accuracy and the stability of the system will be improved. Finally all those experimental data are applied into veritable situations and testifies that all the following judgment can detect the fatigue driving condition accurately and effectively.