| According to the statistics analysis of traffic accident in recent years, it is known that many people died in traffic accident because of fatigue driving. How to effectively detect fatigue driving has highly attracted scientists’attention. In this paper, the combination of computer vision technology and image processing technology is regarded as the core technology, and based on visual recognition, a set of driver fatigue detection algorithm is developed for early warning system.Previous researches on fatigue detection is concentrated to simple situations where a driver drives a vehicle in daytime. In facts, a driver usually drives vehicle in wearing sunglasses in strong light or drives in the night. Therefore the fatigue detection system is difficult to detect fatigue under complex conditions, and bad weather conditions. In order to solve this problem, we propose a visual multi-channel eye classifier based on improved LBP features. Due to the current several mainstream human face database cannot meet the requirements of the system to train the classifier, we propose a new concept of specific face database, and have made a NIR face database with Multi-Poses and Multi-Scenes. AdaBoost algorithm is used to train the eye cascade strong classifiers. At the same time, the traditional LBP (Local Binary Pattern)feature is improved from two aspects, the one is to increase the sample library, the other one is to limit the weight of the sample in the training process. Finally the driving condition of the driver is judged by the PERCLOS criterion.After completion of the design and test of the algorithm, the algorithm is implemented on development platform of visual IoT XC-7600 which is based on the ARM and the Cortex-A9 processor. The main work includes design of the boot of Linux system and Linux kernel system transplant. In addition, the software like Qt and Opencv are installed in Linux system. The experiment show that the algorithm works well on the development platform of visual IoT, and achieves the expected goal in the aspects of fatigue detection accuracy, real-time performance and system robustness. The problems are solved for the situations where a driver is wearing sunglasses or drive at night. |
