Fast algorithm design for driving drowsiness detection in a driving recorder device

• Main Authors: PAI, YUN-JUI, 白筠睿
• Other Authors: HAN, CHIN-CHUAN
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/5zne8u

碩士 === 國立聯合大學 === 資訊工程學系碩士班 === 106 === Driving in drowsiness is a very dangerous driving behavior. Especially, many fatal accidents occur due to driver drowsiness from many news. The study on drowsy driving detection attracts the attention from many academic researches and information technology (IT) companies. Drivers have to put on the sensors on head in sensor-based detection. However, it is uncomfortable and drivers always forget to put on. In this thesis, an image-based drowsy detection has been developed on driving recorders. Currently, drowsiness detection algorithms using images are implemented in general personal computers with high computational power and storage. However, it is expensive and hard to implement on low-end driving recorders because of the cost. We modified the Viola’s face detection and implemented on the embedded systems. After face detection, facial landmarks are identified using face alignment algorithm. This algorithm is a forest tree-based search method with local binary features(LBF). the locations of eye’s landmarks are used to determine the eye and mouth status. In addition, the panning angle of head is calculated according the detected landmarks. The eye status and panning angle of head determine if the drivers are in the dangerous driving status or not. All programs are implemented in C programing language. To evaluate the effectiveness of the proposed algorithm, the program is also implemented on PC for simulation. More than 10 video clips with 3,000 face images are tested in which facial landmarks were manually labelled. The implemented algorithm is compared with that of Open-CV tool kit. The detected errors of facial landmarks are acceptable. 4-6 frames per second (fps) are achieved on the embedded systems.