| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 101 === In modern society, due to the advancement in technologies and the reduction of tariffs, buying a vehicle is not very difficult. The increase of vehicles causes that the traffic accidents are occurred frequently, and it results to pay a lot of costs. According to statistics in accident events, more than ninety percent vehicle accidents in Taiwan causes by human factors.
In this thesis, we propose a computer-vision-based system to track the status of eyes to determine whether the driver is drowsy, and the system warns the driver in advance to prevent the occurrence of dangerous driving conditions, and then traffic accidents are reduced. In Taiwan, owing to the exam-oriented environment, the proportion of wearing glasses in high school students is up to eighty percent. Therefore, there is a high probability that a driver wears glasses. In image processing systems, the facial frame with eye glasses will affect the judgment of open-closed eyes conditions, and then we develop a system that can analyze the open-closed eyes status with wearing glasses.
The proposed system mainly detects the position of the nosepiece. Firstly, we use the Canny edge filter, which is combined with the Gaussian filter, and the edges of the facial image are captured, and then we get thin edges of the images. The advantages are reducing noises and reducing the difficulty of the subsequent judgment. Secondly, we use the nosepiece filter to search out the nosepiece features. The inverted U-shaped filter is applied to match the candidates of nosepiece features, and the correct position of the nosepiece is found. Finally, the nosepiece is used as a reference point, and the left and right glasses in facial video frames are found respectively, and then eye positions located in the glasses are also detected. In general, the positions of eyes will be located into the glasses frame, and we assume that the eyes must be located within the glasses frame. For driver drowsiness detections, the frequency of the open-closed eyes status can be used to identify whether the driver dozes or not. In this thesis, to judge the open-closed eyes status, the vertical projection is used to detect the gray level of the white region inside eyes and the gray level of eyeballs.
In this study, two other experimental methods for feature point searches are also discussed, where the first one uses the color space based method to find the nosepiece, and the second one uses the ellipse template matching based method to find the nosepiece. We compare the accuracy and the execution speed among these three methods, and also analyze the advantages and disadvantages among these methods.
|
|---|
