Facial Drowsiness Signs Detection Algorithm Using Image Processing Techniques For Various Lighting Condition

• Main Author: Yuri, Nur Fatin Izzati
• Format: Thesis
• Language: English; English
• Published: UTeM 2017
• Subjects: T Technology (General); TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utem.edu.my/id/eprint/23088/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=107134; TJ211.42.L64 2017

For the past few years, drowsiness signs detection systems have been developed as one of the initiative to reduce car crashes. However, various luminance intensities are one of the major problems in the development of a drowsiness signs detection system. This research studies the suitable image processing techniques to be implemented in a drowsiness signs detection algorithm for various lighting conditions. Four lighting conditions are proposed with the average range of 0 luminance value to 175 luminance value. In this project, the algorithm is developed based on four main algorithms which are the detection algorithm, the tracking algorithm, the preprocessing algorithm and the drowsiness signs analysis algorithm. Viola-Jones algorithm is utilized for face detection. Upon acquiring the face location, the knowledge-based method is implemented to locate the eye and the mouth. After that, Kanade Lucas Tomasi algorithm is applied for tracking purpose. Based on the tracked face and the tracked facial components, the region of interest is selected. Image processing techniques are applied to the eye region and the mouth region to fix the image intensity and to enhance the features of the image. In order to analyse the drowsiness signs portrayed by the eye and the mouth, the operation to determine the eye state and the mouth state is determined. The distance between eyelid is computed to determine the eye state. Meanwhile, the height of the mouth opening is computed to determine the mouth state. There are three drowsiness signs that are analysed for the eye region, namely, the eye blink count, the duration of the eye closure and the percentage of time that the eye is closed. As for the drowsiness sign in the mouth region, the yawning count is computed. This thesis presents a small-scale drowsiness signs database for four lighting conditions. The performance of the algorithm is validated by using the developed database under four luminance intensities and achieved promising results. The performance of the drowsiness signs detection algorithm is fully dependent on the performance of the eye state detection and the mouth state detection. For eye state detection, the proposed technique possessed an accuracy of 98.71 % for 0 luminance value, 97.10 % for 2 luminance value, 98.30 % for 5.2 luminance value and 98.8 % for 174.9 luminance value. As for mouth detection, the proposed technique possessed an accuracy of 99.45 % for 0 luminance value, 98.03 % for 2 luminance value, 99.6 for 5.2 luminance value and 99.7 % for 174.9 luminance value. The proposed technique yielded the overall accuracy of 98.22% for eye state detection and the overall accuracy of 99.23% for the mouth state detection. In conclusion, the proposed technique managed to yield high accuracy for four lighting conditions and could be improved for further research to be implemented in a real time environment.