Improvements of Driver Fatigue Detection System Based on Eye Tracking and Dynamic Template Matching

Driver fatigue detection plays an important role in intelligent transportation systems for driving safety. Therefore, it becomes an essential research issue these years. Recently, Horng and Chen proposed a real-time driver fatigue detection system based on eye tracking and dynamic template matching. In their work, the driver fatigue detection system consists of four parts: face detection, eye detection, eye tracking, and fatigue detection. However, their work suffers from an exhaustive search in eye tracking with the conventional mean absolute difference (MAD) matching function. To remedy the low accuracy in matching and inefficiency in search, in this paper, we first propose two new matching functions, the edge map overlapping (EMO) and the edge pixel count (EPC), to enhance matching accuracy. In addition, we utilize fast search algorithms, such as the 2D-log search and the three-step search algorithms, to expedite search. The experimental results show that the 2D-log search with the EPC matching function has the best performance on eye tracking; it only requires 22.29 search points on average to achieve 99.92% correct rate of eye tracking, as comparing to the original work which requires 441 search points with only 96.01% correct rate. By theoretical analysis, the total amount of computations for eye tracking in the 2D-log search with EPC only takes up to about 10% of the original work. These improvements make the driver fatigue detection system more suitable for implementations in embedded systems. Key-Words: Intelligent transportation system; Driving safety; Driver fatigue detection; Eye tracking; Template matching.