Abstract Every year, thousands of vehicles are involved in crashes which are attributed to the onset of driver drowsiness. As a result, there are numerous drowsy driver assistance systems (DDAS) available on the market; however, many of these technologies rely on a single predictor of driver drowsiness (e.g., eye closures, lane position, steering). Relying on only one predictor of drowsiness makes the system susceptible to periodic intervals in which data is unavailable due to failure of the single sensor, usage outside of the sensor’s envelope of operation, or driver’s individual differences. Driver drowsiness measures can be classified as either driver-based (those measures derived from the human) or vehicle-based (those measures derived from the vehicle). For driver-based measures, PERCLOS (a measure of slow eye closure) is considered to be a robust measure of driver drowsiness. Machine-vision (MV) slow eye-closure sensors have been developed to estimate the percent of eye closure and calculate the PERCLOS value. However, these MV slow eye-closure sensors’ ability to detect the eye closures is challenged by eyewear, ambient illumination, and head movement. For vehicle-based drowsiness metrics, lane position appears to be a key indicator of driver drowsiness. Lane position is typically estimated through MV technology detecting lane edge markings on the forward roadway scene. The absence of lane edge markings on roadways or instances of low contrast between lane markings and the surrounding scene make it difficult for this MV lane position sensing technology to accurately measure the vehicle’s position within the lane. Typical causes of low contrast lane markings include poor lane marking quality, artificial overhead lighting, or headlight “blooming.” Therefore, a multi-measure approach, that uses multiple distinct sensors, can offer not only sensor redundancy but also provide a data fusion approach whereby both measures provide more robust drowsiness detection than either measure could alone. This chapter describes the salient measures of driver drowsiness, the concept of data fusion in DDASs, and provides a case study of a prototype DDAS that integrates two drowsiness metrics (i.e., PERCLOS and Lane Position) to form an enhanced drowsiness estimate that may prove to be a more robust measure in a real-world, field application as compared to a single metric system.
Advances in Drowsy Driver Assistance Systems Through Data Fusion
Handbook of Intelligent Vehicles ; 895-912
2012-01-01
18 pages
Article/Chapter (Book)
Electronic Resource
English
Drowsy and Fatigued Driver Warning, Counter Measures, and Assistance
Springer Verlag | 2012
|IEEE | 2018
|