Introduction
Despite substantial developments with in-vehicle drowsy driver detection, a robust, affordable commercial product has yet to tap into the market and impact driver drowsiness occurrences. Most existing systems are expensive (to create, maintain, what?), high cost, obtrusive, have operational limitations, or lack of robustness in the trucking domain, general effectiveness or driver acceptance.
In a recent report from congress, fatigue was cited as an associated factor in 13% of all large truck crashes. However, in most cases, drowsy driving is usually underestimated due to unreported off-road crashes, inability to verify drowsiness as a contributing factor and erroneous driver reports. Transecurity wanted to further investigate ways to reduce these occurrences and offer affordable solutions.
It is unlikely a single, real-time measure of driver drowsiness is a realistic solution. Therefore, a multi-measure algorithm based on variables that are likely to be available and can be obtained in real time will be required to create a robust system. In particular, such an algorithm must be able to handle missing or imperfect data sources, without significantly compromising overall accuracy. Such a system will likely make use of driver, vehicle, and environmental factors.
Project Goals/Phase 1
When it comes to research and development, Transecurity always establishes clear, specific, goals to help keep the team on track and enable us to collect clear data we can use to develop accurate, dynamic products. For this project we focused on the following goals:
- Develop a database of drowsy driving events from existing naturalistic driving data sets. Events will be used generate and validate candidate multi-measure algorithms.
- Discriminate analysis and classification techniques will be used to identify predictive measures, generate candidate multi-measure algorithms and test the algorithms for accuracy.
- A video-based eye and head tracking system will be enhanced and improved to optimize the ability to accurately measure PERCLOSE and other drowsiness-related behaviors.
- The optimal algorithm will be implemented into an in-vehicle processing system making use of eye tracking, lane tracking and other driver, vehicle, and environmental measures.
- An on-road demonstration of the system will be created to measure the functional potential of the multi-measure algorithm.