Deriving Pattern in Driver's Observability in Road Turns & Traffic Lights: Eye-Tracking based Analysis

Abstract

As one move towards driverless cars, there will always be a big worry of how autonomous cars would behave in the presence of vehicles driven by humans. In the co-existence model, it is essential for autonomous systems to 'understand' the behavior and gazing patterns of the drivers across different road turns and traffic lights. It is essential to understand that each road turn in a city is different due to angle of turn, building environment, etc. Hence, one needs to understand the gaze patterns of drivers across different turns and traffic lights. This paper is a long-drawn effort for measuring driver's observability and deriving driver's observance pattern in real-time. In this regard, an experimental model is provided and clustering-based technique is applied that would measure driver observability. More than 100 segmented readings were extracted from the video of five different vehicles and drivers under two different road conditions. It was observed that - while taking left or right turns and waiting in traffic light, the focus of drivers in front of their cars also changes considerably, with significant variation in the driver's gaze point, both horizontally and vertically. This is a important pattern/result in the design of adaptive driver assistance system and future driverless cars.