Chapter 8 - Driver Lane-Change Intention Inference

Abstract

In this chapter, driver intention inference will be discussed from two different aspects, which are host driver intention inference and surrounding driver intention inference. Specifically, the host driver's intention follows the driver intention inference framework that we introduced in Chapter 2. The main motivation is to predict the driver's tactical maneuver before it is initiated. The second part introduces the surrounding driver intention detection, which can be transformed into a vehicle path prediction task. As the behaviors of the surrounding driver cannot be recognized directly, the driving intention of the surrounding drivers can only be estimated based on the outer vehicle status. Regarding the host driver's intention inference, with the development of advanced driver assistance system (ADAS) and the intelligent vehicles, drivers need to share their control authorities with the intelligent control units. It is critically important for the ADAS to understand the drivers' intent and their following maneuvers. Therefore in this section, an intention inference system that particularly focuses on the lane-change maneuver on the highway is proposed. First, a general driver's intention mechanism and framework are introduced. Then, the vision-based intention inference system, which use multiple low-cost cameras and vehicle data acquisition system to capture the multimodal sensory inputs are proposed. The vision system is designed to simultaneously capture both the in-vehicle and outer vehicle features. To predict the driver's intention, a deep recurrent neural network (RNN) with long short-term memory (LSTM) units is proposed to deal with the time-series driving data and temporal pattern. The experiment data are collected with multiple drivers on the highway environment. Statistical analysis for the lane-change maneuver sequences is performed. It is found that drivers tend to perform mirror checking 6 s before the lane-change maneuver, and the time interval between steering the wheel and crossing the lane is about 2 s. Before the driver initiates a lane-change maneuver, the RNN model achieves 95% inference accuracy for the lane-change intention.