Behavior signal processing for vehicle applications
Chiyomi Miyajima, Pongtep Angkititrakul, Kazuya Takeda · APSIPA Transactions on Signal and Information Processing · 2013
Within the past decade, analyzing and modeling human behavior by processing large amounts of collected data has become an active research topic in advanced human-machine interaction systems. The research community strives to find improved ways to explain and represent meaningful behavioral characteristics of humans in order to develop efficient and effective cooperative interactions between humans, machines, and the environment. This paper provides a summary of progress achieved to date of our research on behavior signal processing, with a focus on the driver-vehicle-environment interaction. First, we describe the method of data collection used to develop our real-world driving corpus, which contains multimodal driving signals capturing relevant information regarding driver, vehicle, and environment. Then, the paper provides an overview of our signal processing and data-driven approaches used to analyze and model driver behavior for a wide range of practical vehicle applications. We then perform experimental validation using the realistic driving behavior of several drivers. In parti cular, the vehicle applications include driver identification, be havior prediction (i.e., car following and lane change), driver fr ustration (emotion) detection, and driver education. We hope that this paper will provide some insight to researchers who have interest in this field, and help identify areas and applications where further research is needed. I. I NTRODUCTION Human behavior plays an important role in any system involving human-machine interaction. In regards to driv- ing, when analyzing driver-vehicle-environment interactions (Fig. 1), human errors contribute to more than 90% of fatal traffic accidents. Understanding human/driver behavior ca n be useful in preventing traffic collisions (21), as well as enha ncing effectiveness of the interaction between driver, vehicle, and en- vironment. The study of driver behavior is a very challenging task due to its stochastic nature with high degree of inter- and intra-driver variability. To cope with these issues, ov er the past decades data-centric approaches have gained much attention in the research community (8), (14), (15), (18). I n this research, we focused on the understanding of human behavior from a signal processing perspective, and on developing a methodology to analyze and model the extracted meaningful behavioral information. To analyze and model driver behavior, the first step is to collect a reasonable amount of realistic multi-modal ob- servations. Here, observations or driving signals represe nt behavioral variables as a time series which possesses particular dimensions of behavioral characteristics. We took extra care in designing and developing our instrumented vehicle to collect a broad range of driving signals which could represent relevant