Agreeing to Cross: How Drivers and Pedestrians Communicate (1702.03555v1)

Abstract: The contribution of this paper is twofold. The first is a novel dataset for studying behaviors of traffic participants while crossing. Our dataset contains more than 650 samples of pedestrian behaviors in various street configurations and weather conditions. These examples were selected from approx. 240 hours of driving in the city, suburban and urban roads. The second contribution is an analysis of our data from the point of view of joint attention. We identify what types of non-verbal communication cues road users use at the point of crossing, their responses, and under what circumstances the crossing event takes place. It was found that in more than 90% of the cases pedestrians gaze at the approaching cars prior to crossing in non-signalized crosswalks. The crossing action, however, depends on additional factors such as time to collision (TTC), explicit driver's reaction or structure of the crosswalk.

• The paper introduces the JAAD dataset with over 650 pedestrian crossing samples derived from 240 hours of diverse driving footage.
• The analysis reveals that non-verbal cues, such as gazing and gesturing, occur in over 90% of crossing events, emphasizing their critical role in communication.
• The study’s insights advocate for autonomous systems to adapt to complex human signals, enhancing safety in mixed traffic environments.

Analysis of Non-Verbal Communication in Traffic Dynamics: A Study of Driver-Pedestrian Interactions

This paper provides an essential contribution to the understanding of interactions between drivers and pedestrians at pedestrian crossings, with a particular focus on non-verbal communication. Authored by Amir Rasouli, Iuliia Kotseruba, and John K. Tsotsos, the paper introduces the Joint Attention in Autonomous Driving (JAAD) dataset, comprising over 650 samples of pedestrian behavior across various environments, including city, suburban, and urban roads.

The paper is bifurcated into two primary contributions. First, it introduces a novel and comprehensive dataset valuable for analyzing pedestrian behaviors during road crossings under differing conditions. The dataset, drawn from approximately 240 hours of driving data, allows for the examination of pedestrian behaviors in varied street contexts and weather conditions.

Second, the paper offers a detailed analysis of how non-verbal communication cues are used by pedestrians and drivers during crossing scenarios. A significant finding is that pedestrians engage in some form of non-verbal communication in over 90% of crossing events. Typically, this involves gazing at approaching vehicles as a primary communication cue. Such behavior depends on multiple factors, including time to collision (TTC), the structure of the crosswalk, and explicit reactions from drivers.

In dissecting these interactions, the researchers categorized pedestrian behaviors into groups based on action sequences such as "Precondition + Attention + Crossing" and "Action + Attention + Reaction," among others. This nuanced understanding reinforces the complexity of traffic interactions extending beyond simple mechanical models.

Strong empirical results indicate that non-verbal cues play a crucial role in crossing behaviors. For example, non-verbal communication from pedestrians, including looking or gesturing, often coincides with or prompts similar responses from drivers, facilitating smoother navigation of pedestrian crossings. This contrasts with less complex models that traditionally forecast pedestrian behavior based solely on factors like speed and trajectory.

The dataset and analysis also highlight the situational dependency observed in crossing occurrences. The crossing behavior more prevalently occurs where there are explicit signals such as traffic lights or designated crosswalks. Simultaneously, in scenarios lacking such structures, pedestrian attention and interaction with drivers increase, emphasizing a need for adaptive pedestrian navigation strategies in autonomous systems.

From a practical standpoint, these findings hold significant implications for the design of autonomous vehicles and intelligent transportation systems. Particularly, the integration of mechanisms to interpret non-verbal human cues is critical for improving the safety and efficacy of autonomous systems in mixed traffic environments.

Looking ahead, the research suggests that future work could explore analyzing pedestrian gait patterns relative to attention cues during crossing events. Additionally, capturing driver data, including gestures and reactions, could augment the understanding of implicit communication in road use dynamics.

Overall, this paper underscores the complexity of human-traffic interactions and the critical importance of incorporating non-verbal communication analysis in the development of advanced autonomous driving systems. The JAAD dataset stands as a crucial resource enabling further exploration in this area.