Abstract

View Abstract PDF

DARPA‚Grand Challenge in the early 2000s kick-started an aggressive pursuit (by technology and automotive companies alike) of driverless cars. These vehicles face a myriad of challenges in urban settings however, chief among them is ensuring safe interactions with pedestrians. Most state-of-the-art autonomous vehicles use a combination of radar, LIDAR, and cameras to detect and track pedestrians. A significant amount of work has focused on inferring pedestrian intent (e.g., predicting when and if a person will attempt to cross a roadway) using machine learning-based computer vision techniques. One issue with these approaches is that the sensors they rely on may be unusable in certain weather and lighting conditions. To mitigate such issues, we should always build autonomous systems with redundancy by utilizing an array of techniques and sensors that ensure safety and reliability. To that end, this work discusses a novel algorithm for predicting a pedestrian‚Äôs intent to cross a roadway. This algorithm runs on modern smartphones and it broadcasts pedestrian intent information to nearby vehicles. Paired with existing computer vision-based techniques for pedestrian crossing inference, this new combined system comprises a more robust solution for autonomous vehicles.