Machine Learning for Autonomous Vehicle Pedestrian Intent Prediction

Abstract

Pedestrian safety is one of the most important key points for implementing autonomous vehicles in urban environments. The Transportation Research Board (TRB) researchers have defined pedestrian safety as a process to ensure that pedestrians can cross and walk beside roads and streets without encountering excessively high accident and injury risks. Common problems for pedestrian safety include adverse traffic impacts and the difficulty of integrating walking and riding into local laws and regulations. Recognizing pedestrian walking intention is essential for predicting the trajectory of pedestrians and is a critical problem in autonomous vehicle safety. Behavior understanding and predicting the trajectory of pedestrians have received increasing academic attention during the last decade. Nevertheless, most of the existing work used traditional techniques (feature-based) for pedestrian intent prediction from sensor data, which may not provide enough discriminatory power.