Autonomous vehicles that understand road agents: Detection, tracking, and behavior prediction

autonomous vehicles;detection;tracking;behavior prediction;driver behavior

Xu, Ke ; Driggs-Campbell ; Katherine Rose

关键词: autonomous vehicles;    detection;    tracking;    behavior prediction;    driver behavior;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/108028/XU-THESIS-2020.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Object detection, object tracking and behavior prediction are three fundamental problems towards human-level road agent understanding. In this thesis, we introduce a joint object detection and tracking model for real-time autonomous driving applications. Comparison with two state-of-the-art models on a research dataset shows that our model has the best detection performance and comparable tracking performance. We implement our algorithm on a real autonomous driving vehicle and conduct public road test to prove the robustness and reliability of our system. We further explore the task of vehicle behavior prediction for high-level understanding of road agents. We introduce the Fusion Seq2Seq model and compare it with two other baseline models. Experiments on a driver behavior dataset shows that our model can reasonably predict ego-vehicle actions.