2020/04/02 by Dominik Notz, Notz, Dominik, Felix Becker +5
Computer Science · Engineering · #Advanced Neural Network Applications #Autonomous Vehicle Technology and Safety #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Robotics (cs.RO) #Video Surveillance and Tracking Methods #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2004.01288
openalex publication_date 2020/04/02 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Collecting realistic driving trajectories is crucial for training machine\nlearning models that imitate human driving behavior. Most of today's autonomous\ndriving datasets contain only a few trajectories per location and are recorded\nwith test vehicles that are cautiously driven by trained drivers. In particular\nin interactive scenarios such as highway merges, the test driver's behavior\nsignificantly influences other vehicles. This influence prevents recording the\nwhole traffic space of human driving behavior. In this work, we present a novel\nmethodology to extract trajectories of traffic objects using infrastructure\nsensors. Infrastructure sensors allow us to record a lot of data for one\nlocation and take the test drivers out of the loop. We develop both a hardware\nsetup consisting of a camera and a traffic surveillance radar and a trajectory\nextraction algorithm. Our vision pipeline accurately detects objects, fuses\ncamera and radar detections and tracks them over time. We improve a\nstate-of-the-art object tracker by combining the tracking in image coordinates\nwith a Kalman filter in road coordinates. We show that our sensor fusion\napproach successfully combines the advantages of camera and radar detections\nand outperforms either single sensor. Finally, we also evaluate the accuracy of\nour trajectory extraction pipeline. For that, we equip our test vehicle with a\ndifferential GPS sensor and use it to collect ground truth trajectories. With\nthis data we compute the measurement errors. While we use the mean error to\nde-bias the trajectories, the error standard deviation is in the magnitude of\nthe ground truth data inaccuracy. Hence, the extracted trajectories are not\nonly naturalistic but also highly accurate and prove the potential of using\ninfrastructure sensors to extract real-world trajectories.\n