Robust Tracking of Reference Trajectories for Autonomous Driving in Intelligent Roadside Infrastructure

Abstract

High quality reference data is crucial for the development of autonomous driving applications. Unfortunately, datasets including fixed, reproducible static environments that contain manifold interactions between traffic participants are not widely available. In this paper we propose a camera based trajectory estimation framework that enables the generation of reference trajectory data in stationary roadside infrastructure. We develop a Simple Online Realtime Tracking (SORT) algorithm that tracks objects in image space utilizing the tracking-by-detection paradigm with a deep neural network detector. By projecting tracks to a ground model, we are able to gather cartesian and georeferenced trajectories for manually driven and autonomous vehicles in the field. We evaluate the framework in stationary roadside infrastructure in the Test Area Autonomous Driving Baden-Württemberg, Germany. A vehicle equipped with inertial measurement unit and differential GPS is used to generate ground truth positions that are compared with our framework.