Automated Multi-Object Tracking for Autonomous Vehicle Control in Dynamically Changing Traffic

Abstract

Public road traffic is rich in examples of dynamic objects suddenly appearing/disappearing in/from the Field of View (FoV) of an autonomous ego vehicle, such as when target vehicles zoom out by accelerating from the ego vehicle or sensor detections deteriorate temporarily due to shadows and other environmental effects. Thus, the guidance and control system should capture the motion of moving obstacles by a perception and tracking module capable of track management features including but not limited to track initiation and termination. This paper presents such a module that executes multi-target tracking with the linear integrated probabilistic data association (IPDA) filter in conjunction with a model predictive control (MPC) scheme for path and reference speed tracking and obstacle avoidance. From the perception module, all the confirmed tracks are made available to an Interacting Multiple Model (IMM) subsystem which predicts the motion of target vehicles to constrain the optimization problem in the MPC. The paper includes illustrations of the proposed scheme with practical traffic scenarios, which show that the ego vehicle is able to autonomously react to random changes in the number and/or state of target vehicles as well as to occasional missed detections due to environmental effects.