Analyzing track management strategies for multi object tracking in cooperative autonomous driving scenarios

Abstract For autonomous driving to operate safely it is crucial to perceive surrounding objects correctly. Not only detection but also state estimation (track) of a perceived object is urgent. The state is required to enable a safe motion planning, since it allows to predict the future position of an object. To include only valid information, the state estimations must be maintained to determine which track is active and which is not. Mostly, a simple count-based approach is used. For this, we present an investigation of two common approaches from non-cooperative track management in comparison to two new management strategies to maintain tracks in a cooperative scenario. We evaluate them using three simulated scenarios with a varying rate of cooperative vehicles. A confidence-based approach was able to increase the average precision by up to 9 percentage points.