On track-to-track data association for automotive sensor fusion

When fusing data from more than one information source, it is important to associate the correct pair of the data available from the information sources to achieve an optimal fused result. The responsibility of the task of proper association lies on the data association method used in the system. The design of the data association method has to be done considering the requirements of the application, quality and quantity of the information provided by the sensors. Parameters such as: whether the fusion is carried out in a centralized or decentralized architecture, whether the data available from the sensors is raw and unprocessed data or already processed by the built-in signal processing system of the sensor, plays a role in the finalization of the data association problem. The data association problem in a decentralized sensor fusion setting, also known as track-to-track association, is discussed in detail in this paper. The information sources used in this paper are environment perception sensors based on different measurement principles used for automotive safety and autonomous driving functions. The sensors deliver kinematic and as well as non-kinematic information on the tracked targets. To make use of this non-kinematic target information, attribute based association methods in-addition to the traditional data association methods and results based on the real world data are presented in this paper. A video recorded under real world test conditions that include sensor data and results will be made available for the community.