Augmented vehicle tracking under occlusions for decision-making in autonomous driving

摘要: This paper reports on an algorithm to support autonomous vehicles in reasoning about occluded regions of their environment make safe, reliable decisions. In driving scenarios, other traffic participants are often from sensor measurements by buildings or large like buses trucks, which makes tracking dynamic objects challenging.We present a method augment standard object trackers with means 1) estimate the state agents and 2) robustly associate estimates new observations after tracked reenters visible region horizon. We perform estimation using dynamics model that accounts for behavior hybrid Gaussian mixture (hGMM) capture multiple hypotheses over discrete behavior, such as along different lanes turning left right at intersection. Upon observations, we them existing terms Kullback-Leibler divergence (KLD). evaluate proposed simulation real-world traffic-tracking dataset vehicle platform. Results show our can handle significantly prolonged occlusions when compared system.