Tracking to improve detection quality in Lidar for autonomous driving

Enabling Lidar systems to detect objects at very long ranges has the potential to be extremely valuable for autonomous driving applications, but is challenging due to noise. In this work, we leverage information from multiple consecutive frames to improve the detection capabilities of Lidar systems. We develop a mathematical model whose solution gives a low memory and low computation algorithm that detects some fraction of the objects present while keeping the number of false positives small. Performance of the proposed method is characterized using simulations of a realistic Lidar chain.