Moving objects observation from running automobile based on real-time map-size free localization

摘要

This paper proposes a method of observing moving pedestrians and cars by detecting and tracking moving objects at map coordinates using a light detection and ranging (LIDAR) sensor mounted on a running automobile. The pro-posed method has two key techniques: 1) real-time 6-degrees-of-freedom robot localization using 3D maps, and 2) tracking multiple moving objects, which applies the automobile's speed. The localization method keeps the load of the memory constant by switching the likelihood fields that convert divided 3D maps to migrate them extensively. Moving objects are detected using filtered LIDAR input based on estimated robot positions, and they are tracked by a method that deals with changing the number of multiple objects. Then, the system stores the moving objects' trajectories at map coordinates, along with the objects' class, time, and velocity. The proposed system generates a map embedded with the trajectories of moving objects by integrating the aforementioned technologies. We implemented our method with an automobile, and the experimental results demonstrated its effectiveness in an urban area.