Local Volumetric Hybrid-Map-Based Simultaneous Localization and Mapping With Moving Object Tracking

In this paper, we present a novel framework for solving two environmental perception tasks concurrently-simultaneous localization and mapping, and moving-object tracking-by using a Velodyne laser scanner. To extract proper input data for these tasks from the sensor, several sensor data preprocessing algorithms are first addressed. For the simultaneous localization and mapping problem, we propose a local volumetric hybrid-map-based approach using Rao-Blackwellized particle filters. We represent the static environments with the hybrid map consisting of feature and 3-D grid maps. This framework basically allows us to utilize the traditional approaches using a single map. In addition, we derive a new sampling formula by combining a feature measurement likelihood to the traditional grid-map-based approach, and this significantly improves the accuracy and efficiency of the algorithm. The proposed moving-object tracking algorithm is achieved based on the geometric and multiple motion models. We introduce a robust extraction and parameterization method of the geometric shape based on predefined contour models. Then, the geometric shape is inferred with a multiple-base-point method. We establish three motion models, which are utilized for tracking in an adaptive way by using the well-known interacting multiple model algorithm. The algorithms proposed are evaluated using the data sets collected from our test vehicle in the complex urban scenarios. The experimental results show that our approach works well even in real outdoor environments and outperforms traditional approaches.