Likelihood-field-model-based vehicle tracking with Velodyne

Abstract

Vehicle tracking is an important module for an Autonomous Land Vehicle (ALV) navigation in urban environments. In this paper, we propose a novel vehicle tracking module with a Bayesian filter based on the likelihood field model for our ALV, which is equipped with a Velodyne LIDAR and an Inertial Navigation System (INS). At each time step t, Scaling Series importance sampling algorithm is ran on the associated measurements of each tracker with an uniform prior to obtain a weighted particle set. Each particle represents a possible prior pose of the tracker in the current scan. Then for each particle, its weight is adjusted with the weighted particles at time step t − 1 via Bayesian recursion equation to capture the vehicle dynamic model. These final weighted particles approximate the posterior belief of the corresponding tracker and the one with the maximum weight is chosen as the output of the tracking result at time step t. Both the quantitative and qualitative performance of our vehicle tracking algorithm is validated on the Velodyne data collected by our ALV in various environments.