Tracking objects of arbitrary shape using expectation-maximization algorithm

We address the general object tracking with arbitrary shape using rangefinders, which is a key module for detecting surrounding traffic and infrastructure for an autonomous driving vehicle. An Expectation-Maximization (EM) algorithm with locally matching is proposed for motion estimation between two consecutive range images. The complexity of the algorithm is O(N) with N the numbers of scan points. Quantitative performance evaluation of the algorithm using a benchmarking vehicular data set. Results of road tests show the effectiveness and efficiency of the implemented system.