Self-Alignment-by-Tracking With a Mobile Single Robot

In this paper, we address multiple object tracking (MOT) using a single mobile robot without any prior knowledge of its location or orientation. In scenarios where both objects and the robot are moving, continuous object tracking becomes challenging due to the lack of a consistent coordinate system. To overcome this, we propose a self-alignment-by-tracking mechanism to maintain spatio-temporal consistency. We formulate an optimization problem aimed at maximizing multiple object tracking accuracy (MOTA), but its combinatorial nature makes it difficult to solve directly. To address this, we decompose the problem into two subproblems: (1) consecutive spatial alignment of objects over time sequences to a common reference coordinate system and (2) object tracking within the same coordinate system for temporal consistency. Compared to conventional tracking processes, our proposed self-alignment-by-tracking mechanism requires an additional step of spatial alignment before the data association step, which may introduce inter-frame delays. However, we mitigate this issue by employing a geometry-based sorting mechanism, allowing us to search only a subset of the sorted neighboring objects. In addition, robust spatial alignment can be achieved by generating pseudo-environment objects through a closed-form expression, ensuring reliability even under challenging matching conditions.We analyze theoretical computational complexity, along with Monte Carlo simulations and real-world experiments, which validate the superiority of our proposed MOT process in terms of both accuracy and computational efficiency.