A widely-used approach for extended object tracking is based on random matrices, where the scattering matrix, i.e., measurement spread, is used to update a symmetric positive definite random matrix representing an elliptic extent. However, for lidar data, a mismatch between the assumed measurement model and observed data hinders the estimation quality of the method. We propose adaptions to the random matrix approach in order to facilitate the application for tracking a rectangular extended object based on contour measurements. Specifically, we derive a suitable scaling factor for the scattering matrix of measurements in this setting. Furthermore, we propose a simple yet effective estimation scheme for the target center, adapting the shape estimate accordingly. The resulting algorithm closely follows the framework of the random matrix approach. A detailed comparison with a variety of state-of-the-art trackers is carried out in a simulation based on real-world lidar parameters, confirming the effectiveness of the approach.

A novel concept for indoor self-localization based on rotating artificial landmarks with known locations using short-range radar is proposed. First, a processing pipeline for extracting range and angle measurements to the landmarks from a raw radar image is introduced, which consists of a neural network for distance estimation and a basic angle-of-arrival estimator. Second, a particle filter for tracking the pose based on the range and angle measurements is developed. Due to the ability of radar to measure range rate, i.e., the velocity in the direction of a detection, it is possible to robustly detect and localize rotating landmarks with the help of their micro-Doppler pattern. In this way, localization is possible even under difficult conditions (e.g., light changes). Experiments with a wheeled mobile robot and common office fans as landmarks demonstrate the effectiveness of the approach for indoor localization.

Successful tracking of an extended object requires accurate estimation of the target’s shape. In order to evaluate the estimation performance of such tracking algorithms, the target shape must be incorporated by a suitable metric or score. In this context, a common task is to determine a consolidated scalar score for the estimation accuracy across a set of different target types. We highlight problems exhibited by existing scores in this case, particularly if the data set consists of objects of diverse sizes. To this end, we focus on elliptical targets. Furthermore, we present a scale-invariant adaption of the commonly used Gaussian Wasserstein Distance, which does not suffer from the highlighted problem.

Recursive estimation of the orientation and spatial extent of an object from sparse measurements is a challenging yet crucial problem. To this end, we propose a problem-tailored particle filter that employs a special importance density for efficient sampling from high likelihood regions. This importance density is obtained with the help of a suitable analytic method for determining the length and width of the object. Due to the geometric meaning of the state vector, a consolidated estimate from the individual particles is obtained based on the Gaussian Wasserstein distance. The resulting filter can be employed with different analytic methods, and is found to improve estimation accuracy beyond state-of-the-art algorithms in a challenging scenario. Especially for low measurement rates, the proposed filter yields improved results compared to the reference methods.

Multi-sensor fusion can improve environment perception, e.g., by increasing the field of view in collective perception, where intelligent vehicles communicate. Track-to-track fusion in a collective perception scenario makes it necessary to associate tracks from multiple sensors. Especially in settings with many sensors that have limited field of view, track-to-track association can be quite challenging. In this work, we develop a stochastic optimization-based approach for an arbitrary number of sensors with a limited field of view, which utilizes a cluster likelihood to sample joint associations. The proposed method clearly outperforms a greedy approach and computes the most likely associations with only a few samples. We evaluate the approaches on simulated data in static and dynamic scenarios.