Data Association for Grid-Based Object Tracking Using Particle Labeling

Abstract

Estimating surrounding objects and obstacles by processing sensor data is essential for safe autonomous driving. Grid-based approaches discretize the environment into grid cells, which implicitly solves the data association between measurement data and the filtered state on this grid representation. Recent approaches estimate, in addition to occupancy probabilities, cell velocity distributions using a low-level particle filter. Measured occupancy can thus be classified as static or dynamic, whereby a subsequent tracking of moving objects can be limited to dynamic cells. However, the data association between those cells and multiple predicted objects that are close to each other remains a challenge. In this work, we propose a new association approach in that context. Our main idea is that particles of the underlying low-level particle filter are linked to those high-level objects, i.e., an object label is attached to each particle. Cells are thus associated to objects by evaluating the particle label distribution of that cell. In addition, a subsequent clustering is performed, in which multiple clusters of an object are extracted and finally checked for plausibility to further increase the robustness. Our approach is evaluated with real sensor data in challenging scenarios with occlusions and dense traffic.

Grid-Based Object Tracking With Nonlinear Dynamic State and Shape Estimation

“…This section gives a brief overview of our overall grid-based environment estimation approach [4]- [6] in order to understand the input of this work and the pre-processing steps of it. These processing steps are exemplarily illustrated in Fig.…”

“…The individual cells are thereby directly associated to predicted object tracks, i.e., information about the existing tracks is considered before those cells are abstracted by a clustering. We use our particle labeling association approach as presented in [6]. The basic idea is to link particles χ ∈ X t of the underlying low-level particle tracking with the high-level objects by attaching an object label to each particle.…”

“…The radar velocity measurements z c v,t are represented in the measurement grid structure as separate layers, as defined in ( 4), containing the radial velocity v r z , azimuth angle θ r z , and the position x s of the corresponding sensor origin at the current time t. The radar measurements thereby typically have a higher spatial uncertainty than those of the lidar sensors, thus resulting in a lower occupancy evidence and multiple cells containing the same radial velocity measurement, which in turn is useful for the velocity weighting of the low-level particle tracking. The particle labeling association [6] used in this work, however, discards cells with a low dynamic occupancy evidence, as only cells with a higher occupancy evidence should be considered to form the measurement minimum bounding box of the object tracking.…”

“…This means that if none of the evaluated features contradicts with the class pedestrian, that class should dominate the others. This results in a more robust association, since the extracted size is also used for the gating area of the association [6], which is selected more conservatively that way.…”

“…This enhances and also simplifies the detection and tracking of moving objects, since only measurement data classified as dynamic have to be considered. Parts of our overall grid-based approach have been presented in our previous work on the dynamic grid estimation [4], the extraction of new objects by clustering dynamic cells [5], and the association of dynamic cells with predicted objects [6]. This paper follows on from those concepts and focuses on the subsequent object state estimation.…”