GPU-GLMB: Assessing the Scalability of GPU-Accelerated Multi-Hypothesis Tracking

TL;DR

This work addresses the computational bottleneck of multi-hypothesis tracking with Generalized Labeled Multi-Bernoulli (GLMB) filters by introducing a variant that supports multiple detections per object without explicit geometry modeling. The authors show that breaking inter-detection dependencies in the update step enables fully parallel, GPU-friendly sampling and batched particle updates, implemented in PyTorch as GPU-GLMB. Extensive experiments demonstrate real-time performance on server GPUs for scenarios with up to 20 objects and 100 hypotheses, with favorable cardinality and tracking accuracy when appropriately configured. The results indicate strong potential for integrating GPU-GLMB into distributed sensing networks and motivate future work to scale to larger, more complex multi-object/multi-sensor environments.

Abstract

Much recent research on multi-target tracking has focused on multi-hypothesis approaches leveraging random finite sets. Of particular interest are labeled random finite set methods that maintain temporally coherent labels for each object. While these methods enjoy important theoretical properties as closed-form solutions to the multi-target Bayes filter, the maintenance of multiple hypotheses under the standard measurement model is highly computationally expensive, even when hypothesis pruning approximations are applied. In this work, we focus on the Generalized Labeled Multi-Bernoulli (GLMB) filter as an example of this class of methods. We investigate a variant of the filter that allows multiple detections per object from the same sensor, a critical capability when deploying tracking in the context of distributed networks of machine learning-based virtual sensors. We show that this breaks the inter-detection dependencies in the filter updates of the standard GLMB filter, allowing updates with significantly improved parallel scalability and enabling efficient deployment on GPU hardware. We report the results of a preliminary analysis of a GPU-accelerated implementation of our proposed GLMB tracker, with a focus on run time scalability with respect to the number of objects and the maximum number of retained hypotheses.

Figures (4)

• Figure 1: Visualization of testbed ground truth track.
• Figure 2: These plots show metrics for the case of 20 objects, a maximum of $100$ hypotheses, and the L40S GPU. (a) shows the true cardinality (e.g. the true number of objects in the environment) at each time point. (b) shows the cardinality error of the tracker. (c) shows the number of hypotheses at each time point. (d) shows the update run time at each time point. For (a)-(c) the results are the mean of 10 runs and the shaded region is the standard error of the mean.
• Figure 3: These plots show the per-update mean run time of GPU-GLMB on several hardware platforms as described in Table 1. Each line shows the per-update mean run time for a different number of ground truth objects. The shaded regions correspond to one standard error of the mean.
• Figure 4: These plots show the average cardinality and tacking errors as a function of the number of objects and maximum number of hypotheses computed using the L40S GPU.