Deep Fusion of Multi-Object Densities Using Transformer

摘要

The fusion of multiple probability densities has important applications in many fields, including, for example, multi-sensor signal processing, robotics, and smart environments. In this paper, we demonstrate that deep learning based methods can be used to fuse multi-object densities. Given a scenario with several sensors with possibly different field-of-views, tracking is performed locally in each sensor by a tracker, which produces random finite set multi-object densities. To fuse outputs from different trackers, we adapt a recently proposed transformer-based multi-object tracker, where the fusion result is a global multi-object density, describing the set of all alive objects at the current time. We compare the performance of the transformer-based fusion method with a well-performing model-based Bayesian fusion method in several simulated scenarios with different parameter settings using synthetic data. The simulation results show that the transformer-based fusion method outperforms the model-based Bayesian method in our experimental scenarios. The code is available at https://github.com/Lechili/DeepFusion.