Neuromorphic Extended Object Tracking and Classification Using Spiking Neural Networks

Extended object tracking systems are becoming more relevant for autonomous systems due to new innovative sensor systems and the cost reduction of existing sensors. Data from such sensors can be modelled as sets of spare asynchronous spikes, opening the door to the use of Spiking Neural Networks (SNNs). SNNs offer a more energy-efficient method of processing data when compared to Artificial Neural Networks (ANNs). By leveraging SNNs for extended object tracking, we illustrate in this paper how to create fast and robust tracking systems capable of directly processing data from multiple potential data sources. We also show how to easily enhance tracking systems with the ability to classify or estimate the sizes of objects. As an example, we propose an SNN architecture capable of jointly performing extended object detection, tracking and classification. We compare our architecture to a state-of-the-art Poisson multi-Bernoulli mixture (PMBM) tracker and discuss how SNNs can be applied in a wider set of scenarios.