Aim

This project consists of developing robust neuromorphic methods for multiple object tracking

Objectives

1. Identification of  tractable network architectures to discriminate spatio-temporal patterns in spiking data
2. Leveraging (permutation) invariances in multi-object tracking to improve detection and tracking performance
3. Detection of distributed objects within structured and dynamic clutter
4. Comparison of deterministic and stochastic spiking networks for uncertainty quantification within data-driven object trackers

Description

Spiking neural networks (SNNs) can offer increased processing speed and reduced power consumption, especially when implemented on dedicated hardware (neuromorphic chips or FPGAs). Standard SNNs are typically fed with spiking data (e.g., streams of binary values), and the output of each layer remains spiking. Thanks to their ability to capture spatio-temporal patterns, such networks appear as promising candidates for data-driven object detection and tracking tasks, as detection events from sensor networks can be interpreted as asynchronous spikes. Such architectures could be particularly relevant to handle complex motions and dynamic and structured clutter.

While traditional Bayesian tracking methods are particularly valuable for their ability to model and quantify uncertainties, they are often limited by the simplicity of the motion, observation and clutter models used, which can lead to erroneous uncertainties in real-world tracking scenarios. Here we aim at leveraging data-driven strategies within the Bayesian framework to learning dynamics from data and thus provide more realistic uncertainties. Bayesian methods have recently been proposed to train SNNs but the application of these methodological tools to solve real world tracking problems still need to be demonstrated. This is what we intend to achieve in this project.

The findings of this project find applications in a variety of detection and tracking scenarios, from UAVs (radars) to space debris (optical imaging) and seabed inspection (sonar), where reducing false-alarm and mis-detection rates is paramount.