MOTIA3: Technical report WP3 activities

This document presents the work conducted in Work Package 3 (WP3) of the MoTIA3 project, focused on applying neuromorphic vision sensors and Spiking Neural Networks (SNNs) for object detection, tracking, and classification in industrial environments. Neuromorphic sensors, which operate asynchronously by transmitting only pixel changes as events, offer significant advantages in energy efficiency, latency, and bandwidth requirements compared to traditional frame-based cameras. The Spiking Fusion Object Detector (SFOD) SNN was selected for its state-of-the-art performance, despite not being initially included in the project’s target software libraries. This decision addressed the lack of pre-trained models and neuromorphic data needed for training. The SFOD demonstrated effectiveness in neuromorphic vision tasks, particularly for low-light cone detection using a rover-type robot provided by eSRobotics. Future work aims to enable real-time SNN inference and autonomous driving algorithms within the robot. While neuromorphic vision data from a mosquito sexing application under a Sterile Insect Technique program aligned with state-of-the-art benchmarks, further improvements are required to meet biofactory standards. Additionally, tests with Synsense's Speck hardware and libraries validated its use for spike-based image classification, paving the way for integrating object detection capabilities in future implementations.