The Optimization of Object Detection Framework for Event Flow Based on Spiking Neural Network and Spatial Attention Mechanism

Event cameras are a new type of neuromorphic sensor that has been gaining increasing attention. Unlike traditional frame-based cameras, they operate and output data in a completely different manner. Inspired by the human visual system, event cameras capture changes in light intensity at each pixel using photosensitive elements and output discrete event signals at the corresponding pixel locations. They offer advantages such as low latency, high dynamic range, and minimal sensitivity to lighting conditions. The Spiking Neural Network (SNN) is a type of neural network model that mimics the human nervous system by transmitting information through spikes between neurons, which aligns well with the operation of event cameras. Despite the advantages of SNN in terms of speed and power consumption, training deep SNN models for complex tasks is typically challenging, limiting their performance in current implementations of object detection using event cameras. In this paper, we introduce a spatial attention mechanism and optimize the Single Shot MultiBox Detector(SSD) Head based on the SSD object detection algorithmic framework. The optimized model demonstrates good performance on the complex Genl event stream car detection dataset, achieving a mAP of 32.5%.