Diagonal Region Division-Based Fly Neural Network on Omnidrectional Collision Detection

Abstract

Camera-based road vehicle collision detection is a major challenge in the field of intelligent transportation, particularly it is still open to borrow motion sensitive neurons to construct computational models for multi-vehicle collision detection. To fill this gap, a bio-inspired fly visual collision detection neural network with presynaptic and postsynaptic neural networks is proposed to execute vehicle collision early warning in complex scenes. The former network includes four sub-neural networks which share four visual neural layers, each with a specific visual neuron; the latter network involves in one lobula plate layer and three spiking neurons. The experimental results have validated that the fly neural network can successfully execute collision detection when confronted with some approaching object(s) in real time.