Abstract:Modern cars are equipped with both active and passive sensor systems that can detect potential collisions. In contrast, locusts avoid collisions solely by responding to certain visual cues that are associated with object looming. In neurophysiological experiments, I investigated the possibility that the 'collision-detector neurons' of locusts respond to impending collisions in films recorded with dashboard cameras of fast driving cars. In a complementary modelling approach, I developed a simple algorithm to reproduce the neuronal response that was recorded during object approach. Instead of applying elaborate algorithms that factored in object recognition and optic flow discrimination, I tested the hypothesis that motion detection restricted to a 'danger zone', in which frontal collisions on the motorways are most likely, is sufficient to estimate the risk of a collision. Furthermore, I investigated whether local motion vectors, obtained from the differential excitation of simulated direction-selective networks, could be used to predict evasive steering maneuvers and prevent undesired responses to motion artifacts. The results of the study demonstrate that the risk of impending collisions in real traffic scenes is mirrored in the excitation of the collision-detecting neuron (DCMD) of locusts. The modelling approach was able to reproduce this neuronal response even when the vehicle was driving at high speeds and image resolution was low (about 200 × 100 pixels). Furthermore, evasive maneuvers that involved changing the steering direction and steering force could be planned by comparing the differences in the overall excitation levels of the simulated right and left direction-selective networks. Additionally, it was possible to suppress undesired responses of the algorithm to translatory movements, camera shake and ground shadows by evaluating local motion vectors. These estimated collision risk values and evasive steering vectors could be used as input for a driving assistant, converting the first into braking force and the latter into steering responses to avoid collisions. Since many processing steps were computed on the level of pixels and involved elements of direction-selective networks, this algorithm can be implemented in hardware so that parallel computations enhance the processing speed significantly.

All‐Optically Triggered In‐Sensor Collision Detection and Warning Based on 2D Complementary Material Devices

A low-power biomimetic collision detector based on an in-memory molybdenum disulfide photodetector

A Multiple Vehicle Sensing Approach for Collision Avoidance in Progressively Deployed Vehicle Networks.

Collision Selective Visual Neural Network Inspired by LGMD2 Neurons in Juvenile Locusts

LGMD and DSNs Neural Networks Integration for Collision Predication.

A New Strategy to Improve the Performance of MoS2-based 2D Photodetector by Synergism of Colloidal CuInS2 Quantum Dots and Surface Plasma Resonance of Noble Metal Nanoparticles

A Visual Neural Network for Robust Collision Perception in Vehicle Driving Scenarios

A Biomimetic Visual Detection Model: Event-Driven LGMDs Implemented With Fractional Spiking Neuron Circuits

Robustness of Bio-Inspired Visual Systems for Collision Prediction in Critical Robot Traffic

Enhanced Visible to Near-Infrared Photodetectors Made from MoS2-based Mixed-Dimensional Structures

Ultra-low power neuromorphic obstacle detection using a two-dimensional materials-based subthreshold transistor

Multiframe-integrated, In-Sensor Computing Using Persistent Photoconductivity

O-LGMD: An Opponent Colour LGMD-Based Model for Collision Detection with Thermal Images at Night.

An Ultrasensitive Bi2O2Se/In2S3 Photodetector with Low Detection Limit and Fast Response toward High-Precision Unmanned Driving

Simplified bionic solutions: a simple bio-inspired vehicle collision detection system

A directionally selective collision-sensing visual neural network based on fractional-order differential operator

An insect vision-inspired neuromorphic vision systems in low-light obstacle avoidance for intelligent vehicles

Enhancing LGMD-based model for collision prediction via binocular structure

A Novel Cppm Anti-Crosstalk Collision Avoidance Lidar with Ultra-Low Laser Power

Simulating the obstacle avoidance behavior day and night based on the visible-infrared MoS2/Ge heterojunction field-effect phototransistor

Simulating the Obstacle Avoidance Behavior Day and Night Based on the Visible-Infrared MoS 2 /ge Heterojunction Field-Effect Phototransistor