Research On Uav Looming Detection Based On Bionic Vision

With the continuous innovation of artificial intelligence,machine vision and sensors,the application areas of UAVs are becoming more extensive and diverse,such as agricultural production,rescue and disaster relief,power inspection and military fields.Since the actual operating environment of UAVs is complex and varied,with various obstacles,obstacle detection and avoidance is of great importance to UAV operations.Among the visual obstacle avoidance solutions for UAVs,the insect bionic vision-based imminent collision detection algorithm has unique advantages such as low power consumption,high response speed and universal environmental adaptability.This paper presents an in-depth study of the locust and fruit fly visual nervous systems and their artificial visual network models,and proposes a looming detection algorithm from a computer vision perspective that incorporates locust LGMD(Lobula Giant Movement Detector)and Drosophila’s LPLC2(lobula plate/lobula columnar,type II)neurons.A simulated and actual UAV flight platform was built to verify the effectiveness of the algorithms.The main work of this paper can be summarized as follows:1.Radially symmetric motion is studied in detail from the perspective of motion vision.A mathematical definition of radially symmetric motion and a criterion for its determination are proposed for the first time,and a metric for measuring the degree of radial symmetry of image motion is also proposed.At the same time,a preliminary exploration of the application of radially symmetric motion in looming detection is made.2.A method for modelling LPLC2 neurons based on the principle of lateral inhibition is proposed,mimicking the structure of their four synaptic fans.Numerical experiments show that the computational model reproduces well the directional selectivity and sensitivity to radially symmetric motion of LPLC2 neurons.3.A looming detection algorithm is proposed that incorporates locust LGMD and Drosophila LPLC2 neurons.The model consists of two signal processing pathways,one mimicking LGMD for filtering image velocity information and the other mimicking LPLC2 for extracting image symmetry information,and the two pathways are passed through an enhancement module based on a self-attention mechanism to obtain the final output.The sensitivity and robustness of the looming detection algorithm are verified with simulated and real scenes.4.The UAV simulation and actual flight platform was built to verify the looming detection and obstacle avoidance algorithm proposed in this paper.The UAV flight simulation platform was built using a combination of Air Sim and ROS,and the feasibility of the looming detection algorithm proposed in this paper was verified through experiments.A quadrotor UAV flight experimental platform was built to verify the feasibility of the looming detection algorithm.