模拟复眼视叶神经网的目标运动方向检测模型  

Detection model of target motion direction simulating the optic lobe neural network of compound eyes

作　　者：徐梦溪[1] 施建强 郑胜男[1] 韩磊[1] XU Mengxi;SHI Jianqiang;ZHENG Shengnan;HAN Lei(School of Computer Engineering,Nanjing Institute of Technology,Nanjing 211167,China;School of Energy and Power Engineering,Nanjing Institute of Technology,Nanjing 211167,China)

机构地区：[1]南京工程学院计算机工程学院,江苏南京211167 [2]南京工程学院能源与动力工程学院,江苏南京211167

出　　处：《智能系统学报》2024年第3期546-555,共10页CAAI Transactions on Intelligent Systems

基　　金：江苏省自然科学基金项目(BK20221399).

摘　　要：如何对杂乱背景中物体(目标)的运动方向做出准确可靠的检测与感知,是计算机视觉研究领域中一个重要问题。自然界中,飞虫(如苍蝇、蜻蜓等)高适应性和高可靠性的感知目标运动是一种自然特性,本文基于飞虫−果蝇视叶神经纤维网最新的生理学研究成果,提出一种基于果蝇视觉感知目标运动方向的多层级检测模型系统。通过对不同场景下拍摄的视频序列样本进行实验和测试,并与2-Q运动检测器模型、基于ON和OFF信号通道处理运动信息的检测模型等进行了对比,验证了其在杂乱背景下对于目标水平和垂直方向运动检测的有效性和鲁棒性。How to detect and perceive the motion direction of objects(targets)in a chaotic background accurately and reliably is an important research area in the field of computer vision.In nature,flying insects(such as flies and dragonflies)can sense the movements of their target with high adaptability and reliability.Based on the latest research results of the physiology of the optic lobe neuropile layer of the flying insect drosophila,a multilevel model system for detecting target motion direction based on visual perception of drosophila is proposed.By shooting the video sequence of samples under different scenarios for experiment and test and comparing with the 2-Q motion detector model based on ON and OFF signal channel processing motion information,the effectiveness and robustness of the proposed system for horizontal and vertical motion detection of video targets under a chaotic background are confirmed.

关 键 词：视频 目标检测 运动方向检测 昆虫复眼 神经计算 人工神经网络 多层级模型 视叶神经网 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]