基于FastestDet的多目标靶定位方法  

Method of multi-target precision positioning based on FastestDet

作　　者：易磊 易雅雯 YI Lei;YI Yawen(School of Electrical&Electronic Engineering,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学电气与电子工程学院,湖北武汉430074

出　　处：《现代电子技术》2024年第16期97-102,共6页Modern Electronics Technique

摘　　要：为解决无人机在自动多目标靶定位领域计算资源有限、实时性要求高、定位准确率有限等问题,提出一种基于FastestDet的多目标靶定位方法。首先,通过轻量级目标检测算法(FastestDet算法)实现目标靶检测;其次,对所获得的目标靶图像进行透视变换,并截取仅含目标靶的透视变换图像;再对图像进行去噪、灰度化等处理,通过轮廓检测获得目标靶心圆轮廓后,计算透视变换后的靶心坐标;最后,通过逆变换计算原采集图像实际目标靶心坐标。实验结果表明:所提方法采用的目标检测算法网络参数最小,单张图像处理时间最短,占用计算资源少,实时性高,识别准确率达到90.4%,目标靶识别准确率较高;与其他方法相比,该方法的综合定位误差最大可降低92.09%,定位准确率高,可行性强。In order to solve the problems of limited computing resources,high real-time requirements,and limited positioning accuracy of UAVs in the field of automatic multi-target precision positioning,a multi-target precision positioning method based on FastestDet is proposed.The target detection is realized by means of the lightweight target detection algorithm-(FastestDet algorithm).The perspective transformation is conducted on the obtained target image and the perspective transformation image containing only the target image is captured.The denoising,grayscale processing,etc.are performed on the image to obtain the target center circle contour by means of the contour detection,and calculate the target center coordinates after perspective transformation.The actual target center coordinates of the original captured image are calculated by means of the inverse transformation.The experimental results show that the target detection algorithm used by the proposed method has the smallest network parameter,the shortest single image processing time,less computing resources,high real-time performance,and a recognition accuracy of 90.4%.The target recognition accuracy is relatively high.In comparison with other methods,this method can reduce the maximum error of comprehensive positioning by 92.09%,and has high positioning accuracy and strong feasibility.

关 键 词：无人机 目标靶定位 FastestDet算法 目标检测 透视变换 轮廓检测 

分 类 号：TN820.4-34[电子电信—信息与通信工程] TP271.5[自动化与计算机技术—检测技术与自动化装置]