摘要
为了实现无人机(UAV)航拍图像中多运动目标的实时检测与识别,将静止目标和运动目标分别定义为背景和前景,利用图像稳化技术将航拍图像序列中的每帧与相邻帧对齐,克服UAV飞行动作对摄像机转动拍摄图像的影响;选取图像中的行人和车辆作为前景,分别使用哈尔(Haar-like)特征和级联分类器对图像中的目标进行检测和识别;利用密集光流计算两幅连续图像的运动矢量,从而区分静止目标(背景)和运动目标(前景),最终图像结果仅保留运动目标所在区域;将文章方法用于DroneVehicl航拍数据集实验,每秒平均帧数达到47.08 fps,检测精度为94%,并且表现出较高的召回率和F统计量,可达到实时检测与识别效果。
In order to realize real-time detection and recognition of multiple moving targets in aerial images of unmanned aerial vehicle(UAV),the stationary target and moving target were defined as background and foreground respectively,and the image stabilization technique was used to align each frame in the aerial image sequence with the adjacent frame,so as to overcome the influence of UAV flight movement on the image captured by camera rotation.The pedestrian and vehicle in the image are selected as the foreground,and the haar-like feature and cascade classifier are used to detect and recognize the targets in the image respectively.The dense optical flow is used to calculate the motion vectors of two continuous images,so as to distinguish the stationary target(background) and moving target(foreground),and the final image results only retain the region of the moving target.The proposed method was applied to the DroneVehicl aerial data set experiment,and the average frame per second reached 47.08 FPS,the detection accuracy was 94%,and showed high recall rate and F statistics,which could achieve real-time detection and recognition effect.
作者
栾桂芬
LUAN Guifen(Taizhou electromechanical higher vocational and technical school,Department of Information Engineering,Taizhou 225300,China)
出处
《计算机测量与控制》
2022年第1期221-228,共8页
Computer Measurement &Control
关键词
无人机
运动目标
图像稳定
目标检测
密集光流
UAV
Moving target
Image stabilization
Target detection
Dense optical flow
