基于无人机序列图像的全局运动目标检测

Global Motion Target Detection Based on UAV Sequence Images

无人机在执行图像采集任务时,系统与目标之间产生的相对运动会引起序列图像相邻帧坐标系的全局变化,导致常用方法无法准确检测目标。因此,提出了基于无人机序列图像的全局运动目标检测方法。该方法利用时域、空间域与频域的特点以及优势互补原理,选取运动特征、颜色整合特征、亮度特征和频谱残差4个分量用于模型构建,通过设计图像目标区域的判定法则将各分量融合。最后,通过试验对通用图像数据库进行检测并与常用方法进行对比,检测结果与量化分析表明,该方法具有更高的检测精度,可满足无人机运动条件下的目标检测要求。

During the execution of the image acquisition task of the unmanned aerial vehicles(UAV),the relative motion between the system and the target causes the global change of the coordinate system of adjacent frames of the sequence images.The change makes the common methods unable to detect the target accurately.Therefore,a global motion target detection method based on UAV sequence images is proposed.Using the characteristics of time domain,space domain and frequency domain as well as the principle of complementary advantages,four components including the motion feature,the color integration feature,the brightness component feature and the spectral residuals are selected to construct the model.The components are integrated by designing the decision rule of image target area.Finally,with experiments,the general image database is detected and the method is compared with the common methods.The detection results and quantitative analysis show that the method has higher detection accuracy and can meet the target detection requirements under the condition of UAV motion.