本文针对小目标检测难题及其实时性提出一种固定翼无人机对远距离车辆目标的自动检测方法.通过对图像进行场景感知和语义分析,在感兴趣区域重点搜索目标,同时提出一种融合梯度响应的显著性检测手段.为有效排除周边背景环境和物体的干扰,提出一种结合无人机位姿信息的目标尺寸预估算法,并采用数学形态学进行疑似目标的筛选和剔除.通过公开数据库及无人机航拍试验,结果表明本方法在复杂背景环境下对地面静止或运动车辆目标的自动检测精度可达63.6％,公路场景下检测精度达80.43％.%In order to solve the issue of small target detection and its efficiency,this paper proposes an approach on automatic vehicle detection for fixed-wing unmanned aerial vehicles (UAV).Region segmentation and recognition provide context information to search small targets in specific area.A saliency detection method combined with gradient response is also proposed.In order to get rid of other objects during vehicle detection,a method which combines both mathematical morphology and UAV attitude information has been proposed.The framework has been tested on a public dataset and aerial images from our fixed-wing UAV.Extensive experiments demonstrate that the proposed detection framework can detect both still and moving vehicle with 63.6％ average precision (AP),and vehicles on road can be detected with 80.43％ AP.
展开▼
