基于视觉显著性的航拍车辆检测

摘要

According to the characteristics of the human visuaI system,a vehicIe detection aIgorithm based on characteris-tics saIiency anaIysis is proposed in this paper.First,to obtain a feature response map,the bottom edge of the vehicIe infor-mation is used to do image convoIution on the input image.Then caIcuIate the spectraI residuaI of feature response maps on different scaIes to obtain saIiency response maps.FinaI y,in order to obtain the finaI saIiency map,muIti-scaIes saIiency maps has been merged into one map and temporaI enhancement and spatiaI enhancement method is used.%航拍图像与普通图像相比，具有以下特殊性：分辨率较低，细节不充分；检测目标体积小，难检测；背景的纹理干扰较大。这些特殊性使得直接应用经典的显著性检测方法在航拍图像上不能取得理想的效果。针对这个问题，根据人类视觉系统的特点，提出了一种通过特征显著性分析确定目标候选区域的车辆检测算法。首先，用车辆底层边条信息对输入图像卷积，得到特征响应图。然后计算不同尺度下的特征响应图的谱残差，得到显著性响应图。最后，通过融合多尺度下的显著性响应图，并进行时域和空域增强处理，得到最终的显著性地图。对真实场景下的图像进行仿真实验，结果表明该算法能取得较好的检测效果。