机载复杂遥感场景下特定建筑区检测跟踪算法

摘要

Automatic identification and tracking for specific building areas is not only of great significance in airborne remote sensing video data processing,but also one of the key technologies for task-oriented non-hovering airborne platform to search ground and boost fire control system.This paper proposes an identification and tracking algorithm of specific build-ing areas for real complex application scenes.First,to extract local descriptors for offline reference images with target areas. At the stage of target area online identification,in order to ensure the timeliness,a fast SIFT (Scale Invariant Feature Trans-form)feature extraction method based on significant edges is proposed to extract local descriptors for suspected areas in large field of view images to be detected.In addition,the hierarchical feature point matching method is designed to achieve a highly reliable identification of target areas.At the stage of online tracking for target area,adaptive window and trajectory predicting technologies can effectively delineate suspected target areas,and within the limited range to recognize the final tar-gets and track them.Real-measured data simulation results show that the proposed algorithm in complex scene conditions can realize the quick identification and stable tracking for a particular building area.It can also provide a key technical support for the future application of real systems.%特定建筑区域的自动检测及跟踪在机载遥感视频数据处理中具有重要意义，是非悬停机载平台面向任务对地搜寻及火控系统引导的关键技术之一。针对实际复杂应用场景，提出一种特定建筑区检测跟踪算法。首先，将离线获得的目标区参考图进行局部描述子特征提取。目标区在线检测阶段，为保证时效性，提出基于边缘显著性的快速SIFT （Scale Invariant Feature Transform）特征提取方法，对待检测大视场图像中疑似建筑区域提取局部描述子特征；并通过设计层次化的特征点匹配方法实现目标区的高可靠检测。目标区在线跟踪阶段，通过自适应开窗及轨迹预测技术有效圈定疑似目标区，并在圈定的有限范围内进行最终目标确认及稳定跟踪。实测数据仿真结果表明，本文提出的算法能在复杂场景条件下，实现对特定建筑区域的快速检测及稳定跟踪。可为将来实际系统应用提供关键技术支撑。