复杂动态场景下基于惯导信息辅助的红外弱小移动目标检测技术研究

Infrared small target detection under complex background is a challenging task in infrared image processing field. In order to deal with the challenging caused by small target size, complicated background and dynamic scenes change for infrared detection systems on aerial moving platforms, this project researches the technology of INS-aided infrared small moving target detection methods under complicated dynamic background, based on the study of the full strapdown imaging seeker in recent years. As to the INS with small errors, we plane to analysis the relationships between the inertial navigation information and image, and to construct the scenes change model caused by moving platforms. We formulate the background suppression and target enhancement as the patch-based matrix decomposition problem. The infrared small moving target is detected by adaptive segmented methods. To eliminate drift of the INS, we plane to study the theory between the incremental inertial navigation information and image, and to construct the scenes change model in consecutive frames. We formulate the small target detection problem as the selection mechanism of interesting points. The infrared small moving target is detected via suppressing the background points. At last, a series of experiments and evaluations demonstrate the robustness of the proposed methods. The research results will provide the new ideas and methods for the detection of infrared small target, and an in-depth study of the technology has crucial theory meaning and practical values.

红外弱小移动目标检测是当前红外图像处理领域的研究难题之一。针对空中运动载体下红外检测系统面临的目标弱小、背景复杂、场景动态变化等挑战因素，本项目以当前在全捷联成像导引头领域所展开的工作为基础，研究复杂动态场景下基于惯导信息辅助的弱小移动目标检测技术。针对误差较小的惯导系统,研究惯导信息与图像的融合理论，建立载体运动引起的场景变化模型，通过背景差分消除场景的动态变化，将杂波抑制和目标增强转化为基于块图表示的矩阵分解问题，通过分割实现对弱小移动目标检测；为消除惯导系统的漂移误差，研究增量惯导信息与图像的融合理论,并建立连续帧之间的载体运动引起的场景变化模型，将弱小移动目标检测问题转化为特征点筛选问题,通过背景特征点的抑制实现对弱小移动目标的检测；最后通过全面的实验来验证算法的有效性。本项目的研究成果将为红外弱小目标检测提供新思路、新方法，具有极为重要的理论意义和实际的工程应用价值。

红外弱小移动目标检测是红外探测系统中近几年的研究热点，尽管众多学者对红外弱小移动目标检测进行了大量的研究，但在机载复杂背景下的检测仍然面临着诸多的挑战和困难。本项目将惯导信息与图像信息相融合，研究基于惯导信息辅助的红外弱小目标检测算法，为弱小移动目标检测技术提供新思路、新方法。重点研究惯导信息与图像的融合算法，实现对图像的有效校正，消除飞行器运动和姿态变化引起的图像变形；设计基于背景差分的红外弱小移动目标检测算法，提高复杂场景下的运动目标检测能力；研究增量惯导信息与图像的融合算法，实现对目标位置的准确预测，也为后续通过抑制背景特征点来检测运动目标提供基础；构建了一种全新的检测-跟踪-检测框架，将弱小目标检测问题转化为候选特征点的筛选问题，并对候选特征点的检测提取、候选特征点的跟踪和离群目标点的检测等三个核心问题进行了深入研究；最后构建了基于旋翼飞行器的空地红外目标检测试验平台，为算法的研究提供平台支撑。项目组已发表30篇论文（其中SCI检索论文11篇），申请6项发明专利（其中3项已获授权），授权4项软件著作权，出版学术专著1部。