全景聚焦合成孔径成像及其遮挡目标提取研究

High performance object detection and situational awareness under conditions of complex scene with occlusion is one of the key challenges of existing visual detection field. Starting from the innovation of synthetic aperture imaging method, this proposal will address on all-in-focus synthetic aperture imaging combining the object extraction and synthetic aperture imaging. The proposed method will extend visibility of traditional synthetic aperture imaging from focus region to out-of-focus region, and will generate the clear see through imaging result not only in focus region but also in out-of-focus region while addressing the occluded object extraction problem. The main contents are outlined as follows. (1) Out-of-focus background reconstruction method will be researched to solve the problem of defocusing blur in the out-of-focus region of synthetic aperture imaging. (2) Dynamic focus depth estimation based see through synthetic aperture imaging will be researched to solve the problem of refocusing see through imaging caused by object motion. (3) The sharpness function labeling based synthetic aperture object automatic extraction method will be researched to solve the problem of synthetic aperture object extraction in unconstrained environment. This project has important theoretical and application value in the fields of intelligent video surveillance, military detection and anti-terrorism surveillance.

复杂场景与遮挡条件下目标探测与环境感知是当前视觉探测领域的核心难题之一。本项目从合成孔径成像方式的革新入手，研究将合成孔径清晰成像与目标提取统一进行处理的全景聚焦合成孔径成像方法。该方法将传统合成孔径清晰成像范围扩展到焦平面外，在获得聚焦和非聚焦区域均清晰成像的透视成像结果的同时解决遮挡目标提取问题。主要内容包括：（1）针对合成孔径成像时非聚焦区域散焦模糊问题，研究合成孔径非聚焦背景重建方法；（2）针对目标运动导致的无法快速聚焦透视成像问题，研究基于动态聚焦深度估计的合成孔径目标透视成像方法；（3）针对无约束环境下合成孔径目标的提取问题，研究基于清晰度标识的合成孔径目标自动提取方法。本项目的研究在智能视觉监控、军事探测、反恐侦察等多个领域具有重要的理论和应用价值。

本项目将传统合成孔径清晰成像范围扩展到焦平面外，在获得聚焦和非聚焦区域均清晰成像的全景聚焦合成孔径成像结果的同时解决遮挡目标提取问题。依托本项目，构建了高精度高速同步采集与全聚焦合成孔径成像实验系统。提出了包括：全聚焦合成孔径成像及其目标提取方法；基于卷积神经网络的合成孔径聚焦深度评估；基于非刚体轨迹基的线性自标定；基于合成孔径成像的三维重建等一系列围绕相机阵列标定、全聚焦合成孔径成像、提取及应用方法。本项目的研究成果可以广泛应用于军事与反恐中目标探测、战场环境侦测等领域，将为我国开展相关领域的研究工作提供借鉴和参考。