基于三目共光路图像配准的运动目标融合检测方法研究

The main aim of this project is to improve the performance of moving object detection in the presence of dynamic background scenes. On the basis of infrared and visible image registration which is implemented by common-path with trinocular vision, moving object detection of depth, infrared and visible information fusion is studied. Initially, a model of infrared and visible image registration is designed to simulate the registration result of common-path by using a framework of separate optical axes. What's more, with in-depth analysis of aberration between infrared and visible image, a non-rigid image registration method which can reveal the law of global aberration is proposed to achieve image registration. Finally, by using the model of image registration, depth image is obtained, and a moving object detection method based on depth, infrared and visible information fusion is proposed. The project provides a complete solution of fusion detection, including the registration model of ray-path and the image processing method, in order to improve the efficiency and robustness of moving object detection in the presence of dynamic background scenes by making full use of the complementary information of multi-source images.

针对动态复杂背景下运动目标检测困难的问题，采用三目共光路图像配准方法实现红外与可见光图像自动配准，进而开展基于深度、红外与可见光多源信息融合的运动目标检测研究。本项目首先提出模拟共光路的红外与可见光三目视觉图像前端配准模型，使分立光路具备共光路的成像匹配规律；然后根据配准模型分析红外与可见光图像之间形变的成因，提出可以表征图像全局形变规律的图像非刚性配准方法，以提高红外与可见光图像自动配准的效率和鲁棒性；最后利用前端配准模型引入场景深度估计信息，提出基于深度、红外与可见光多源信息融合的运动目标检测方法，实现动态复杂背景下在线的运动目标检测。本项目从前端配准模型到后端数据处理方法，给出一套运动目标融合检测解决方案，旨在结合三目视觉与非刚性图像配准方法，获得稳健的红外与可见光图像配准效果，在此基础上充分利用多源信息互补特性，提高动态复杂背景下运动目标检测的精度与效率。

多源图像融合是机器视觉领域的重要研究内容之一，具有广泛的应用领域和很大的实用价值。尽管图像融合相关方法和理论已取得很多成果，但是尚未从技术实现角度给出完备的数字图像融合解决方案。针对这一问题，本项目基于非刚性图像配准技术实现了多源图像融合的工程化应用。提出了一系列红外与可见光图像非刚性配准方法，利用增强仿射变换模型表征待配准图像之间的全局形变规律，提高图像配准精度，降低计算复杂度。探索了多源图像配准和实时融合的工程化实现方法，研制出了日盲紫外与可见光图像融合系统、以及红外、紫外与可见光三目多源图像融合系统。基于本项目成果所研制的日盲紫外图像融合系列相机已是批量生产的成熟产品，广泛应用于电力设备故障检测和预警领域，有效推动了数字图像融合技术的产业化应用。