基于变形模板的多目标识别与定位

Multiple Target Recognition and Location Using Deformable Templates in Downward Looking Image

典型目标的识别与定位是无人机应用系统的关键技术之一,但实现完全地自动目标识别和定位存在一定的困难。该文提出一种采用变形模板进行多个目标的识别和定位方法,以机载传感器的融合图像为基础,并引入人在回路的思想,设计了基于变形模板的多目标识别和定位算法,重点研究了多目标变形模板库的构造和相应能量函数的定义,而变形模板的优化采用具有保优策略的遗传算法实现。仿真实验表明该方法具有较高的目标识别与定位能力,可以有效解决无人机图像中多个典型目标的识别与定位问题。

The typical target recognition and location is a key technique of the application system for Unmanned Aerial Vehicle （UAV）, but there exists a certain difficulty in entirely automatic target recognition and location. A new method to recognize and locate multiple targets is presented in this paper. Through introducing the deformable template technique and the man-in-the-loop thought, this paper designs the multiple target recognition and location algorithm for the fused downward-looking image from UAV. The emphases of the method are laid on the construction of different deformable templates, the definition of corresponding energy functions, and the optimal search strategy. Results show that the technique of multiple target recognition and location using deformable templates is more effective and can accurately recognize and locate multiple typical targets.