自然景物中树的远景特征识别的研究

Feature recognition of distantly-viewed trees in natural scenery

在分析了目前视觉系统中两种实用运动检测方法——帧间差法和背景差法的基础上,指出树的特征识别的重要性.通过分析远景中树的图像特征,提出了远景中树的特征模型建立和相应的匹配算法.本文提及的树特征识别方法可以减少运动检测计算量并有助于对场景的粗略定位,提及的特征提取方法对具有重复纹理特征的结构建模具有普遍意义.最后,试验结果证明了该方法的可靠性和可行性.

In this study, the importance of tree feature recognition is first presented via two categories of detection methods in a vision system, i.e. frame and background difference methods. By analyzing image features of trees in a distant view, relevant models and matching algorithms are then developed for tree features. Subsequently, this method could facilitate in the calculation workload reduction of motion detection algorithms, initial orientation of natural scenery, and structural modeling of repeated texture features. Based on experimental results, this approach has finally been proven reliable and feasible.