一种有效的基于点云特征人体识别方法

基于视觉信息的运动目标识别是室内移动机器人自主导航的一个重要研究内容,现有的方法经常由于背景噪声干 扰、自然光照、机器人视角变化等因素影响识别稳定性。本文提出一种有效的目标识别方法，滤除地平面点云特征，并采用 分割聚类和支持向量积融合方法识别运动目标。实验表明，本文方法在满足实时性前提下，相对于传统人体骨骼、二维梯度 直方图算法具有更高的准确率，更适合复杂环境下的人体识别。

An Efficient Method of Human Recognition based on Point Cloud

The recognition of moving objects that based on visual information is an important aspect of autonomous navigation for indoors mobile robot. The existing methods have influence on the stability of human recognition because of the background noise, natural lighting, the change of the robots views. This paper presents an efficient object recognition method. This method recognizes and filters the point cloud of ground plane, and fuses the methods of clustering segmentation and support vector machine to recognize moving objects. The experiments show that this method, compared with the algorithm of human bones recognition or two dimensional histogram of oriented gradient, has the higher accuracy, and is more suitable for the recognition of human in the complex environment, meeting the real-time requirement.