基于机器学习的NAO机器人检测跟踪

Detection and Tracking of NAO Based on Machine Learning

为了解决对人形NAO机器人的检测跟踪问题,提出了一种机器学习与特征匹配相结合的方法。向梯度直方图(Histogram of Oriented Gradient)特征是一种在计算机视觉和图像处理中用来进行物体检测的特征描述子,在行人检测中取得了较好的效果。将其应用于人形NAO机器人的检测跟踪,并结合Ada Boost算法通过机器学习的方法,从大量的训练样本中自动抽取HOG特征并建立级联分类器,利用分类器找出视频帧中含有机器人目标的区域,并在此基础上利用SURF(Speed Up Robust Features)特征匹配方法与模板图像进行特征匹配,以提高目标识别的正确率。实验结果表明,该方法对NAO机器人在室内光线无遮挡的情况下取得了稳定的跟踪效果。

To detect and track the NAO robot, a method of combing machine learning and feature matching was pro-posed. Histograms of Oriented Gradient （HOG） feature is a descriptor which is used for object detection in computer vision and image processing. HOG descriptors outperform existing feature sets for human detection. The HOG feature sets was used to detect the NAO robot. A method of combining AdaBoost algorithm and HOG features created a cas-cade classifier. The cascade classifier was used to detect the NAO robot. In order to improve the right rate of recogni-tion, SURF （Speed Up Robust Features） algorithm was used to compare the region of interest and the template, the region of interest was obtained by the cascade classifier. By matching the two pictures,to determine whether the region of interest contains the target. In the experiment, two NAO robots were tracked. The experimental results show that the tracking and recognition of NAO performed steadily on indoor lights without any shelter conditions.