光学与深度特征融合在机器人场景定位中的应用

Application of optical and depth features fusion on robot scene localization

针对移动机器人室内环境的场景定位问题,研究和提出了一种基于视觉光学与深度特征融合的机器人场景匹配定位算法.首先针对摄像机采集到的光学图像和相应的深度图像信息进行预处理,均匀采样后分别对其进行尺度不变特征变换SIFT的特征提取.然后将2组特征进行特征融合,并利用局部线性编码LLC算法进行特征编码.最后应用线性分类器对场景图像进行分类和匹配,得到场景定位信息.在基于PowerBot移动机器人和微软公司Kinect传感器搭建的机器人实时场景定位系统中,针对设计的算法,进行了实验验证.实验结果显示,提出的算法获得了较高的分类准确率,有效提高了机器人场景定位的工作效率,验证了场景定位算法的高效性和可靠性.

In order to deal with the mobile robot scene localization in indoor environment,a robot scene localization approach based on the fusion of visual optical and depth features is proposed.Firstly,the optical images and the depth images collected by a camera are processed.After sampling images,the scale invariant feature transform（SIFT） is implemented and a feature fusion operation of those tw o features is performed.Moreover,the feature is encoded by using the locality-constrained linear coding（LLC） algorithm.Finally,through the scene matching and classification by using a linear classifier,the scene localization results are obtained.Meanw hile,the proposed algorithm is tested on an experimental platform set up by mobile robot Pow erBot and Microsoft Kinect sensor.The experimental results show that the accuracy of this algorithm is high.By this algorithm,the efficiency of the mobile robot scene localization is improved,proving the efficiency and the reliability of this algorithm.