基于kinect深度图像的目标定位与识别

Target Location and Recognition Based on Kinect Depth Image

针对颜色图像受光照及背景影响较大而较难识别目标的问题,利用深度图像作为目标载体,设计并实现了一种简单的目标识别方法。该方法主要包括深度图像的轮廓提取算法和基于轮廓特征识别两部分。先根据提取的轮廓计算矩信息得到图像质心坐标,再根据相机投影公式获得目标质心的真实相机坐标;提取轮廓的矩形描述子,组合轮廓不变矩特征和轮廓形态学特征成为目标的联合特征向量。针对不同的待识别目标,采集足够数量的训练样本,得到各目标的标准特征向量。最后实时提取场景目标的联合特征向量,计算场景目标特征向量与标准特征向量的最小欧式距离,进行识别判断。实验证明了该方法的快速、有效性。该方法应用在装载机械臂的室内移动机器人上,可实现目标定位、识别及抓取操作,具有一定的应用价值。

This paper uses the depth image as a target to design and implement a simple object recognition methods, because it is difficult to recognize the image which is shined by light and its background. The depth contour extraction algorithms and object recog- nition algorithms are used in this method. The moment is calculated according to the extracted contour to obtain the image centroid coordinates and then a real camera coordinates is got based on the camera projection formula. The rectangular contour descriptors are extracted and the contour invariant moments feature and morphological features are combined into a joint eigenvectors. The sufficient data of the training samples is collected to give a standard feature vector to each object. The joint feature vector of the target is extracted in real time and the minimum Euclidean distance between the target feature vector and the standard feature vectors is calculated to give an identification judgment. This method can be used in Mobile Robot loading robot manipulator to locate, identify and grasp the target.