分析比较了Intel推出的RealSense摄像头,与热门的Kinect摄像头之间的异同.针对RealSense没有全平台点云库支持的问题,给出基于Librealsense的数据获取转换流程.针对三维物体识别算法实时性较差的问题,根据目标物体颜色空间的特性,提出了改进的快速点特征直方图描述符算法.新算法利用目标物体的HSV颜色空间特征,提升了描述符间匹配的准确率,同时利用物体色调位图降低了场景描述符计算量.除此以外,利用综合滤波的方式,显著地提升了图像的有效信息量.%This paper analyzes the similarities and differences between Intel's RealSense cameras and the popular Kinect cameras.In order to solve the problem that RealSense does not support the whole platform of point cloud library (PCL),the data acquisition conversion process based on Librealsense is presented.Aiming at the problem of poor real-time performance of 3D object recognition algorithm,an improved fast point feature histogram (FPFH) descriptor is proposed according to the characteristics of target object color space.The new algorithm,uses the HSV color space feature of the target object to improves the precision of the descriptor matching and reduce the computational complexity of the scene descriptor via hue bitmap.In addition,by using the comprehensive filtering method,the effective amount of information is significantly increased.
展开▼
