三维激光点云运动图像动态目标识别

Dynamic target recognition of 3D laser point cloud moving image

传统点云识别方法针对离散噪声,滤波后仍保留了平均距离较小的邻域点,影响了动态目标识别准确率。为此,提出三维激光点云运动图像动态目标识别方法。移除点云图像场景背景点,分别对离散噪声和一般噪声进行滤波处理,精简并聚类点云数据,赋予激光点运动属性,检测动态目标可通行区域,分割其选定范围,提取目标点云特征点,计算任意两点的点对特征,对目标进行全局描述和局部描述,得到边界特征和局部位置方向,实现目标识别。采集车辆行驶的多帧点云图像,进行对比实验,结果表明,此次方法相比传统方法,提高了车辆识别准确率,同时缩短了点云检测时间,识别结果更加高效可靠。

The traditional point cloud recognition method still retains the neighborhood points with small average distance after filtering,which affects the accuracy of dynamic target recognition. Therefore,a dynamic target recognition method for 3 D laser point cloud moving image is proposed. Remove the background points of the point cloud image scene,filter the discrete noise and general noise respectively,simplify and cluster the point cloud data,endow the laser point motion attribute,detect the passable area of the dynamic target,segment its selected range,extract the feature points of the target point cloud,calculate the point pair features of any two points,describe the target globally and locally,and get the boundary features and local features. Target recognition is realized by the position and direction of part.The results show that this method improves the accuracy of vehicle recognition,shortens the detection time of point cloud,and the recognition result is more efficient and reliable.