基于关键点的点对特征三维目标识别算法

针对复杂场景下的三维点云目标识别速度慢，准确率低的问题，提出了一种基于关键点的点对特征三维目标识别算法. 通过直接对关键点建立点对特征，避免了周围邻域局部曲面的特征计算，具有空间维度小和计算速度快的特点. 使用哈希表存储，加快了特征匹配的时间. 利用快速投票方案对模型点云和场景点云进行匹配识别，生成候选位姿，利用贪婪算法对候选位姿进行聚类与筛选，采用ICP算法对物体位姿进行优化，基于配准后的点云重叠情况完成目标识别. 对提出的算法在多个数据集以及真实场景下进行了实验，验证了所提出的识别方法具有可行性和有效性，且对噪声的鲁棒性较强，具有一定的实际工程应用价值. 

3D Target Recognition Algorithm Based on Point-Pair Features of Key Points

In order to improve the speed and accuracy of 3D point cloud target recognition in complex scenes, a 3D target recognition algorithm based on the features of the pairs of key points was proposed. Firstly, point-pair features were established for key points to avoid the feature calculation of local surfaces in the surrounding neighborhood, to reduce space dimensionality and to improve calculation speed. Then a hash table was taken as storage to reduce the time of feature matching, a fast voting scheme was utilized to match and identify the model point cloud and the scene point cloud for the generation of candidate position and pose, and a greedy algorithm was used to cluster and filter the positions and poses. And an ICP algorithm was adopted to optimize the object positions and poses, according to the overlapping rate of registered point clouds to evaluate the point cloud recognition. Finally, some experiments were carried out based on data sets and real scenarios to validate the proposed algorithm. The results show that the proposed recognition method can present better feasibility, effectiveness and robustness to noise, possessing certain practical engineering application value.