为了提高3维物体目标识别系统的性能及降低计算复杂度,提出一种由粗到细的识别方法。该方法利用深度图像所提供的信息,分两步完成识别过程。首先基于轮廓曲线计算其特征点,并映射到原有轮廓空间,以标志点序列表征原由轮廓进行匹配,在识别初期迅速排除模型库中不相似目标和差异较大的姿态,生成目标候选列表用于精确匹配,以提高识别效率。精确匹配采用一种基于局部区域特征的识别方法,以投票的策略获取最佳结果。局部区域由SIFT算子确定位置和数量,区域特征主要由表面指数和法向量夹角组成,具有平移和旋转不变性。为了更进一步提高效率和降低存储空间,模型库的数据分为轮廓和表面信息两部分,分别以标志位序列和哈希表的形式存储。实验结果表明,该方法具有良好的实时性和识别率,对遮挡和干扰有一定的适应性。

In this paper we propose a two-step approach to recognize free-form objects in range images. First, feature points are calculated based on the contour curve, then mapped to the original shape space. Then a landmark list is determined and used to form a rejection classifier, which quickly eliminates a large number of candidates for an efficient recognition. The remaining free-form objects are then verified using a novel local patch-based matching approach, which is robust to occlusions and noise. The key points are determined based on the scale invariant feature transform (SIFT), and the local surface descriptor is characterized by its two 1D histograms. The two histograms show the frequency of occurrence of the shape index values and the angles between the normal of a key point and that of its neighbors. In order to speed up the retrieval of surface descriptors and save the restore space, the modal data includes shape and surface information. The local surface patched of modal are indexed into a hash table. Verification is performed by running the Iterative Closest Point algorithm. Experimental results with ideal range image are presented to demonstrate the effectiveness, efficiency of the proposed approach.The approach is robust to occlusions and noise.