In man–machine cooperative assembly, assembly recognition that determines the current manual working stage is key information to driving automatic computer-aided assistance. Focused on three features of the assembly scene—movable view, process stage, and CAD model template— a view-free and marker-less assembly stage recognition method is proposed in this paper. By constructing the semantic model for the assembly scene and the stage model for CAD parts, a depth image of assembly and a CAD model can be extracted as point clouds. Then we propose the segmented projection contour descriptor to uniformly express the shape information as a series of contours, so the 3D registration issue is converted to a 2D registration issue. The vertex-to-edge Hausdorff distance is proposed in the partial registration to determine the transformation matrix for each pair of contours. Finally, the overall matching algorithm based on the overlay ratio is given, and the best matching stage model indicates the current assembly stage. The recognition and classification experiments are carried out to verify the proposed method. A comparison with traditional Hausdorff distance proves the proposed algorithm performs better in stage recognition. Our study reveals that the proposed view-free and marker-less method can solve the stage recognition issue based on the assembly’s depth image, so as to connect the on-site assembly with the digital information.

在人机合作装配中，确定当前手动工作阶段的装配识别是驱动自动计算机辅助的关键信息。针对装配场景的三个特征-可移动视图，过程阶段和CAD模型模板，提出了一种无视点和无标记的装配阶段识别方法。通过构建装配场景的语义模型和CAD零件的阶段模型，可以将装配的深度图像和CAD模型提取为点云。然后，我们提出了分段投影轮廓描述符，将形状信息统一表示为一系列轮廓，因此将3D注册问题转换为2D注册问题。在部分配准中提出了顶点到边缘的Hausdorff距离，以确定每对轮廓的变换矩阵。最后，给出了基于覆盖率的整体匹配算法，最佳匹配阶段模型指示了当前的装配阶段。进行识别和分类实验以验证所提出的方法。与传统的Hausdorff距离的比较证明了该算法在阶段识别中表现更好。我们的研究表明，提出的无视点和无标记的方法可以解决基于装配体深度图像的阶段识别问题，从而将现场装配体与数字信息联系起来。最佳匹配阶段模型表示当前的组装阶段。进行识别和分类实验以验证所提出的方法。与传统的Hausdorff距离的比较证明了该算法在阶段识别中表现更好。我们的研究表明，提出的无视点和无标记的方法可以解决基于装配体深度图像的阶段识别问题，从而将现场装配体与数字信息联系起来。最佳匹配阶段模型表示当前的组装阶段。进行识别和分类实验以验证所提出的方法。与传统的Hausdorff距离的比较证明了该算法在阶段识别中表现更好。我们的研究表明，提出的无视点和无标记方法可以解决基于装配体深度图像的阶段识别问题，从而将现场装配体与数字信息联系起来。