Object Pose Estimation Based on Multi-Layer Feature Fusion and Hybrid Attention
In the process of industrial robot grasping,aiming at the problem of 6D pose estimation accuracy under the condition of no texture,occlusion and chaotic scene,a 6D pose estimation algorithm for objects that mixes spatial channel attention and multi-layer feature fusion is proposed.A vertically connected bidi-rectional feature fusion pyramid network is designed to realize multi-layer feature fusion and improve the detection performance of target key points.Embedded in the hybrid spatial channel attention mechanism,fo-cusing on the feature information in the two dimensions of space and channel,and enhancing the local rep-resentation ability of the model.Experimental results on LineMod dataset and Occlusion LineMod occlusion dataset show the superiority of the proposed algorithm,and can effectively deal with background clutter and occlusion problems.
NETL
NSTL
万方数据
