| Spatially encoded structured light system(SLS)has many advantages and has been widely applied in many fields.Conventional spatial SLS methods usually adopt the color as coding features,and that makes it lack of robustness to deal with the targets with plentiful color,texture or surface reflections.In this proposal,a novel binary encoded structured light method is investigated.Some binary geometrical elements are used to construct the structured light pattern based on epipolar constraint theory.Compared with traditional color pattern features,binary geometrical features are more insensitive to surface color or reflections.Based on the calibration method of Zhang,a projector calibration method based on the principle of projective transformation is proposed.By defining the intersection of each two adjacent pattern elements as the pattern feature point,a local structure-based feature detection algorithm is proposed for its robust detection and precise localization.Based on the extracted feature points,a topological network can be established to separate the geometrical pattern elements from the structured light image.In the decoding stage,to recognize the separated pattern elements accurately,a decoding algorithm based on convolutional neural network(or CNN)is studied.To make the recognition algorithm more robustly and accurately,an extensive training set should be established firstly.The training samples can be collected from a variety of target surfaces under various lighting conditions.Based on the triangulation principle of structured light system,an algorithm of 3D point cloud reconstruction implemented.The experimental results demonstrate that the robustness of binary shape-coded structured light can be definitely improved with the proposed binary shape coding strategy,template-convolution-based feauture detection algorithm and convolutional neural network-based decoding method.And these proposed techniques make spatial SLS applicable for more real applications with dynamic 3D information requirement. |
