基于深度相机的边缘融合直接位姿估计方法

Edge-fusion-based direct pose estimation method based on depth cameras

针对基于边缘的视觉里程计方法对光度变化敏感、提取边缘不稳定的问题,提出了基于深度相机的边缘融合直接位姿估计方法。融合了传统方法与深度学习的优势进行边缘提取,提高了边缘提取的信噪比;然后将提取的边缘进行基于直接法的相机位姿估计,并使用非参数统计的方法拟合边缘像素残差,给出参与计算的像素权重,使得所提方法对光度变化具有更强的鲁棒性。在TUM数据集上的实验结果表明:该方法取得了较好的位姿估计精度。

The available edge-based visual odometry methods are sensitive to photometric changes and have instability in edge extraction.Aining at this problem,an edge-fusion-based direct pose estimation method based on depth cameras is proposed.This method combines the advantages of traditional methods and deep learning to extract the edge pixels,which improves the signal to noise ratio(SNR)of edge extraction.Then the extracted edge pixels are used in direct camera pose estimation.A nonparametric statistical method is used to fit the residuals of edge pixels and weights are given,which makes the proposed method more robust to photometric changes.The experimental results on TUM dataset show that this method achieves a good precision of pose estimation.