Brief introduction
This blog introduces kinectfusion in the ICP algorithm code, code implementation is the PCL Engineering Pcl_gpu_kinfu_large_scale project file ESTIMATE_COMBINED.CU.
The ICP algorithm can greatly improve the computational efficiency by doing parallel computing with the GPU. The objective function in the GPU minimization ICP algorithm
Kinectfusion in the ICP using the minimum point to the distance between the plane, the calculation of two frames between the position and attitude transformation, ICP algorithm theory introduction See blog:
http://blog.csdn.net/fuxingyin/article/details/51425721
The ICP algorithm calculates the pose transformation between two frames by solving cx=b cx=b, where the expressions of C C and B are:
c=∑i=1k (Pixnin1) ((Pixni) ⊤n⊤i). \begin{equation} c=\sum\limits_{i=1}^{k}\left (\begin{array}{c c c} p_i\times n_i \ n_1\\ \end{array} \right) \left (\ Begin{array}{c C} (p_i\times N_i) ^\top & n_i^\top \ \end{array} \right). \end{equation}
b=∑i=1k ((pixni) tau, Ni) τ⋅ ((PI−QI) τ⋅ni) \begin{equation} b=\sum\limits_{i=1}^{k}{{{({{{{p}_{i}}\times {n}_{i}})}^{\ Tau}},\ {{n}_{i}})}^{\tau}}}\cdot ({{({{p}_{i}}-{{q}_{i}})}^{\tau}}\cdot {{n}_{i}}) \end{equation}
Where Pi P_i is the current point cloud, Qi Q_i is previous point cloud, ni