Multi-sensor fusion positioning (2-point cloud map construction and map-based positioning) 1-loop detection and code implementation
scan context
Core idea: Under the three-dimensional point cloud matching, the amount of calculation is huge. Based on the three-dimensional matching, the initial value problem cannot be solved. The scan context will be dimensioned to the three-dimensional process, and the scan context is not sensitive to the initial value.
Through two cuts, the point set is changed from three-dimensional to two-digit array
. The smaller the value of d, the higher the similarity between them.
Step1: Combine the two-dimensional arrays into one-dimensional and perform coarse registration (one-dimensional iteration) )
step2: After coarse registration roughly determines the range, perform fine registration (two-dimensional iteration)
to calculate the pose:
1. Find the similar matrix
2. Calculate the angle deviation
3. After the initial value is obtained through Scan Context, perform ICP and NDT
problems : Difficult to find similar specific two frames
LeGO-LOAM
Correction of historical frames is a prerequisite for forming a closed loop
Based on histogram
