Go-ICP: Solving 3D Registration Efficiently and Globally Optimally
read more
Citations
cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination
Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration
PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet
Registration of large-scale terrestrial laser scanner point clouds: A review and benchmark
A review of cooperative and uncooperative spacecraft pose determination techniques for close-proximity operations
References
A method for registration of 3-D shapes
Shape matching and object recognition using shape contexts
Comparing images using the Hausdorff distance
Efficient variants of the ICP algorithm
Using spin images for efficient object recognition in cluttered 3D scenes
Related Papers (5)
Frequently Asked Questions (11)
Q2. How long did it take to match 1000 data points to the hand?
In their experiment, for example, to match 1000 data points to about 30,000–40,000 model points took about 30 seconds for bunny and 15 seconds for the hand.
Q3. What is the main reason why ICP is so popular?
Partly due to its conceptual simplicity, as well as its good performance in practice, ICP is one of the most popular algorithms for registration, widely used in computer vision, and beyond computer vision.
Q4. What is the way to solve the heuristic problem?
Methods that make use of local invariant shape descriptors (e.g. spin image [19], shape contexts [4], EGI [24]) are mostly heuristic and do not address the optimality issue.
Q5. What is the simplest way to show the convergence of Go-ICP?
To show the convergence of their method and demonstrate the evolution of the bounds, the authors record the upper and lower bound values of the outer BnB when registering the 1000 data points onto the model pointsets in the previous experiment, and plot them as a function of time as shown in Fig.
Q6. What is the simplest way to speed up the computation of inner BnB?
To speed up the computation of inner BnB, the authors set the initial E∗t to be E∗ without loss of globally optimal registration based on the following insight.
Q7. what is the angular distance between two rotations in the underlying manifold?
(4)The second inequality in Eq. (4) means that, the angular distance between two rotations in the underlying manifold, is less than their vector distance in the angle-axis representation.
Q8. What is the funding for this work?
This work was funded in part by Specialized Research Fund for the Doctoral Program of China (20121101110035), and ARC Discovery grants: DP120103896 and DP130104567.
Q9. What is the purpose of this experiment?
In this experiment, the authors use the Stanford bunny raw scan data1 and a dense hand mesh2 shown in the last two rows of Fig. 5 to test out the real-life efficiency.
Q10. How do the authors test the running time of their method?
The authors test the running time of their method on different numbers of data points (i.e. M ) by sub-sampling the original data, while the initial poses are fixed.
Q11. How does the algorithm solve the problem of the local minima?
To alleviate the local minima issue, previous work has attempted to enlarge the basin of convergence by smoothing out the objective function.