Recognizing Objects in Range Data Using Regional Point Descriptors
read more
Citations
VoxNet: A 3D Convolutional Neural Network for real-time object recognition
Behavior recognition via sparse spatio-temporal features
FAB-MAP: Probabilistic Localization and Mapping in the Space of Appearance
Unique signatures of histograms for local surface description
Automatic reconstruction of as-built building information models from laser-scanned point clouds: A review of related techniques
References
Object recognition from local scale-invariant features
A performance evaluation of local descriptors
Shape matching and object recognition using shape contexts
Approximate nearest neighbors: towards removing the curse of dimensionality
A performance evaluation of local descriptors
Related Papers (5)
Frequently Asked Questions (12)
Q2. What is the advantage of the 3D shape context over the other two descriptors?
A benefit of the 3D shape context over the other two descriptors is that a point-to-point match gives a candidate orientation of the model in the scene which can be used to verify other point matches.
Q3. What is the first decision to make when designing a 3D shape context?
When designing such a 3D descriptor, the first two decisions to be made are (1) what is the shape of the support region and (2) how to map the bins in 3D space to positions in the histogram vector.
Q4. What is the support region of a spin image at a basis point p?
The support region of a spin image at a basis point p is a cylinder of radius rmax and height h centered on p with its axis aligned with the surface normal at p.
Q5. How do the authors translate a 3D shape context into a harmonic shape context?
The authors translate a 3D shape context into a harmonic shape context by defining a function fj(θ, φ) based on the bins of the 3D shape context in a single spherical shell Rj ≤ R < Rj+1 as:fj(θ, φ) = SC(j, k, l), θk < θ ≤ θk+1, φl < φ ≤ φl+1.
Q6. How many distances between RDs and the returned reference descriptors?
To remove outliers caused by unlucky hash divisions, the authors included in the sum in equation (5) only the 80 smallest distances between RDs and the returned reference descriptors.
Q7. What is the main reason for the noise in the range measurement?
high-speed range scanners (e.g., flash ladars) introduce significant noise in the range measurement, making it nearly impossible to manually identify objects.
Q8. What is the problem with placing a hard vote?
The problem is that in placing a hard vote, the authors discard the relative distances between descriptors which provide information about the quality of the matches.
Q9. What is the method used to divide the highdimensional feature space?
The method divides the highdimensional feature space where the descriptors lie into hypercubes, divided by a set of k randomly-chosen axis-parallel hyperplanes.
Q10. What is the cost of using 3D shape contexts?
In this section, the authors briefly explore the cost of using 3D shape contexts and discuss a way to bring the amount of computation required for a 3D shape context query closer to what is used for spin images while maintaining accuracy.
Q11. What is the match for the representative descriptor?
The authors then sum the distances found for each qk, and call this the representative descriptor cost of matching Sq to Si:cost(Sq,Si) = ∑k∈{1,...,K}min m∈{1,...,M} dist(qk, pm) (5)The best match is the reference model S that minimizes this cost.
Q12. What is the way to measure the cost of a descriptor?
Scoring matches solely on the representative descriptor costs can be thought of as a lower bound on an ideal cost measure that takes geometric constraints between points into account.