Fast PRISM: Branch and Bound Hough Transform for Object Class Detection
read more
Citations
Hough Forests for Object Detection, Tracking, and Action Recognition
Voting for Voting in Online Point Cloud Object Detection
Visual Object Recognition
Detecting Surgical Tools by Modelling Local Appearance and Global Shape
Globally optimal consensus set maximization through rotation search
References
Histograms of oriented gradients for human detection
Robust Real-Time Face Detection
SURF: speeded up robust features
Scale-space and edge detection using anisotropic diffusion
Speeded-Up Robust Features (SURF)
Related Papers (5)
Frequently Asked Questions (12)
Q2. What are the future works in "Branch and bound hough transform for object class detection" ?
Future work will aim at deepening the understanding of the approximate bound and improving the splitting strategy of the branch and bound algorithm.
Q3. How much time can the authors reduce the computation time without reducing the accuracy?
The authors see that for values up to β = 0.1 and 0.15, respectively, the computation time (ignoring feature extraction) can be reduced by about a third without decrease in accuracy.
Q4. What is the disadvantage of kernel density estimators?
The disadvantage of kernel density estimators is their strong dependence on the training data (in terms of storage and computation time) which is unfavourable for large training sets.
Q5. What is the finitedifferences implementation of the regularisation matrix?
2. Consequently, the regularisation matrix defined above can be derived as a finitedifferences implementation of ∫R3 α‖W‖2 + β‖∇W‖2dV (18) where the integration is over the invariant space and ∇ denotes the gradient operator.
Q6. How can the authors efficiently process the maximum queries of (9)?
4.3 Maximum Query Using Integral ImagesA possible method to efficiently process the maximum queries of (9) is by means of integral images.
Q7. What was the first generalised Hough transform?
The Hough transform was originally introduced for line detection, while the Generalised Hough Transform (Ballard 1981) presented modifications for finding predefined shapes other than lines.
Q8. What is the hypothesis score for the sliding-window paradigm?
the hypothesis score is computed by the inner productS(λ|I,W) = 〈φ(λ, The author),W 〉 (2) of the footprint φ and a weight function W , i.e., the object model.
Q9. How does the SVMs integrate spatial pyramids?
the authors integrate spatial pyramids directly into Support Vector Machines (SVMs) by modifying their usual L2-norm regularisation.
Q10. What is the criterion for a natural scale-invariant convergence?
a natural scale-invariant convergence criterion is to stop whenever the extent of ̄ in each dimension is less then a threshold (e.g. 0.01).
Q11. What is the underlying idea of the Implicit Shape Model?
More recently, the Implicit Shape Model (ISM) (Leibe et al. 2008) has shown how the underlying idea can be extended to object category detection from local features.
Q12. What is the common setup for learning?
This is a common setup (Dalal and Triggs 2005; Felzenszwalb et al. 2008; Lampert et al. 2009) where learning can be accomplished using discriminative methods (e.g. linear SVMs).